This is a technical deep dive, not a CES recap. We read the specifications so you don’t have to. You’re welcome.

In October 2025, the AdCP coalition staged a controlled demonstration of an agent-to-agent media buy. A buyer agent negotiated with a seller agent. LG Ad Solutions provided the CTV inventory. The brand being advertised was fictional. A non-alcoholic drink invented for the demo. Real money technically changed hands. Two humans were present, approving steps as the agents worked through the transaction.

McKinsey compared it to watching the first banner ad being served.

The comparison deserves a moment. The first banner ad, a 1994 AT&T placement on HotWired, was not a demo. It was a real advertiser, real money, a real audience, and no human watching each impression serve. It fired. The October 2025 demonstration was a controlled proof of concept representing a fictional brand, with two engineers from the two companies present to fine-tune and approve along the way. Calling these the same kind of milestone is the kind of optimism that produces good press releases and imprecise expectations.

That one data point captures the entire state of agentic advertising in 2026. The infrastructure work is genuine. The autonomous-at-scale part remains aspirational. The industry has found this distinction inconvenient and has mostly declined to make it.

Between October 2025 and today, three significant protocol specifications emerged: the Ad Context Protocol (AdCP), the Agentic RTB Framework (ARTF), and the User Context Protocol (UCP, now renamed Agentic Audiences). All three are serious technical work. All three carry the word agentic prominently. None of them, by themselves, delivers an autonomous advertising agent transacting without human oversight at commercial scale. By May 2026, the most credible public data point was one SSP reporting 30 fully autonomous campaigns live, against a market running hundreds of billions of ad transactions daily.

The investor on that earnings call who called 30 campaigns immaterial was arithmetically correct. The more instructive number: every advertiser who ran one came back to run another. These are different signals. The industry has been citing the second as if it disproves the first.

This article examines what the three protocols actually do at an engineering level, how they relate to each other, where they conflict in both technical and governance terms, and what the sufficient conditions for genuine autonomous advertising look like. The specifications have been read. The press releases have been set aside.

* * *

What “Agentic” Actually Means, From First Principles

The word agent in AI research has a precise technical meaning that predates its current marketing career by several decades. The standard definition, established by Stuart Russell and Peter Norvig in their foundational textbook, describes an intelligent agent as a system that perceives its environment and acts upon it. That definition is deliberately minimal. The interesting distinctions come from the properties that separate a genuinely autonomous agent from a very capable automation system.

The four properties are perception, planning, tool use, and autonomous feedback.

• Perception: the system reads its environment through data sources, signals, APIs, logs.

• Planning: the system sets goals, breaks them into sub-goals, and generates action plans.

• Tool use: the system acts through external tools and APIs it can call, systems it can write to, transactions it can execute.

• Autonomous feedback: the system observes the outcomes of its actions, compares them against its goals, and updates its strategy without a human approving each step in the cycle.

The fourth property is the only one that separates an agent from a very expensive if-then statement. The first three describe automation. All four together, with the feedback loop closing without human intervention, describe agency.

Most of what adtech labels as agentic in 2026 handles the first three well and installs a human at the fourth. Campaigns are set up by AI. Optimization recommendations are generated by AI. Bid analysis, audience segmentation, pacing reports. All faster and sharper than before. A human reviews, approves the significant moves, and signs off on anything above the budget threshold that the contract specifies. The AI accelerates the work. The human still makes the decisions that cost real money.

There is an accurate word for this. The word is assisted. Assisted is genuinely valuable and represents real progress. It is not the word being used in the quarterly earnings calls.

The spectrum runs through identifiable stages. At the deterministic end: rules-based bidding with fixed CPM floors. No learning, no goal-setting. In the middle: ML optimization systems that adjust bids continuously toward a target metric without per-step human approval. These have perception and self-correction, but their goal is fixed by the human and their tool access is confined to a single platform. Near the fully agentic end: a system that receives a business objective in natural language, decomposes it into a multi-platform media strategy, executes across channels, observes outcome signals from independent measurement sources, identifies underperformance against the stated objective, and revises the strategy, all without a human in any loop.

That last system does not exist in adtech at commercial scale. The three protocols analyzed here are necessary infrastructure for it. They are not sufficient, and the industry has conflated necessary with sufficient, which is the source of most of the confusion.

* * *

ARTF: Brilliant Infrastructure With a Borrowed Label

The Agentic RTB Framework, version 1.0 finalized by IAB Tech Lab in March 2026 with version 2.0 already in active development, solves a latency problem the industry has lived with since roughly 2015 and discussed at conferences since roughly 2016. The problem is worth explaining carefully because the specification’s solution is the most technically rigorous piece of work in the current protocol landscape, and that technical rigor deserves to be understood separately from what the marketing around it implies.

The current RTB architecture requires any enrichment of a bid request to traverse a network. An SSP receives a bid request from a publisher, wants to apply fraud scoring, contextual categorization, audience segment matching, or custom bidding logic, and calls the relevant vendor’s API. That call crosses the open internet to a separate data center, gets processed, and returns. The round trip runs 600 to 800 milliseconds. The auction timeout is typically 100 to 300 milliseconds. The enrichment loses the race more often than it wins. The bid exits without the data, or misses the auction entirely. The data vendors selling enrichment services are delivering a product the architecture physically cannot use in time.

ARTF’s solution is co-location through containerization, and the concept connects directly to a principle that has transformed every other layer of software infrastructure: edge computing.

Edge computing’s organizing idea is that the latency bottleneck in distributed systems is usually the distance between data and compute, and the solution is to move the compute closer to the data. Content delivery networks do this for web assets: instead of serving a video from a single origin server, it is cached at nodes physically close to end users. CloudFlare Workers, AWS Lambda@Edge, and similar platforms let application logic run at the network edge, cutting hundreds of milliseconds of round-trip latency to single-digit milliseconds. The principle is the same wherever it is applied: reduce the physical and network distance between where the data is generated and where it is processed.

ARTF applies this principle to the RTB bidstream. Rather than the SSP making a cross-internet network call to a vendor’s API, the vendor’s logic is packaged as a container and deployed inside the SSP’s own infrastructure like same data center, same server, or same Kubernetes cluster as the auction engine. The enrichment runs locally. There is no network hop. The 600-millisecond round trip becomes a local function call measured in microseconds. IAB Tech Lab’s specification documents claim up to 80 percent reduction in bid request and response time, from 600-800ms to approximately 100ms.

The technical architecture is more precise than the summary suggests. When a host platform (SSP, DSP, exchange, or ad server) invokes an agent container, it passes a defined subset of the bid request data to the container. The container applies its logic and returns an OpenRTB Patch: a structured set of proposed mutations, each tagged with a declared intent. An intent might be fraud_detection, contextual_enrichment, identity_resolution, deal_activation, or bid_adjustment. The host platform reviews the proposed mutations, applies its own policy, and merges approved changes into the live bid request before the auction proceeds.

The mutation-with-declared-intent model is a deliberate design choice with consequences. Containers propose. The host decides. A fraud container can flag an impression as high-risk and propose its removal, but the orchestrating SSP can override. This means an ARTF container is an intelligent specialist, not an autonomous decision-maker. It has deep expertise in its narrow domain. The consequential decisions about which bids to send, which deals to activate, which impressions to accept, remain with the host platform.

The security specification is explicit and strict. Agent containers have no external network access. All network ingress and egress is prohibited except for communication with the orchestrating host. A container running inside an SSP’s infrastructure cannot call home to the vendor’s servers, cannot exfiltrate bid data, and cannot make secondary API calls during execution. This is the same security model that edge compute platforms apply: the code runs in a sandboxed environment with access only to what the host explicitly provides. The serialization protocol is gRPC with protobuf rather than HTTP with JSON, chosen specifically for lower latency, smaller payload size, built-in schema validation, and better support for streaming patterns.

The pre-ARTF existence of this architecture in production is worth noting for its implications. The Trade Desk’s engineering team developed a custom container-based bidstream augmentation solution with Chalice Custom Algorithms, the custom AI bidding specialist, before the IAB Tech Lab specification existed. Adam Heimlich, CEO of Chalice, confirmed support for the standard at launch and noted the interoperability benefits. What was a bespoke bilateral integration becomes, through ARTF, something any host platform can implement and any vendor can deploy into. Standardization is not glamorous, but this is exactly what standardization is for.

The honest evaluation of ARTF: the engineering is correct, the latency problem is real, the containerization model is proven in adjacent infrastructure contexts, and adoption will come because the specification asks the industry to change architecture rather than business model. The word agentic in the framework’s name is defensible in the narrow sense that the containers are services with perception and action capabilities. It implies a level of goal-directed autonomy that the specification does not claim and the containers do not exercise. Both observations can coexist without contradiction.

* * *

AdCP: The Coordination Layer and the OpenDirect Question

The Ad Context Protocol is the most ambitious of the three specifications and the one generating the most institutional disagreement. It is built on Anthropic’s Model Context Protocol, governs a different time scale than RTB, and targets a category of advertising transaction that the industry has been failing to automate for a decade under a different name. The different name is OpenDirect. This context belongs at the beginning, not the end.

In January 2015, IAB released OpenDirect 1.0, a specification designed to standardize the automated buying and selling of premium direct-sold inventory. The problem OpenDirect was solving: the request-for-proposal and trafficking process for guaranteed direct deals was human-mediated, slow, error-prone, and could not scale. OpenDirect defined a standard API for inventory discovery, campaign setup, order management, and pricing. Implemented on both sides of a transaction, it would allow a buyer’s system to negotiate and configure a direct deal without a human emailing a spreadsheet. OpenDirect 2.0 followed in 2021 with updated features and renewed industry commitment. OpenDirect remains available on GitHub. It is implemented by a minority of platforms. The direct deal process still runs largely on email and spreadsheets, with meaningful OpenDirect adoption concentrated in specific regional markets rather than as a general standard.

AdCP is attacking the same structural problem. Understanding what is different requires understanding AdCP’s technical foundation.

Anthropic’s Model Context Protocol, published in late 2024, answers a specific problem in AI application development: how does a language model connect to external tools and data without custom integration code per tool? Before MCP, every connection between an LLM and an external system required bespoke engineering as in a custom API client, a custom authentication handler, a custom response parser, and ongoing maintenance as external systems changed. MCP standardizes the interface: every tool declares its capabilities in a common schema, and any MCP-compatible model can discover and invoke those capabilities without custom code per tool. By late 2025, the MCP SDK had recorded 97 million monthly downloads across more than 10,000 active servers deployed globally. Anthropic transferred MCP governance to the Linux Foundation in December 2025, with OpenAI, Google, Microsoft, and AWS each confirming support.

AdCP takes MCP as its foundation and adds an advertising-specific coordination layer on top. Where MCP defines how agents connect to external systems generally, AdCP defines what advertising platforms need to expose and understand for agents to negotiate and execute campaigns. The protocol has three deployed modules and one pending.

The Signals Activation Protocol defines how a buyer agent discovers and activates audience signals. Every AdCP-enabled publisher maintains an adagents.json file, a machine-readable declaration of the publisher’s properties, agent endpoints, available audience data, governing consent frameworks, and activation conditions. A buyer agent reads this file and understands what is available before opening a deal conversation, without a human logging into a publisher portal or a sales team picking up the phone. The adagents.json structure is conceptually similar to ads.txt and sellers.json, which the industry already reads at crawl time for supply chain validation. The difference is that adagents.json is read by an AI agent constructing a media strategy, not a crawler checking chain of custody.

The Media Buy Protocol defines how buyer and publisher agents negotiate deal terms. The buyer agent sends a structured brief specifying target audience, flight dates, pricing constraints, brand safety requirements, creative formats, and performance goals. The publisher agent responds with available inventory, contextual data, pricing options, and deal structure. The negotiation proceeds asynchronously, not in 100 milliseconds, but over whatever time frame the business decision warrants. Publisher-side agents can auto-approve campaigns meeting predefined criteria and route larger deals for human review. This asynchronous design is what makes AdCP suited for the deal types where context and relationship matter: programmatic guaranteed, private marketplace, CTV direct, DOOH, audio.

The Creative Protocol defines how buyer agents describe creative assets in structured metadata that publisher agents can parse. Rather than a URL pointing to a static file, an AdCP creative declaration includes format specifications, contextual targeting applicability, brand safety classifications, and for dynamic creative, the decision logic for variant selection. For DOOH and CTV, where creative appropriateness is inseparable from screen context and moment, this matters considerably more than it does for display.

The Curation Protocol, expected later in 2026, defines how deal IDs, PMP packages, and curated audience segments surface to buyer agents as machine-discoverable capabilities. This is the module that most directly challenges the current human-mediated curation sales process, and therefore the one most likely to generate institutional resistance proportional to its utility.

AdCP does not replace OpenRTB. The buyer agent uses AdCP to negotiate the deal and define the terms. OpenRTB then executes impression-level auction against the resulting deal ID at sub-100 milliseconds. AdCP is the contract negotiation. OpenRTB is the delivery.

So why will AdCP succeed where OpenDirect did not? The coalition has made two specific bets. The first is that MCP substantially lowers integration cost. OpenDirect required publishers and buyers to design, build, and maintain a custom API product. MCP-based AdCP integration means any MCP-compatible AI model can discover and invoke an AdCP endpoint without custom code per integration. The second is that AI agents change the demand economics. A human media buyer could not operate 500 simultaneous direct deal negotiations. An AI buyer agent can. The supply of publisher relationships that could be automated existed in 2015 but could not be extracted because the automation infrastructure did not. Whether these two changes are sufficient to overcome the commercial dynamics that defeated OpenDirect is the most important open question in adtech infrastructure right now, and it has an empirical answer that will be clear within 18 months.

Anthony Katsur, CEO of IAB Tech Lab, has publicly called AdCP deeply flawed and resource consuming, arguing that the Tech Lab’s existing standards already address what AdCP claims to solve. The criticism has technical substance. There is genuine overlap between AdCP’s scope and what extended IAB standards could cover. It also comes from a body that governs the existing programmatic infrastructure. AdCP, at its most ambitious, creates an agent coordination layer that routes around meaningful portions of that infrastructure. Reading the criticism with awareness of the institutional position from which it comes is not an accusation of bad faith. It is basic analytical practice.

* * *

UCP: The Protocol That Holds the Stack Together

The User Context Protocol, donated by LiveRamp to IAB Tech Lab on November 3, 2025, and formally renamed Agentic Audiences, is the foundational data layer that neither AdCP nor ARTF can function optimally without. It receives roughly one-third the conference stage time of the other two protocols, partly because it requires explaining embeddings before it makes sense, and partly because it does not have a dramatic origin story involving a fictional beverage brand.

The problem: when a buyer agent negotiates with a publisher agent through AdCP, both sides need to describe audiences. When an ARTF container performs identity resolution or audience enrichment in real time, it needs to represent and compare audience signals. The current method for expressing audience identity in programmatic involves hashed email addresses, cookie-based user IDs, segment membership lists, and proprietary identity keys that do not map cleanly across platforms. These representations are verbose to exchange, slow to compare, increasingly constrained by privacy regulation, and fundamentally fragmented.

UCP’s answer is embeddings. An embedding is a dense numerical vector that encodes the semantic meaning of something in a format machine learning systems can process efficiently. Rather than a list of segment memberships or a raw identifier requiring a panel match, systems exchange a learned vector that encodes the audience signal itself. The specification defines vectors ranging from 256 to 1,024 floating-point dimensions. Comparing two embeddings requires computing cosine similarity, a dot product normalized by vector magnitude which modern CPUs execute in microseconds across millions of comparisons per second. Comparing two segment membership lists requires set intersection across potentially hundreds of segments in different taxonomies. For ARTF containers operating within the auction’s time budget, this is not a marginal efficiency gain.

UCP defines three signal layers. The identity layer represents who the user is in a privacy-preserving format, attributes at a level of abstraction that cannot be trivially reversed to personally identifiable information. The contextual layer encodes the current session: content being consumed, time of day, device, behavioral signals from the active visit. The reinforcement layer encodes historical advertising response: which formats drove completion, which creative characteristics performed, which frequency levels produced saturation versus incremental effect.

The governance detail is worth stating clearly. UCP now resides in IAB Tech Lab’s GitHub repository. AdCP, governed independently by AgenticAdvertising.org, depends on a signal format standardized by the body whose CEO called AdCP deeply flawed. This is not necessarily a problem as open standards governance has worked in adversarial conditions before. It is a dependency that produces a specific structural dynamic: IAB Tech Lab controls the data layer that makes AdCP most efficient. This gives the Tech Lab indirect leverage over AdCP’s performance even if AdCP achieves broad adoption. Whether that leverage is exercised consequentially is a future variable, not a current fact.

* * *

The Stack, the War, and the Question Nobody Is Asking Directly

The logical relationship between the three protocols is a stack. UCP defines the data format for audience signals. AdCP defines the coordination protocol for agent-to-agent deal negotiation using those signals. ARTF defines the execution infrastructure for how agents operate inside the RTB auction using those same signals in real time. Each layer addresses a distinct problem. The architecture has internal coherence.

A buyer agent using AdCP negotiates a CTV campaign using UCP-formatted audience embeddings to describe the target audience in a format the publisher agent can compare directly against its available signals. When the deal executes, ARTF containers inside the SSP apply identity resolution and contextual enrichment in real time, in compatible embedding format, within the auction’s latency budget. The protocols are designed to interoperate. Technically, the stack works.

Governmentally, it is more complicated.

AdCP is governed by AgenticAdvertising.org, a new independent organization formed last year. ARTF and UCP are both governed by IAB Tech Lab, the established industry organization. These are not neutral parties in the same debate. IAB Tech Lab is the standards body for the existing programmatic infrastructure. AdCP, in its most ambitious form, routes around meaningful portions of that infrastructure. When Katsur calls AdCP deeply flawed, and when O’Kelley argues that adtech needs an open coordination layer independent of the existing standards body, they are making technical arguments that are also commercial and institutional arguments. The protocol debate is real. It is also a proxy for the more consequential question: who writes the rules for AI-mediated advertising transactions as that market matures?

Nobody is discussing this directly as a governance question, because governance questions do not have obvious answers and are harder to keynote than protocol comparisons. But the dollar value attached to controlling the standards layer of an industry measured in hundreds of billions annually makes the governance question the more important one. Observers following this debate should track it as a standards-war question alongside the technical one.

* * *

Seven Things the Industry Is Not Saying Clearly Enough

ARTF is infrastructure engineering wearing an agentic badge because 2026 requires agentic in the name. The problem it solves has been documented since 2015. The solution of co-locating compute with data, reducing the enrichment call from a network traversal to a local function call is the edge computing principle applied to RTB. The specification is technically rigorous. The container security model is correctly designed. The gRPC and protobuf serialization choices are right for the latency requirements. None of this requires the word agentic to be true. The word is there because that is the current vocabulary, and current vocabulary is how standards get adopted. This is observation, not criticism.

AdCP is OpenDirect 2.0 with a better technical foundation and an unresolved adoption question. The MCP foundation genuinely lowers integration cost in a way OpenDirect’s custom API approach did not. The AI agent demand dynamic genuinely changes the economics of publisher-side implementation in a way that did not exist in 2015. Whether these two differences are sufficient to produce different adoption outcomes than OpenDirect achieved across two versions is an empirical question with a known answer timeline. If broad two-sided adoption is not visible by mid-2027, the OpenDirect parallel will be difficult to avoid.

UCP is the most technically important protocol of the three for production deployment and the least discussed at industry events. Both of the other protocols perform better with UCP-compatible embedding infrastructure than without it. The industry is debating the protocols at the coordination and execution layers while the data layer remains in early community development. Building floors before the foundation is standardized is a common industry error with known consequences.

The controlled October 2025 demonstration was not the first autonomous advertising transaction in the way the first banner ad was the first online advertising transaction. The first banner ad was a real advertiser buying real reach with no humans watching each impression. The October 2025 demo was a proof of concept for a fictional brand, with engineers from both companies present, in a controlled environment. Calling these equivalent milestones creates expectations the current technology does not meet. This matters because unmet expectations produce the disillusionment that causes genuinely good infrastructure to be written off before it matures.

The governance war is the real story. The technical differences between AdCP and ARTF are worth understanding. The governance question of who controls the standards layer for AI-mediated advertising is the question that determines which technical approach shapes the industry at scale. It is not being discussed in those terms because governance wars are less conferencing-friendly than protocol comparisons.

No measurement protocol exists for the agentic future, and nobody seems to be in a hurry to build one. Autonomous agents improve through feedback loops. Feedback loops require outcome data that arrives faster than a quarterly MMM cycle and is more accurate than last-click attribution. The ANA’s December 2023 study found that only 36 cents of every dollar entering a DSP effectively reached the consumer. The remaining 64 cents distributed through a decision chain that is not fully legible to any single party. Autonomous agents making allocation decisions on top of broken measurement inputs will automate the existing problems more efficiently, not transcend them. Three protocols address execution. Zero address measurement.

None of the protocols currently mandates that autonomous decision logs be accessible to advertisers. A genuine agentic system produces a complete audit trail of every decision, which signal drove which allocation, which bid strategy was applied, why. That log is simultaneously the mechanism by which agents improve and the most comprehensive supply chain audit the industry has ever been positioned to produce. Whether it is accessible or internal to the platform running the agent is a design choice, not a technical constraint. The specifications leave this choice to the implementers. The implementers’ commercial incentives do not uniformly point toward accessibility.

* * *

What Genuine Agentic Advertising Actually Requires

Genuine autonomous advertising agentic systems that set goals, execute across platforms, observe independent outcome signals, and revise strategy without per-step human approval require five conditions that the current state of the industry does not meet. The protocols being built address one of the five. This is important context for evaluating the timeline claims currently in circulation.

First: interoperable identity that works across the open web without bilateral matching agreements. UCP’s embedding approach is the most promising technical candidate in the current landscape. It is not yet a standard. It is a proposal in active community development. Every agentic capability that depends on matching audience signals across platforms depends on this layer being stable and broadly adopted. Every timeline for autonomous agents that does not specify when identity interoperability arrives is incomplete.

Second: outcome measurement that closes the feedback loop within the campaign window. Agents that observe outcomes quarterly cannot update strategy within a campaign. Agents that rely on platform-reported attribution are optimizing toward a metric the platform controls. Independent, timely, cross-platform measurement is not a nice-to-have for autonomous advertising, it is the mechanism by which the autonomous part works. Without it, what is being automated is the execution, not the intelligence.

Third: standardized tool access across platforms via protocols like AdCP. This condition is being actively addressed. The rate at which buy-side and sell-side platforms implement working AdCP endpoints determines the rate at which buyer agents can operate across the open web without falling back to human negotiation. The protocol is correct. The adoption timeline is the variable.

Fourth: accountability structures for decisions made at autonomous scale. When a guardrailed autonomous agent makes a decision that produces a brand safety incident, or burns significant budget on demonstrably wrong inventory, the legal and contractual question of responsibility is not currently resolved. The industry is running a proof of concept at small scale while the liability frameworks are incomplete. This works at 30 campaigns. It becomes consequential at 30,000.

Fifth: commercial incentives that reward transparency in the decision layer. An autonomous agent’s decision log is the complete record of every allocation decision, every signal consulted, every tool called and is the most comprehensive supply chain audit the industry has ever been positioned to produce. It is also a direct threat to every margin that currently exists in the opacity of programmatic decision-making. Whether the industry builds the agentic future in a way that makes this log accessible or maintains it as internal platform data is not a technical question. It is a commercial choice, and the commercial pressures point in both directions simultaneously. The protocols do not resolve it. The contracts will, eventually, and the outcome will either validate or undermine the transparency claims attached to agentic advertising.

* * *

The Honest Summary

Three protocols exist. All three are doing serious work. ARTF addresses a real latency problem with correct engineering, will achieve meaningful adoption because it asks for architectural change rather than business model change, and is correctly labeled as agentic in the narrow sense that containers are specialist services with perception and action capabilities. AdCP proposes a genuinely novel coordination layer for agent-mediated deal negotiation, is built on a sounder technical foundation than its OpenDirect predecessor, and faces the same adoption challenge that defeated that predecessor across two versions and a decade. UCP provides the data layer both other protocols need to function at their stated potential and is the least mature of the three in terms of standardization.

Thirty campaigns at one SSP early this year is a proof of concept, not a revolution. Every advertiser who ran one came back to run another, which is the correct signal to watch: not the absolute number but the repeat behavior. Repeat behavior means the technology is useful enough to choose again. It does not mean scale has been achieved. The distance between useful enough to repeat and deployed across the industry at autonomous scale contains all five of the conditions listed in the preceding section, none of which the protocols by themselves resolve.

The agents have not arrived. The protocols have. The industry called both the same thing, which generated excellent conference programming and imprecise expectations. The protocols are necessary infrastructure for the autonomous future being described. Necessary infrastructure is valuable. It is also not the same as the future being described, and the industry’s habit of conflating the two will produce a disillusionment cycle that the good engineering underneath does not deserve.

The soup is real. Some of it is nutritious. The menu promised something more elaborate. The kitchen is still being built.

Radio City Music Hall, 10:30 AM, May 11, 2026

Radio City Music Hall seats 5,960 people. On the morning of May 11, 2026, most of those seats are filled with the media and advertising industry’s most important buyers, sellers, and people who are important at industry conferences. At precisely 10:30 AM, NBCUniversal takes the stage. Seth Meyers makes a joke about his employer’s streaming strategy. The room laughs, because everyone in it earns enough money to find streaming jokes funny. Roger Goodell is waiting for his cue somewhere in the front rows.

Over the next four days, Fox, Amazon, Disney, Netflix, Warner Bros. Discovery, and YouTube will each take their own stage in their own Manhattan venue. Billions of dollars of committed, pre-negotiated, guaranteed advertising inventory will change hands. No auction will take place. No bid will be entered. No DSP will log a win rate. Two parties will agree on a price and execute a deal, in a format invented before the modern internet existed and that has outlived every technology built to replace it.

The insertion order turned 32 this year. It is having a great run.

While the industry spent twenty years building real-time auction infrastructure to displace the guaranteed deal, the guaranteed deal kept the budget. Premium Programmatic budgets grew 52 percent year-on-year in Q1 2024 among large US advertisers, nearly three times faster than open exchange. According to eMarketer, 76.2 percent of US programmatic digital display ad spending is expected to flow through programmatic direct channels by 2026. The pipes changed. The deal structure did not.

To understand why, look at the money. Really look at it.

In 2025, Google, Meta, and Amazon collectively generated approximately $530 billion in advertising revenue against a global digital advertising market of roughly $900 billion. That is 59 percent to three companies. Add TikTok’s estimated $25 billion and Apple Search Ads’ estimated $5 billion, and five platforms account for 62 percent of all global digital advertising revenue. The remaining 38 percent is fragmented with every DSP, SSP, exchange, DMP, CDP, clean room vendor, measurement platform, brand safety company, identity graph provider, and independent adtech startup on the planet is fighting over and that includes the walled gardens’ share of fragmentation too.

The specific numbers, because specificity is the enemy of comfortable vagueness: Google’s advertising revenue for FY2025 was approximately $265 billion, verified from quarterly earnings, with Q1 at $66.8 billion, Q3 at $74 billion, and Q4 at $82.3 billion. Meta’s advertising revenue was approximately $196 billion, up 22 percent, per its 10-K filing. Amazon’s Advertising Services came in at $68.6 billion, up 22 percent, per its 8-K. These are SEC filings, not estimates.

On the independent side: The Trade Desk reported FY2025 revenue of $2.896 billion. Magnite: $714 million. PubMatic: $283 million, down 3 percent. Criteo: $1.9 billion, up 0.6 percent. Roku’s advertising revenue for 2025 was $2.4 billion. LiveRamp: approximately $790 million. DoubleVerify: approximately $680 million. IAS: approximately $600 million.

The Trade Desk’s $2.9 billion is genuinely impressive for an independent technology company. It also represents roughly 1.1 percent of what Google reported in the same year. Google doesn’t think of The Trade Desk as a competitor. It thinks of it the way a whale thinks of a barnacle: useful for demonstrating ecosystem diversity, not a strategic threat.

Walmart Connect generated $6.4 billion in global advertising revenue in 2025, growing 46 percent. Netflix’s advertising revenue more than doubled to approximately $2.2 billion; the company now runs its own in-house ad server, the Netflix Ads Suite, built from scratch and launched in 2025 after Netflix spent two years moving away from Microsoft’s Xandr infrastructure. Disney’s streaming advertising across Hulu, Disney+, and ESPN+ reached approximately $5.3 billion. Every one of these is a walled garden. Amazon’s audience data requires Amazon DSP. Walmart routes through The Trade Desk and its own systems. Netflix inventory is accessible programmatically through The Trade Desk, Google’s DV360, and Magnite as SSP, all sitting on Netflix’s own ad server. Disney runs DRAX, the Disney Real-time Ad Exchange, which covers its entire streaming portfolio and integrated with Amazon DSP in June 2025.

The market structure is not a crisis. It is arithmetic, and it explains every other paragraph in this article.

The Rebrand Machine: Reinventing the Same Ideas With Better Names

Adtech has a superpower that has nothing to do with algorithms or machine learning. The ability to take an existing concept, rename it, give it a three-letter acronym, build a conference track around it, and charge enterprise prices for the renamed version. The industry has been running this play since roughly 2010 with remarkable consistency.

The insertion order was invented in 1994. In 2008, the industry automated it and called it programmatic advertising. In 2012, when publishers wanted to protect premium inventory, the negotiated IO became a “Private Marketplace” or “Deal ID.” In 2018, the guaranteed-impression, fixed-CPM, non-auctioned deal was rebranded “Programmatic Guaranteed.” In 2025, these same guaranteed deals are being dressed up with AI labels and called “agentic-powered curated inventory.” The document AT&T and HotWired signed in 1994 has been renamed four times. It still says: I will buy X impressions from you at Y price. The pipe changed. The deal did not. The conference registration fee stayed the same.

The data vocabulary follows the same pattern. Third-party cookies became “identity graphs.” Identity graphs became “first-party data activation.” First-party data activation required a “data clean room.” Clean rooms, which are secure database environments with controlled access, have existed in enterprise IT since the early 1990s. Adtech discovered them in 2020 and immediately added platform fees. DMPs became CDPs when cookies started dying. CDPs became “data collaboration platforms” when CDPs weren’t delivering. At each renaming, the conference session was booked, the Digiday article was written, and the vendor deck said “transformative.”

The contextual targeting revival is the most revealing example. Before behavioral tracking at scale existed, advertising was targeted by context: car ads on automotive sites, finance ads on business pages. The industry spent fifteen years building cookie-based behavioral infrastructure and told publishers their editorial context was largely irrelevant to pricing. When cookies started disappearing, contextual targeting was repackaged as “the privacy-first solution” and positioned as an innovation. The industry went backwards, called it forward, and put it in a keynote. Nobody in the room looked embarrassed.

Now consider what “agentic AI” has actually produced for independent adtech so far. PubMatic’s AgenticOS drove 250 agentic deals in full year 2025, per its earnings release. PubMatic processes 78 trillion impressions per quarter. Two hundred and fifty agentic deals is not a product line. It is a press release with a number in it. Magnite partnered with Cognitiv to “enrich the programmatic bidstream with AI,” which in practice meant adding a machine learning layer on top of RTB infrastructure it already operated. The Trade Desk’s Kokai platform redesign arrived late, was adopted slowly, and its Q1 2026 guidance implied just 10 percent year-on-year growth, down from 25 percent the previous year. Nobody mentioned that in the keynote either.

Compare that to what the actual AI platforms built. Meta’s Advantage+, Amazon’s Performance+, and Google’s Performance Max were created from scratch on first-party data at scale. According to Meta, Advantage+ drove a 22 percent improvement in return on ad spend for e-commerce advertisers in 2025 trials. Amazon’s ML bidding system optimizes across hundreds of billions in gross merchandise value with real-time signals that no independent DSP can access. You can’t rebrand your way to these capabilities. You need the audience first, and the audience is inside the walls.

True Programmatic Has Never Actually Happened. Here Is Why That Matters More Than Ever.

“Programmatic” has been doing heavy lifting in advertising vocabulary for fifteen years, and the industry’s dirty secret is that it covers two completely different activities. One is a real-time competitive auction where the highest-value buyer wins an impression the moment it becomes available, based on live audience data, live pricing signals, and live creative decisioning. The other is automated execution of a deal two humans negotiated weeks or months ago. The industry calls both programmatic. One is genuinely revolutionary. The other is an API wrapper on a fax machine.

In premium CTV, PMPs account for about 60 percent of programmatic transactions and programmatic guaranteed deals about 35 percent, per adwave.com’s Q3 2025 analysis. Open auction, where any buyer can bid with no prior negotiation, represents under 20 percent of all programmatic CTV, concentrated in FAST channels like Tubi and Pluto TV.

In digital out-of-home, private deals account for 95 percent of spend on the Place Exchange platform.

In audio, Spotify’s premium inventory trades through direct and PMP. The programmatic pipes are live and functioning. What they are mostly transporting is guaranteed deals.

Why PG Exists and Why It Is So Stubborn

PG persists for one reason technology can’t fix: both parties prefer the certainty. A broadcaster selling premium CTV inventory upfront knows approximately how many people will watch a series based on historical viewing data. An advertiser buying that inventory knows approximately what reach they are getting. Both sides accept demographic estimates rather than individual guarantees in exchange for knowing the price and the committed volume in advance.

A publisher who switches from PG to real-time pricing takes on real risk: no committed revenue floor, no guaranteed fill rate, and the volatility of a market that prices based on who is actually watching rather than who was projected to watch. An advertiser who switches from PG gets more precise audience targeting but loses the ability to plan creative resources against a known delivery schedule. PG isn’t a technology failure. It is rational risk management by both parties in a market where uncertainty is expensive.

The CTV ad format gap shows how deep this problem runs. In December 2025, IAB Tech Lab released standardized specifications for six CTV ad formats for public comment: pause ads, menu ads, squeezeback formats, overlays, in-scene insertions, screensaver ads. The comment period closed January 31, 2026. The squeezeback, which compresses content to display an ad alongside it, has been a standard part of British linear television since the early 2000s. Sky Sports was running animated bugs and credit squeezes during Premier League coverage before smartphones existed. The L-band format appeared on BSkyB before most people working in CTV advertising today had graduated secondary school. An industry still defining what a pause ad looks like in December 2025 is not ready to replace the upfront with live audience auctions. You need the product before you can build the market.

Fubo TV claimed a “first” in 2025 when it offered programmatic buying of pause ads via Magnite’s ClearLine: one publisher, one format, one platform. Had the industry genuinely been innovating CTV infrastructure for a decade, this would have happened in 2019.

Agentic AI Can’t Reach Its Potential in a PG/PMP World

This is where the industry’s AI narrative runs into its most uncomfortable wall, and the wall is made of spreadsheets from Q4 2025 media plans.

Agentic AI in media buying is designed to do three things at scale that humans do poorly: real-time audience evaluation, dynamic bid optimization, and rapid creative personalization. When an AI agent evaluates an impression opportunity, it should be asking, in about 200 milliseconds, who is this viewer, what context are they in, what should I pay for this moment, and which creative version will perform best? That is what real programmatic at individual impression level looks like. And that is precisely what can’t happen inside a PG deal.

When 80 percent of premium CTV inventory is locked in a PG deal negotiated six months before the campaign runs, the AI agent is not making a 200-millisecond decision. It is executing a six-month-old decision made by a human media planner in a quarterly planning cycle. The CPM is fixed. The placement is fixed. The audience commitment is fixed. The creative rotation may have some flexibility, but everything that makes AI genuinely valuable, real-time decisioning, live audience optimization, dynamic pricing, has been removed from the equation before the campaign brief was written.

PG deals commit to specific impressions at a specific CPM regardless of who is actually watching at delivery time. The audience guarantee is demographic, Women 25-54 with household income over $75,000, not individual. The AI agent knows the composition of that demographic better than any human planner. It still cannot change the price, change the placement, or walk away from the deal. The parameters were set in February. The agent is managing delivery in November. This is not intelligence. It is a very expensive reminder system.

PubMatic’s 250 agentic deals points in the right direction, getting AI agents to negotiate and execute deals rather than humans, but negotiating a PG deal with AI isn’t fundamentally different from negotiating it with a junior planner using better software. The agent needs real-time inventory to make real-time decisions. In a market that is 80 percent pre-committed, the agent optimises the margins around a plan rather than building the plan from live signals.

The industry is building AI agents for the programmatic market that currently exists: mostly PG and PMP, mostly pre-negotiated, mostly operating at campaign level rather than impression level. What AI agents actually need to reach their potential is the programmatic market that doesn’t yet exist at scale: real-time priced, authenticated individual-level impressions, where the agent can optimise against current audience data, current campaign performance, and current creative signals simultaneously, in 200 milliseconds, across every impression the brand buys. Until that market exists, “agentic AI” in adtech is mostly a planning productivity tool wearing a very exciting name badge.

How CTV Moves From Upfronts to Real-Time Decisioning

Upfronts don’t disappear because the industry decides they should. They disappear when the conditions that created them, anonymous mass audiences, imprecise measurement, and the need for revenue certainty, are replaced by something better. Two of those three conditions are already changing in streaming, and the third is starting to.

Every viewer of Netflix is logged in. Every viewer of Disney+ is logged in. Every Amazon Prime Video viewer has an Amazon account with purchase history. Every Roku user is registered. For the first time in television’s history, the identity of every streaming viewer is known before the ad runs, not probabilistically, not via panel data, but deterministically, because authentication is a condition of access to the platform. That is the condition that makes the upfront eventually negotiable.

Disney understood what this means. In April 2025, the company expanded biddable ad technology across its streaming platforms, making live content from Hulu and Disney+ available through programmatic biddable integrations. Not a PG deal. A real-time bid on a verified individual at a specific content moment. An advertiser can tell the system: I want to reach 35-49-year-old male car intenders who haven’t seen my creative in the last 14 days, at this content moment, at this price. The system executes with no upfront commitment required.

SCTE 35 signals fire at every ad break in streaming broadcasts. Dynamic Ad Insertion is already live on Amazon’s Thursday Night Football, YouTube’s NFL Sunday Ticket, and Disney’s ESPN+. A DSP can, in principle, receive those signals, evaluate the authenticated viewer’s profile, bid accordingly, and deliver the correct creative within the latency window of a streaming ad break. For server-side DAI, that window is 200-500 milliseconds. Technically achievable today, at scale, on authenticated streaming audiences.

Upfronts survive as a risk management product for advertisers who need revenue certainty and reach guarantees. Real-time biddable grows as a performance product for advertisers who need audience precision. Premium inventory doesn’t defect from one model to the other overnight; what changes is the share. Industry analysts project addressable CTV to reach $28 billion in the US by 2030, growing at 32 percent annually, per eMarketer. Some portion of that will be real-time, authenticated, impression-level decisioning. Not the end of the upfront. The beginning of the market the upfront can’t serve.

Live Sports: The Last Guaranteed-Deal Fortress, and the Path Through

Live sports is the most expensive, most watched, and most resistant category in advertising. The NFL, the Premier League, the NBA, the Olympics, these are not just programming. They are the last mass-audience live events in an era of fragmented, on-demand attention. A concurrent audience of 100 million people watching the same moment is extraordinarily rare, and advertisers pay for that scarcity. The Super Bowl 30-second spot costs approximately $7 million. That price reflects the value of the shared cultural moment, not the precision of the targeting. An open auction wouldn’t clear above it, because the value is in the moment itself, not the individual profile.

But within those 100 million Super Bowl viewers, there are meaningful segments worth targeting precisely: fifteen million male auto-intenders aged 35-55, eight million people actively in-market for insurance, four million households with children under 12. A truly programmatic sports model wouldn’t replace the mass-reach buy. It would add an authenticated addressable tier alongside it, for advertisers who need precision more than reach.

The technical infrastructure for this is already in place for streaming. SCTE 35 signals fire at every ad break in streaming sports broadcasts. DAI is already live on Amazon’s Thursday Night Football, YouTube’s NFL Sunday Ticket, and Disney’s ESPN+. In principle, these platforms can replace a national broadcast spot with an individually targeted ad in the streaming version, differentiated by the viewer’s authenticated identity. In practice, three things prevent this from working at scale today.

Rights holders won’t move to real-time pricing as long as upfront commitments pay more reliably. The NFL earns approximately $10 billion per year from broadcast partners who committed upfront and guaranteed. Trading that certainty for the possibility of higher CPMs from addressable targeting requires data that makes an airtight case. The data exists on authenticated streaming platforms. It hasn’t yet been compiled and shared in a way that closes the argument. Creative is a second problem: sports advertisers produce hero TV spots, not personalized variants, and serving different ads to different viewers in the same game break requires multiple creative versions and trafficking infrastructure that most advertiser operations aren’t built to handle. And measurement feedback is too slow for agents to optimize live bids. The viewer sees the car ad during halftime, researches the model three days later, visits a dealer two weeks after that. An agent can’t make live bid adjustments without near-real-time outcome signals, and that signal chain across authenticated view, purchase intent, and conversion is a multi-year infrastructure project.

Addressable sports advertising develops first in streaming-only events where every viewer is authenticated, second for performance-oriented advertisers like DTC brands and insurance companies who have faster outcome signals and clearer ROI models, and third during lower-stakes games where publishers are willing to experiment with pricing. The Super Bowl upfront isn’t going anywhere in the next decade. By 2030, a meaningful share of streaming sports inventory, perhaps 20 to 30 percent, will trade in real-time based on authenticated audience data rather than upfront demographic projections.

For agentic AI, live sports is actually the most promising near-term use case, not for buying the impression in real-time, but for negotiating the deal intelligently. An AI agent that can evaluate a rights holder’s sports inventory package, compare it against audience delivery alternatives, model expected outcomes across multiple creative versions and audience segments, and propose an optimal deal structure in seconds rather than weeks is a genuine productivity leap. The real-time auction for the live Super Bowl impression is a five-to-ten-year problem. Intelligent deal negotiation for the streaming sports upfront is a two-to-three-year problem. Sequence the ambitions accordingly.

Measurement: Last Touch Is the Industry’s Operating Reality. Build Around It, Not Against It.

Last-touch attribution is not a fiction. Anyone who calls it one has a consultant to sell you. Last-touch is the industry’s operating reality: it has been since the first ad server, it is now, and it will be for the foreseeable future. One problem: it’s the only measurement metric the entire industry agrees on, nobody independent is verifying it, and the platforms that benefit most from it are the same platforms that control the last step before purchase. Google paid search, Meta retargeting, Amazon Sponsored Products. Calling that a coincidence requires more optimism than the data supports.

Last-touch works as an operational tool. If someone clicks a Google ad and buys, that click happened, and measuring it is legitimate. The trouble starts when last-touch becomes the only model, because it eliminates all accountability for the awareness, consideration, and intent-building that happened before the final click. The CTV campaign that ran three weeks before the search. The editorial that shaped the decision. The social post that introduced the brand. None of those get credit. All the credit goes to whoever was standing at the door when the customer arrived, regardless of who built the road. Google, Meta, and Amazon built the last mile. The rest of the industry built the road. Last-touch pays only for the last mile.

With AI, this goes from bad to worse. When a customer asks ChatGPT for a product recommendation and ChatGPT names a brand, last-touch attribution gives OpenAI 100 percent of the credit. The brand that spent months on CTV, social, and display building awareness gets zero. The publisher whose editorial coverage informed ChatGPT’s model knowledge of that brand gets nothing either. And OpenAI has no attribution model at all; it launched its ad product in February 2026 and has no independent measurement infrastructure in place. The industry is walking into the AI advertising era with a measurement framework that was already broken and is about to break in entirely new ways.

Accept last-touch as the baseline. Then mandate independent verification of it, at the same level the industry mandated viewability verification in 2014. When the MRC established viewability standards, it didn’t ask Google and Meta to self-report. It created an independent audit standard that third-party measurement companies implemented. Attribution needs the same treatment: not as an optional add-on that gets cut in the first budget negotiation, but as a baseline campaign condition. An advertiser buying on ChatGPT at $60 CPM should require an independent partner to verify delivery and measure downstream outcomes, the same way it requires DoubleVerify to verify viewability on YouTube. The tool exists. The mandate does not.

Measurement fails the same way across programmatic, CTV, retail media, and AI. The people who would have to fix it benefit from it remaining unfixed. The publisher runs the ad server and reports impressions. The retailer runs the attribution and reports ROAS. The platform optimizes its own algorithm and reports performance improvements. The ANA documented that publishers receive roughly 51 cents of every dollar spent in open programmatic display. The other 49 cents goes to fees, intermediaries, and technology layers insulated from measurement because transparent measurement would reveal what they cost.

Retail media was supposed to be the exception, the closed-loop attribution the industry had been promising since the first ad server launched. Skai and Stratably’s 2026 State of Retail Media survey found 70 percent of brands claim to meet or exceed their retail media goals. Fifteen percent strongly trust their measurement. Advertisers are operating across an average of six retail media networks, each with its own attribution window, each with its own definition of incrementality. As Profitero’s Chief Growth Officer Mike Black put it: “As an industry, we throw the word Incrementality around a lot, but do any of us EVEN AGREE on what it means?” No. Eleven dashboards by end of 2026. Eleven definitions. Zero consensus.

AppsFlyer and Adjust in mobile, DoubleVerify and IAS in display and video, have built genuine businesses as measurement partners. Access is their Achilles heel. Their access depends entirely on platforms granting permission. Meta selectively expanded third-party measurement access in 2025. IAS called this a milestone. It was a milestone in Meta’s permission granting, not in independent accountability. When a verification vendor celebrates receiving access, it is not acting as an auditor. It is acting as a guest who is grateful to be let in. Guests don’t set the terms. Major buyers, holding companies, major CMOs, and the ANA need to make independent measurement a non-negotiable condition of spend. You don’t negotiate brand safety with Google by asking Google to self-report. You require an independent layer. Attribution should be no different.

New Walled Gardens: Interoperable Walls Are Still Walls

In April 2025, a federal judge ruled that Google had willfully monopolized the publisher ad server and ad exchange markets. The industry celebrated. The remedy is still pending. AdX is still running. Google’s ad business is structurally intact. While the lawsuit was being litigated, new walled garden infrastructure was being assembled by companies the antitrust case didn’t name.

Netflix launched its ad-supported tier in November 2022 with Microsoft’s Xandr as its primary technology partner. By May 2024, Netflix had decided to build its own stack. At its 2024 upfront, Netflix president of advertising Amy Reinhard announced the company was building the Netflix Ads Suite, its proprietary in-house ad server, and adding The Trade Desk, Google’s DV360, and Magnite as programmatic partners. The Netflix Ads Suite was live by 2025. Microsoft, for its part, shut down Xandr Invest, its demand-side platform, by February 2026, the quiet ending to a $1 billion acquisition it had made from AT&T in 2022. Xandr’s SSP and curation layer continue to operate, but the DSP is gone. Netflix inventory now trades through The Trade Desk, DV360, and Magnite. Control stays with Netflix. The buyers just rent access.

Disney runs DRAX, the Disney Real-time Ad Exchange, covering Hulu, Disney+, ABC, ESPN, and FX inventory. Disney integrated DRAX with Amazon DSP in June 2025. Roku’s advertising revenue reached $2.4 billion in 2025, driven by OS-level data across 100 million active households. Samsung Ads operates a controlled marketplace using ACR data that no independent DSP can access without Samsung’s permission. Paramount+ and Pluto TV are consolidating ahead of a likely merger with Warner Bros. Discovery later in 2026. Every premium CTV platform is converging on the same model: control the supply, control the data, control the measurement, and accept buyers’ money on your terms.

What separates the new CTV and retail media walled gardens from Google and Meta is that they can’t source demand exclusively. They need DSPs and agencies to bring budgets. The industry can call them “interoperable walled gardens,” meaning they control the supply, data, and measurement while cooperating with the demand ecosystem rather than owning it outright. A wall that accepts visitors on a guest pass is still a wall. Visitors don’t set the terms. Visitors don’t audit the data. And visitors get shown the door when the host decides the relationship no longer serves their interests.

Then the AI walled gardens launched, making everything else look almost transparent by comparison. ChatGPT started showing ads in February 2026 at $60 CPM with 800 million weekly active users, the fastest-growing new ad platform since TikTok. Google Gemini is extending existing programmatic infrastructure into AI-generated responses, with ads appearing in 25.5 percent of AI Overviews. Microsoft Copilot is integrating advertising into AI-assisted Office 365 workflows. None of these have an OpenRTB specification. None have a GARM brand safety framework. The transparency architecture of adtech, ads.txt, sellers.json, OpenRTB, the supply chain object, was built for web pages. Conversational AI interfaces are not web pages.

Zero-click searches rose from 56 to 69 percent between 2024 and 2025. Organic traffic to US websites fell from 2.3 billion to under 1.7 billion visits in the same period. Business Insider lost 55 percent of its organic search traffic between 2022 and 2025. All AI platforms combined account for 1 percent of publisher referral traffic. AI is consuming content at scale and answering user questions directly without sending anyone to the source. Traditional search volume is forecast to fall 25 percent by end of 2026. The content that trained these models was created by publishers who aren’t being compensated for training the systems now reducing their traffic. No antitrust case covers this yet.

Credit Where It’s Due, Skepticism Where It’s Warranted

The industry does have people actually building things. They deserve acknowledgement alongside the honest observation that most of the work is harder than the press releases imply and earlier-stage than the conference keynotes suggest.

Index Exchange: the cloud tax is real and they are attacking it. At the IAB Annual Leadership Meeting in February 2026, CEO Andrew Casale outlined containerised models running directly on exchange infrastructure, eliminating the compute cost per bid request that SSPs pass to publishers while accelerating processing from milliseconds to microseconds. Rethinking the physics of an auction rather than renaming it. Worth watching. Skeptic’s note: it is still an SSP competing on efficiency in a market where buyers are cutting the number of SSPs they work with. Efficiency helps. It doesn’t reverse the trend.

UID2.0 is honest infrastructure. Adopted by Index Exchange, Magnite, PubMatic, and Nielsen, live at 400-plus publishers, UID2.0 is deterministic, consent-based, and auditable. The average authenticated match rate of 47 percent is below the cookie era’s 68 percent, but the data is honest. Skeptic’s note: 47 percent match rate means 53 percent of impressions still carry no durable identity signal. The identity problem is smaller than it was. It isn’t solved.

Prebid.org is the most under-appreciated infrastructure in adtech. Open-source header bidding software that runs programmatic auctions for thousands of publishers without taking a percentage of revenue. Infrastructure built as a public good, without a sales team, used by every major publisher in the world. When The Trade Desk and Prebid had their Transaction ID dispute in 2025, over a data field intended to reduce auction duplication and fraud, it was a disagreement between the largest independent DSP and the rails that actually enable open internet buying.

Amazon Marketing Cloud got democratized. AMC became free for all Sponsored Ads advertisers in September 2025, putting a clean room that was previously available only to enterprise managed-service clients within reach of any brand running an Amazon campaign. Amazon remains a walled garden. It is becoming a more honest one, at least on the measurement side. Progress is progress, even when it comes with an Amazon Prime logo.

Moloco is building the right model. Starting in mobile app advertising where it grew by outperforming walled garden algorithms, now expanding into CTV and retail media, Moloco’s bet is that the ML algorithm is the product and a strong enough model can sit on any supply. The company is private and doesn’t disclose revenue, which makes independent verification of its claims impossible. But the model is directionally correct: bilateral, ML-led, outcome-accountable.

IAB Tech Lab released CTV format specs. Finally. The December 2025 CTV Ad Portfolio specification, with standardized definitions for pause ads, squeezebacks, and overlay formats, creates the foundation for programmatic buying of these units at scale. IAB Tech Lab CEO Anthony Katsur acknowledged the industry should expect “several false starts” in agentic standards work. Implementation takes two to three more years. The specification exists. Linear TV had these formats for thirty years without a specification because direct deals came with format briefs. In a programmatic world, you need the spec first. Now there is one.

Stop Building Pipes. Start Proving Outcomes.

The open internet is not a lie. A lie implies intent. The people who built RTB infrastructure in 2008 were solving a real problem. The people who built DMPs in 2012 were solving a real problem. The people who built clean rooms in 2020 were solving a real problem. Each time, the solution worked. Mostly. Then the solution became the product, the product became the business model, and the business model became the thing that had to be protected from measurement that would reveal its cost.

That’s the half-truth. The open internet solved distribution, not accountability. It monetized reach, not outcomes. The inventory people actually want, premium streaming, branded audio, high-intent retail surfaces, AI conversational interfaces, was never sold through open auction. The auction monetized what was left. Magnificently, for a while.

Three platforms generate 59 percent of global digital ad revenue. The entire independent adtech sector generates roughly $12 to $15 billion combined. Google reports more than that in a single quarter. AppLovin built $4.7 billion in revenue in 2025 by owning both sides of the transaction: mobile game inventory and a machine learning DSP, growing 44 percent year-on-year, structurally aligned with advertiser outcomes. The commercial model that works in adtech is the one that proves value to both parties in the same deal. The model running out of road takes a percentage of transactions it didn’t improve and can’t measure.

The industry should stop asking “how do we grow the open internet’s share?” and start asking “how do we build advertising infrastructure that proves it worked for everyone, verified by someone with no stake in the answer?”

Five changes. The industry knows what they are. It hasn’t made them.

1. Mandate independent measurement as a campaign condition, not a budget line item. Every campaign above a meaningful threshold should require an independent measurement partner the same way viewability verification became standard after the MRC established that standard in 2014. Not optional. Not negotiated away when budgets get tight. A condition. AppsFlyer, Adjust, DoubleVerify, and IAS will only achieve meaningful industry access when buyers make that access non-negotiable. The platforms won’t grant it voluntarily. They never have.

2. Build AI attribution standards now, not in 2028. ChatGPT launched advertising in February 2026. The IAB and MRC haven’t started a working group on AI attribution standards. Building this infrastructure from specification to independent audit to industry adoption takes five to eight years. By 2028 the platforms will have written their own measurement standards, and the industry will be asking nicely for access to data the platforms control, exactly the position it was in with Facebook’s self-reported video metrics in 2016.

3. Build bilateral platforms. Stop calling DSP and SSP different things. Amazon is both. The Trade Desk builds supply paths. Magnite enables demand access. The DSP/SSP classification is a billing convention from 2010. The test for 2030 is simple: can you show the publisher independently verified earnings, and the advertiser an independently verified outcome? Yes on both means you are relevant. No on either means you are overhead.

4. Standardize retail media measurement before advertisers stop caring. Eleven retail media networks by end of 2026. Eleven attribution models. The Retail Media Alliance is attempting cross-network standards. Trade bodies with optional compliance don’t fix this. Independent third parties with enforcement authority do. Advertisers spending across multiple RMNs should be able to compare results in a common framework without buying each retailer’s proprietary clean room at full price.

5. Fund the auditors like they matter, because they do. ANA’s supply chain research, Jounce Media’s SPO analysis, IAB Tech Lab’s format work, these produce the most honest outputs in adtech. They are consistently underfunded relative to the marketing budgets of the companies they audit. The industry spent 49 cents of every publisher dollar on fees and intermediaries. Two cents of that directed toward independent audit infrastructure would change the accountability of the entire ecosystem. An industry that can’t fund honest accounting of its own supply chain has no standing to complain that walled gardens don’t share their data.

Today is May 11, 2026. Radio City Music Hall is running. Billions in guaranteed deals are being signed in New York. The CTV format specs are in adoption phase, arriving roughly a quarter-century after linear TV standardized the same formats. The AI walled gardens have launched, have no independent measurement, and are growing faster than any advertising platform in history. The retail media dashboards are generating numbers 70 percent of advertisers cheerfully report as successes and 15 percent trust. The publisher receives 51 cents per dollar. The supply chain absorbs the rest, mostly without audit.

Markets adjust when the cost of not adjusting becomes visible. The cost is becoming visible, in declining open exchange CPMs, in shrinking publisher traffic, in independent adtech market caps that peaked years ago, in ad budgets migrating to environments the industry has no tools to measure.

The Latin verb is agere. It means to act on behalf of another. From it comes “agent”: a person who acts on behalf of a principal. From it comes “agency”: a company staffed with such people. And from it, in 2025, came “agentic”: the adjective the technology industry chose for software that autonomously receives instructions, makes decisions, and executes tasks without a human at each step.

Same root. Same meaning. Different invoice.

Advertising agencies have been charging clients for agere since the 1840s. An AI agent running on Claude or Gemini does agere at machine speed, around the clock, for the price of a cloud bill. Either somebody in the etymology department was not paying attention, or somebody was paying very close attention and assumed the industry would not notice.

The industry noticed. The industry gave it a standing ovation.

That was Miami Beach, late April 2026. POSSIBLE 2026, the adtech Global conference at the Fontainebleau hotel, where a badge costs $2,995 and the agenda had no session titled “Should Agencies Be Afraid.” The sessions had better titles. “The Future of AI-Powered Marketing.” “Agentic AI and the CMO of Tomorrow.” The conversations between sessions were considerably more anxious.

Jeff Green, CEO of The Trade Desk, took the main stage with media executive Michael Kassan and delivered the most precisely calibrated statement of the conference. “You can’t keep taking money out of the middle,” he said, “providing a higher level of service in a more complicated ecosystem, and do that for less and less.” He called this moment “one of the greatest opportunities that I think any of us will ever see in this space.” He described the ad supply chain as being “cleaned up,” with fewer tolerated “toll collectors in the middle.” The agencies in the audience nodded. They understood what he was describing. They did not applaud that specific sentence.

In the same week, at Marketecture Live in New York, Green confirmed to Ari Paparo that The Trade Desk is running a closed beta allowing some advertisers to create campaigns directly through Anthropic’s Claude, with no human media planner configuring the platform.

Sympathy for the agencies in Miami. A tool that bypasses them in New York. Same person. Same week.

A few days after POSSIBLE ended, Daniel Miessler, founder of Unsupervised Learning, published a post on May 2, 2026 titled “Most Companies Aren’t Anywhere Near Ready for AI.” He was not writing about adtech specifically. He did not need to. “A massive percentage of companies are haphazardly successful despite themselves,” he wrote. “AI can do basically nothing for these companies. In fact, it could make it worse, because now it helps people flail more impressively. Like with more backflips and charts and stuff.”

The companies succeeding with AI, he observed, are “somehow magically the same companies that already know what they are doing” and can describe their goals, workflows, metrics, and costs clearly and consistently. Adtech holding companies have never found those questions easy to answer at the best of times.

This is the industry and this is the moment.

What the Conference Stage Says vs What the Trading Desk Does

Before the paradox can land properly, the technology needs to be described precisely, because the version on most conference stages is about 40% accurate and 60% aspiration.

Most companies right now are deploying conversational AI. You type a query. It answers. You ask it to draft a brief. It drafts. A human reviews every output before anything happens in the real world. Useful. Not agentic.

An agentic AI system works differently at the architecture level. The large language model, whether Claude, Gemini, or GPT-5, serves as a reasoning core that receives a high-level objective, decomposes it into sequential sub-tasks, calls external tools and APIs, observes the results, and re-plans when something fails, all without a human prompt at each step. The human sets the goal once. The system executes every step.

In advertising, the technical interface is Anthropic’s Model Context Protocol, released late 2024. MCP is a standardized connector that lets an LLM communicate securely with external systems: a publisher’s inventory API, a DSP’s bidding endpoint, a data clean room, a measurement dashboard. Write one MCP-compliant server and any MCP-compatible AI agent can talk to it, without a bespoke integration for each platform.

The Ad Context Protocol, launched October 15, 2025 by Scope3, Yahoo, PubMatic, Triton Digital, Optable, and Swivel, is built directly on MCP. A buyer’s AI agent sends a natural language brief. “Find premium video inventory, in-market auto buyers, Southeast Asia, 25-34, $18-22 CPM, Q3, $500,000 budget.” The seller’s AI agent packages available inventory, audience match rates, and pricing into a structured response. No insertion order email. No RFP sitting in an inbox for three days. No account manager on hold. The transaction is asynchronous and can accommodate a human approval loop, but the operational overhead is gone.

This works. In specific, constrained, production situations. Butler/Till, an independent agency, ran a December 2025 campaign for the beverage brand Clubtails using PubMatic’s AgenticOS powered by Claude. AI recommended tactics, executed media buys, and optimized in real time. Butler/Till described this as freeing their team for “higher-value strategic planning, creative development, and measurement.” Which is exactly the right framing. It is also worth asking: what specifically was the human doing during that buy that the AI could not have done without them? That question was not on a POSSIBLE panel. It is the only question that matters.

IAB Tech Lab’s Agentic RTB Framework, released for public comment November 2025, is technically distinct from AdCP and constantly conflated with it. Where AdCP automates deal negotiation above the auction, ARTF uses containerization to put third-party AI agents inside the auction itself. A fraud detection agent does not sit in a remote data center reviewing impressions post-bid. It runs inside the DSP at sub-millisecond latency, before the bid fires. The host platform gets the output without seeing the agent’s code. Anthony Katsur’s stated target: up to 80% reduction in bid request and response latency by eliminating cross-network data center hops. The Trade Desk, Netflix, Paramount, Yahoo, Index Exchange, and Chalice have all expressed support.

AdCP and ARTF are at different layers and do not technically conflict: AdCP above the auction, ARTF inside it. In principle they are complementary. In practice they represent competing governance structures and different commercial interests. AdCP is governed by the new Agentic Advertising Organization with Brian O’Kelley and former IAB CEO Randall Rothenberg on the interim board. O’Kelley built Right Media (sold to Yahoo) and AppNexus (sold to AT&T, eventually becoming Microsoft Invest). His track record is building infrastructure that becomes a market standard and generates substantial commercial value for its architects. That is context, not accusation.

Google, The Trade Desk as a full AdCP signatory, and Amazon DSP have not joined the AdCP consortium. When those three sit out, “industry standard” is an aspiration. ARTF has broader institutional support from the platforms that process the majority of programmatic volume. Katsur told The Current: “We are a shiny-penny industry, and agentic is a shiny penny, but I don’t think it’s blockchain. There is something of substance there.” The substance in ARTF is engineering-level. The substance in AdCP for creative workflow automation is real. For media buying at scale, it is running into the same structural walls Ari Paparo documented when he tried to build this at Google in 2010 and described again in his November 2025 Marketecture analysis: buyers won’t be price takers, sellers won’t reveal real rate cards, small publishers lack the data scale to deliver meaningful performance, large publishers have no incentive to automate their best inventory relationships. These are incentive problems. AI executes faster through them. It does not dissolve them.

Gartner published a formal prediction in June 2025: over 40% of agentic AI projects will be canceled by end of 2027 due to escalating costs, unclear business value, or inadequate risk controls. “Most agentic AI projects right now are early-stage experiments or proof of concepts that are mostly driven by hype and are often misapplied,” said Anushree Verma, senior director analyst at Gartner. Of thousands of vendors claiming agentic capabilities, Gartner estimates around 130 have genuine ones. The industry has a name for the rest: “agent washing.” We replaced greenwashing with agent washing without pausing for breath. McKinsey research found that while nearly two-thirds of enterprises have experimented with AI agents, fewer than 10% have scaled them to deliver measurable value.

This is the gap between the conference stage and the trading desk.

ARTF vs AdCP: different layers, different camps, and the three platforms processing most of global programmatic volume are not in the AdCP room.

The 36-Cent Scandal That AI Will Now Run at Machine Speed

The principal-agent problem is a foundational concept in economics. A principal hires an agent to act on their behalf. The agent has private information the principal does not have. Their interests can diverge. The agent can exploit that gap. The classic undergraduate textbook example is the advertising agency. This is not a metaphor. It is in the textbooks.

Agencies earn a percentage of media spend. Their financial incentive is to spend more, not less. They often have supply relationships with certain SSPs, exchanges, and data vendors that are not always fully disclosed to clients. The ANA’s December 2023 Programmatic Media Supply Chain Transparency Study quantified this formally, tracking $123 million in ad spend across 21 major brands between September 2022 and January 2023. Finding: only 36 cents of every dollar entering a DSP reached the consumer. Transaction costs, primarily DSP and SSP fees, consumed 29 cents. Non-viewable, non-measurable, fraudulent, or made-for-advertising inventory took 35 cents. Agency fees and brand safety costs were explicitly out of scope, meaning 36 cents is the optimistic version of the story.

The industry knew. It preferred clients not have those specific numbers in a published research report.

Now put agentic AI on top of this.

Vinny Rinaldi is VP of Media and Marketing Technology at Hershey, overseeing north of $2 billion in combined media and trade marketing investment. He is an operator, not a conference pundit. In a LinkedIn comment thread this year that circulated widely across the industry, he wrote the most important single paragraph produced by adtech in 2026:

This is not a complaint about technology. It is a precise description of how the principal-agent problem migrates into an AI system. If an agency builds the client’s agentic stack, the agency sets the objective function. If that objective includes preferences for supply partners where the agency has financial relationships not disclosed to the client, those preferences travel silently into the algorithm. The AI produces CPMs and reach figures that look neutral. The misalignment runs at machine speed. The audit trail points at a model, not a meeting.

This is a worse version of the transparency problem, not a better one.

Hershey’s practical response is the most instructive production case study in adtech right now. The company built agentic marketing mix modeling using Mutinex and Tracer, running on Claude and Gemini, using its own data, through analytics partners it controls directly. “We were getting the full read of 2024 data midway through 2025, while we were planning for 2026,” Rinaldi told Adweek. Always-on agentic MMM compresses the lag from annual to monthly. “You take all of our trade dollars into account, anywhere from two plus billion dollars of investment, you can now start to make decisions off of on a monthly basis.” The agency that previously delivered the annual readout on PowerPoint, six months after the data had passed its use-by date, is not in this workflow.

Ari Paparo served at Google’s DoubleClick business, where he was responsible for products sold to agencies and advertisers, before co-founding Beeswax, a DSP he sold to FreeWheel, and then founding Marketecture. He is methodical about distinguishing where agentic AI works from where it runs into structural walls. On creative workflow automation, he is genuinely positive: AdCP’s creative protocol solves a real, time-consuming, human problem. On media buying, he returned to his 2010 experience at Google in his November 2025 Marketecture analysis: “Buyers don’t want to be price takers. Sellers don’t want to reveal their rate cards.” His conclusion: “The use of agents ends up being more appropriate for reinforcing the role of intermediaries, rather than enabling a breakthrough to empower buyers and sellers.”

AI makes the middlemen faster. It does not make them unnecessary. Unless they have nothing left that only they can provide, which leads directly to what Green said in New York.

What $150 Million Says About Where AI Value Accrues

In March 2026, Jeff Green purchased approximately $150 million of Trade Desk stock personally. The Trade Desk reported $2.896 billion in 2025 revenue, with growth decelerating to 18% year-over-year, and Q1 2026 guidance came in below analyst expectations. The market was pricing genuine uncertainty about what large language models mean for a programmatic DSP.

Green’s thesis, delivered across a LinkedIn post, a detailed blog essay, and a Marketecture Live session with Paparo: “Agentic AI will accrete the most value to companies that already have deep customer trust, that have scaled, refined and objective datasets, and that prioritize objectivity.” He cited scale as the reason programmatic is uniquely suited to AI: 20 million ad impression opportunities every second, decisions required in 10 milliseconds or less. “I don’t think there is an industry in the world that is more conducive to AI than programmatic advertising.”

At POSSIBLE, Green described “complexity” as no longer being “an excuse for opacity.” The supply chain, he said, is being “cleaned up,” with fewer tolerated toll collectors. Which brings us to the subplot that is almost too neat. Adweek reported in February 2026 that Dentsu and WPP had quietly exited The Trade Desk’s OpenPath direct publisher program, citing transparency concerns and what they described as hidden fees. OpenPath was designed specifically to create a more transparent supply path. Green, the industry’s loudest critic of supply chain opacity, found himself accused of opacity by the agencies he critiques for opacity. In adtech, everybody is somebody else’s toll collector, and everybody else’s toll collector is a “trusted partner.”

Paparo’s April 2026 Marketecture piece distilled DSP survival to exactly two variables: unique data and unique inventory. “That’s it. I didn’t make this up. It was told to me over and over by not one, but tens or hundreds of potential buyers.” If agentic AI automates every execution function a DSP differentiates on today, the moat is what remains. Every DSP without proprietary data or exclusive supply has already lost the argument. The invoice just has not been itemized yet.

Google and Amazon Are Not Applying for Conference Panels

Google, Meta, and Amazon together controlled 54.7% of the global digital advertising market in 2025, according to WARC’s Global Ad Forecast. Their combined share is forecast at 56.2% in 2026. They are building agentic advertising infrastructure on their own terms, inside their own ecosystems, with no requirement to participate in any open standard.

Google’s AI Max, its AI-based bidding product for search and shopping, was described by AdExchanger’s on-the-ground POSSIBLE 2026 coverage as having “hundreds of thousands” of advertisers using it. Google’s Universal Commerce Protocol, co-developed with Shopify, Walmart, Target, Etsy, Wayfair, and more than twenty other partners, creates an agentic checkout layer inside Google’s AI surfaces. A consumer describes what they want in Gemini or AI Mode in Search. The agent handles research, comparison, and purchase without leaving Google’s environment. No open web ad placement. No DSP bid. No agency media plan.

Amazon owns the transaction data, the inventory across streaming and display and retail media, and increasingly the AI that bridges them. Amazon’s figures for its Rufus shopping assistant showed a 60% increase in purchases per session among customers who used it in Q3 2025. For advertisers whose primary metric is ROAS with closed-loop attribution, Amazon offers a complete system in which AI handles discovery, targeting, delivery, and measurement inside one platform that also owns the transaction. The open web cannot replicate this structurally.

Anthropic’s run-rate revenue passed $30 billion in 2026, up from approximately $9 billion at end of 2025. Over 1,000 enterprise customers are each spending more than $1 million annually. Claude is available simultaneously on AWS Bedrock, Google Cloud Vertex AI, and Microsoft Azure. The same reasoning engine is embedded in The Trade Desk’s campaign creation beta, in PubMatic’s AgenticOS, and in Hershey’s MMM platform at the same time.

The open web’s strategic response to all of this: two competing protocols, monthly standards boot camps at the IAB Ad Lab, and a public webinar on the roadmap. WARC’s data shows walled gardens absorbing roughly 1.5 percentage points of global ad spend from the open web per year. Agentic AI does not correct that trajectory. It accelerates it. The environments that benefit most from fully autonomous execution are the ones with closed data loops. Those are the walled gardens. The open web can close individual transactions. It cannot close the loop.

The billing math most agency principals are not running in client conversations. AI-native competitors are pricing 20-30% below traditional cost basis on the same deliverables.

Eight Ways an Agency Gets This Wrong, With Numbers

At CES 2026, WPP launched Agent Hub inside WPP Open, a curated marketplace of “Super Agents” working across client briefs. Omnicom unveiled Omni as an agentic AI operating layer. Both announcements received applause from audiences that included employees whose jobs those systems are being built to eventually handle. The cognitive dissonance in the room was something you could almost measure.

A Q1 2026 survey of 250 agencies by digitalapplied.com catalogued eight structural failure modes in adtech agency AI adoption. The most consequential:

62% of mid-market agencies still bill by the hour. Agentic delivery compresses production task completion time by 80-90%. An agency using AI to do in two hours what previously took twenty, and billing hourly, has destroyed 90% of its revenue on that engagement without a single conversation about repricing. The repricing conversation is also the “why exactly do you still need us at this margin” conversation. Most agency principals are choosing not to have it.

Only 11% of mid-market agencies have a mature evaluation harness: the systems to measure whether their AI output is better than the output a platform produces natively for free. Without this, the margin argument collapses the first time a client asks the question.

34% utilization of agentic engineering talent hired by agencies. The engineers landed in innovation labs, adjacent to the production teams that actually need the capability rather than integrated into delivery. Expensive headcount generating impressive slide decks while client work runs on human labor and monthly retainer.

Omnicom has been the most candid holdco about the underlying logic, which is worth acknowledging. CTO Paolo Yuvienco said at POSSIBLE that Omnicom has already “tested the pipes” with AdCP and run agent-to-agent media buys for several clients. The goal, he said: move budget into “working media” and less into “the machinery around it.” The machinery he wants to eliminate is operated by middlemen. The joke is that Omnicom is the machinery. It is automating its own operational layer and describing this as a competitive advantage. Even they cannot say it with a straight face in private.

WPP’s market cap fell from roughly £25 billion in 2016 to below £3 billion by early 2026. Dentsu cut 8% of its global workforce. WPP’s Elevate28 plan targets £500 million in gross cost savings by 2028. That saving is not coming from renegotiating server contracts.

Four Things Worth Paying Agency Margin For, Honestly

Not everything is being automated. Stating otherwise would be analytically lazy.

Original creative strategy. Not execution. Not variant generation. The upstream brief that explains why a brand occupies cultural space no competitor can easily claim. AI generates ten thousand variants and optimizes them programmatically. It did not write the brief that made those variants worth making. Hershey’s “Holiday Bells” campaign is 35 years old. Rinaldi used agentic AI to compress measurement cycles to monthly. He did not use it to write the brief that made 35 years of a campaign possible.

Navigating client organizations. A media strategy exists inside a political ecosystem: CFO risk tolerance, CMO career ambitions, procurement history, board quarterly expectations. Navigating those forces and serving as a credible external voice when internal consensus breaks down is irreducibly human. AI surfaces the analysis. It cannot read the room.

Building first-party data infrastructure. Rinaldi’s “input layer” argument cuts both ways. Agencies helping clients build clean, structured, activated first-party data through clean rooms, identity resolution, and direct publisher partnerships are creating infrastructure with compounding value. The AI needs the data. Someone has to build it. Genuine defensibility here, with one strict caveat: it only works if the agency does not hold the data in a way that recreates the conflict of interest Rinaldi described.

Independent measurement. As AI agents run campaigns autonomously and produce reports that look neutral, brands need someone they trust to audit those reports. A genuinely independent measurement function, structurally separated from any undisclosed supply relationships, that validates AI-generated outcomes against real business results, has growing and underserved demand. The agencies capable of providing it are the ones that have already given up the margin games that make structural independence impossible. That is a small group. It will be a valuable one.

These four things are not what most holdco agencies currently charge premium margin for.

The Bill

Agentic AI did not create the crisis in the adtech agency model.

Twenty years of opacity, principal-based buying, and undisclosed supply relationships created the crisis. The ANA’s 2023 study quantified it precisely: 36 cents per dollar reaching consumers from a DSP, 29 cents to transaction costs, 35 cents to low-quality or fraudulent inventory, and agency fees out of scope. The industry knew. It preferred clients not have those numbers in a published research report with a methodology they could audit.

Agentic AI is the mechanism through which clients will finally have an alternative that does not require trusting the intermediary to report honestly on its own performance. Jeff Green called this “one of the greatest opportunities that I think any of us will ever see in this space” at POSSIBLE. He is right. The opportunity is not equally distributed. It belongs to the entities with unique data, closed measurement loops, and the technical capacity to run AI against real business outcomes. It does not belong to companies whose primary competitive advantage is complexity. Complexity can now be routed around, at roughly 1.5 percentage points of global ad share per year and accelerating.

Miessler’s observation from May 2026 applies to adtech with surgical precision: the companies succeeding with AI are the ones that already know what they are doing and can describe it clearly. The ones that cannot will use AI to fail more impressively, with better charts and more confident keynotes.

The Latin word at the center of all of this is agere.

An industry named after human agents is building software agents to do what human agents do. It is presenting this as transformation. It is charging for the transition. It is convening annual conferences with $2,995 badges to discuss when it will be complete.

For the production layer of agency work, the answer is already.

The turkeys voted for Thanksgiving. They named the stuffing “Agentic AI.” They organized a panel about it at the Fontainebleau and charged $2,995 to attend.

The caterers have been booked.

Every May, New York City hosts its most elaborate annual performance in advertising. This year is no different. NBCUniversal is back at Radio City Music Hall on May 11, where it will celebrate the 100th anniversary of the NBC broadcast network while also, quietly, selling guaranteed ad inventory for the coming television season to the same agencies that have been buying guaranteed inventory in this same ritual since the 1950s. YouTube’s Brandcast takes over Lincoln Center on May 13. Disney, Amazon, Netflix, and approximately three dozen others follow across two weeks of open bars, celebrity appearances, and sizzle reels. The IAB moved its NewFronts to March this year, which means programmatic companies had their moment slightly earlier, but the same cast of DSPs, SSPs, measurement vendors, and ad-tech platforms are still in the room at every upfront presentation, standing alongside the streamers, claiming that programmatic CTV has finally arrived.

It has not, or rather it has arrived in the same way that a passenger train has arrived at a station that still uses 1950s ticketing infrastructure. The ride is smoother. The engine is quieter. The seat is more comfortable. The ticket still has to be bought at the window, in advance, at a fixed price, negotiated between human beings who then confirm the transaction over email. The programmatic revolution in television is real at the infrastructure level and largely theater at the commercial level. Understanding why requires understanding something that almost nobody in those hotel ballrooms will say out loud: that the operating system of the television set is now the most powerful piece of commercial infrastructure in modern advertising, that the companies who built it are using it to concentrate supply in ways that would give an economics professor pause, and that one of the most credible attempts to challenge that concentration is being made by The Trade Desk, which saw the problem clearly, built Ventura as a direct response, and now faces the question of whether its current architecture is sufficient to actually shift the balance of power.

Let us be specific about what is happening, and why it matters.

Who Controls Supply Controls the Market. The OS Companies Figured This Out First.

Before the technology, the economics. The television advertising market is, at its core, a supply and demand problem that has taken on a peculiar shape that standard market equilibrium theory was not quite designed to describe.

In a perfectly competitive market, producers set prices based on their costs, buyers choose among identical goods, and the market clears at an equilibrium where supply equals demand. Television advertising is not this market. Premium CTV advertising inventory is not a fungible commodity. A 30-second slot in an NFL playoff game on Peacock is not interchangeable with a 30-second slot in a FAST channel running reruns at 2am, even if both arrive in the same programmatic pipe. The playoff slot is scarce, contextually valuable, and brand-safe in a way that the advertiser will pay a substantial premium to obtain. Simulmedia has documented that premium CTV publishers price their inventory at two to six times the CPM of national linear television. The premium is not arbitrary. It reflects genuine scarcity of the contextual environment that major brand advertisers require.

In oligopoly theory, when a small number of suppliers control the majority of a scarce good, those suppliers hold market power. They can price above competitive equilibrium, restrict supply to maintain that premium, and structure the market to make substitution difficult. This is precisely what has happened in premium CTV inventory, and the operating system is the mechanism through which it has been achieved.

An AdExchanger investigation published in November 2025 documented the specific mechanics. Many OEMs and streamers take as much as half of a publisher’s ad inventory by contract. They aggregate that share, repackage it as premium direct supply, and label the remainder as lower quality, even when the audience, content, and impression value are identical. I remember SpringSever Ad Server (now acquired by Magnite) even built a feature called “Inventory Router” that adapted the revenue sharing between content producers, distributors, OEMs, and other partners. Like the first ad in the ad pod belonged to HBO, the next one was Roku. The same ad break, the same household, the same moment of viewer attention, is classified as premium or remnant based entirely on which entity controls its sale. The entity that controls the OS controls the classification. The entity that controls the classification controls the price. The same OS operators also decide which bidstream signals are visible to external buyers. When impressions flow through the OEM’s own auction, data is abundant: genre, device ID, IP address, viewing history, household demographic. When publishers sell through independent channels, those signals often disappear or are sold back to the publisher as a fee-bearing add-on. Missing data depresses bid density even on quality inventory, further suppressing yields for publishers who try to operate outside the OS operator’s commercial architecture.

Tatari’s CEO Philip Inghelbrecht made the concentration point directly in a March 2026 interview: 90 percent of all streaming impressions placed for brands and agencies come from the same top ten publishers. In sports, anything that looks and smells like a playoff or a bigger game is not going to be found on The Trade Desk as an example. That inventory is sold directly, by publishers who prefer it that way and will only sell it that way, because the economics of direct selling are better than the economics of open-exchange bidding at these CPM levels. Jounce Media’s data adds a dimension to this: approximately 125 million US streaming devices per month flow through trusted direct supply chains. Tens of millions of additional devices appear in the bidstream through ad networks reselling inventory from other networks. Jounce’s conclusion is unambiguous: there are two types of resold CTV auctions, those that waste 20 percent of a marketer’s budget and those that waste 100 percent. The premium inventory that every advertiser actually wants exists in a concentrated, controlled environment. The open bidstream that DSPs access is partly real and partly fabricated.

This is the supply control problem in CTV, and it is the problem that The Trade Desk’s Ventura is designed to address. The architecture Ventura has chosen focuses on programmatic transparency: OpenPath for direct supply connectivity, Unified ID 2.0 for privacy-preserving identity, OpenAds for auction fairness, and OpenPass for single sign-on authentication. The Ventura Ecosystem, launched in February 2026 with V, the operating system formerly known as VIDAA powering more than 50 million Hisense and Toshiba devices globally, and Nexxen as founding collaborators, attempts to open the supply chain by giving OS partners direct access to this programmatic infrastructure. The logic is internally consistent and the ambition is real. Whether the current emphasis is sufficient to shift structural market power is a different question, one this article will return to.

What the Old Box Knew That the New Screen Has Forgotten

For most of television’s commercial history, the set-top box was the advertising infrastructure. NDS Group, founded in 1988 by scientists from the Weizmann Institute of Science and later acquired by Cisco in 2012 for approximately 5 billion US dollars before being spun out as Synamedia in 2018, built the conditional access systems and middleware that processed every signal from broadcaster to screen. Its VideoGuard conditional access system protected content in more than 114 million pay-TV households. Its MediaHighway middleware ran in more than 180 million set-top boxes cumulatively. The name changed when it became Synamedia. The structural problem it was built around, which is how video content reaches screens and who controls that process, remained.

The set-top box operator had something that no smart TV OS operator has fully replicated: sovereign control over the complete signal chain from content ingest to screen display. Content came in, the box processed it, the box modified the ad break, the box rendered the result. At no point did the operator need an application’s cooperation, a publisher’s SDK, or a streaming service’s agreement to implement a format. Sky’s AdSmart in the United Kingdom, launched in 2012, was delivering household-level addressable advertising in live linear broadcast when Netflix’s ad tier did not exist. Germany’s SevenOne Media and RTL Ad Alliance were running dynamic ad insertion in linear streams through set-top box software as early as 2016. The credit squeeze, where a broadcaster compresses end credits into a smaller window to create advertising space alongside, and the L-band overlay, where an advertisement occupies the left side and bottom of the screen while content continues in the remaining window, were deployed in broadcast environments using this architecture before most streaming executives had heard the terms.

The smart TV era disrupted this architecture without replacing its most important commercial capability. The television manufacturer now runs the operating system. Applications from Netflix, Disney, Peacock, and hundreds of others run on top. Content arrives through those applications. But the OS manufacturer does not control the video rendering pipeline of a Netflix stream the way a set-top box operator controlled the signal. The OS controls the home screen, the device identifier, what runs natively, and what data passes through at the device level. Formats like squeezeback require modifying the displayed video frame in real time, which requires either the streaming application to implement it deliberately or the OS to have rendering-pipeline control over third-party video playback that most smart TV platforms have not built as a cross-application service.

Roku’s position in this story is worth examining specifically because it is the most structurally interesting of any company operating in CTV advertising today. Roku is simultaneously a partner and a competitor to almost everyone. It maintains a UID2 data integration and supply partnership with The Trade Desk, while also operating OneView, its own DSP that directly competes with The Trade Desk. It struck a deal with Amazon in June 2025, announced at Cannes Lions, giving Amazon DSP exclusive access to Roku’s OS-level audience signals, creating combined reach across an estimated 80 million US CTV households, while Amazon Fire TV is Roku’s largest device-level competitor. It licenses its OS to TCL and Hisense, who are also being targeted by The Trade Desk’s Ventura Ecosystem. Roku reported approximately 85 million active accounts and platform revenue of approximately 2.7 billion US dollars in 2024, the majority from advertising. Nobody in CTV advertising can ignore Roku. Nobody can fully trust or fully rely on it either, which is a structural characteristic that has suited Roku quite well.

What the Publisher Can Control and What Belongs to the OS

One of the more consequential misunderstandings circulating in CTV discussions is the assumption that individual streaming applications, if sufficiently motivated, can implement the full range of advanced advertising formats on their own. They cannot, and understanding precisely where the capability boundary falls explains why OS-level solutions matter and why fragmentation has proven so persistent.

A streaming publisher operating its own application on a smart TV has genuine capabilities that are often underutilized. Within its own app, a publisher can implement pause ads by detecting when the viewer presses pause on the remote and triggering a full-screen or partial-screen ad overlay through its own player UI layer. Publishers like Netflix, Peacock, and Hulu have done this. A publisher can overlay QR codes on its creatives using its own rendering layer. It can run shoppable ad formats within its own content environment, allow viewers to interact with branded elements via remote, and optimize dynamic creative for its own first-party audience data. For in-stream ad replacement, any publisher that controls its own playout pipeline can use server-side ad insertion to replace ad slots dynamically, managing creative quality normalization and audience targeting within its own infrastructure. A publisher can also control frequency capping within its own application, prevent competitive adjacency within its own ad breaks, and report viewability from its own measurement instrumentation.

What a publisher cannot do unilaterally is substantial. Cross-application frequency capping requires OS-level visibility, because the application has no signal about what another application has already delivered to the same household in the same viewing session. Home screen placements, the highest-attention inventory on any device, are owned entirely by the OS operator and are not available to third-party publishers. Device-level ACR data, which captures viewing behavior across all inputs connected to the television regardless of which application or source is active, is collected and controlled by the OS manufacturer. This data is what makes Samsung Ads and LG Ad Solutions able to offer cross-platform audience intelligence that no individual streaming publisher can replicate. Screensaver ads, which run when the television is idle and no application is active, are OS-initiated events that individual publishers cannot control. Menu ads, which integrate into the TV’s navigational interface, are OS-owned surfaces entirely.

The formats that fall in the middle are the most interesting. A squeezeback, which reduces the content frame to approximately 60 percent of screen size while displaying an advertisement alongside, requires modifying the primary video display in real time. If the streaming application controls its own video player and has implemented the format within its own app, it can do this. But it must build this for every platform separately, because each OS has different rendering characteristics and different player implementations. There is no shared service. A pause ad that works precisely on Disney+ requires Disney to build it for Samsung’s Tizen, then build it again for LG’s webOS, then for Roku OS, then for Amazon Fire TV, then for Google TV. Standardization reduces the specification debate, but it does not eliminate the implementation work. An OS operator that built squeezeback, overlay, and L-band delivery as a cross-application platform service would eliminate that duplication for every publisher on its platform simultaneously.

Six Formats Standardized in December. Still Barely Deployed in April.

In December 2025, the IAB Tech Lab published its CTV Ad Portfolio establishing technical specifications for six advertising formats that exist outside traditional commercial breaks: pause ads, menu ads, squeezeback, overlay, in-scene insertion, and screensaver ads. The IAB collected more than 100 format descriptions from over 40 companies, including Disney, NBCUniversal, FreeWheel, Google, Yospace, The Trade Desk, and Tubi, as part of its Ad Format Hero initiative begun in 2024. The specifications went to public comment until January 31, 2026.

A TripleLift report published in March 2026 found that 49 percent of CTV advertising impressions are still delivered as traditional 30-second or 15-second spots. The same report found that campaigns combining high-impact formats with standard spots delivered 33 percent higher brand recall, 11 times higher brand consideration versus industry benchmarks, and 76 percent higher purchase intent. Engagement rates for interactive CTV formats reached 1.94 percent in Q2 2025 according to eMarketer, nearly double the rate from Q2 2024. Shoppable CTV formats have been measured converting at five times the rate of standard video in some deployments. The evidence that these formats outperform the dominant creative format is not ambiguous. The industry continues to run the dominant creative format.

The reason is partly commercial inertia and partly a genuine technical barrier. The commercial barrier is that 15-second and 30-second spots require no new creative production, no new agency workflow, no platform-specific integration, and no explanation to the client. The technical barrier is that advanced formats delivered consistently across every CTV platform require either OS-level implementation as a platform service or per-publisher implementation that must be repeated for every platform. Samsung’s GameBreaks, launched in 2024, are interactive ads that convert commercial breaks into branded mini-games within Samsung TV Plus, Samsung’s own FAST channel. They work because Samsung controls both the OS layer and the content layer within TV Plus. DirecTV Advertising’s SVP of Product noted in 2025 that pause ads showed strong results but advertisers were not scaling them in part because of the custom creative work required across different platform implementations. The IAB’s build-once, serve-everywhere aspiration for the six standardized formats is correct as a direction. Getting every OS operator to implement those standards uniformly requires each platform to deprioritize its own proprietary implementations long enough to build something that benefits competitors alongside its own channels.

Why Television Still Sells Itself in Upfronts, and What Video Technology Has to Do With It

The upfront market’s persistence is not purely institutional inertia. It reflects a technical reality about video advertising that is different from display advertising, and the OS layer is central to either perpetuating or resolving it.

In display programmatic, the impression is available in real time because the user’s browser loads a page, makes an ad call, and the creative is fetched after the auction resolves in roughly 200 milliseconds. Television advertising does not work this way. A live stream with a SCTE-35 ad cue marker firing signals that an ad break is approaching within seconds. The SSAI system needs an ad ready to stitch into the manifest before the break point arrives. The creative must be pre-transcoded and quality-normalized to match the content’s bitrate and resolution, and cached at CDN edge locations before the break fires. Yospace, which serves broadcast clients including Channel 4 in the UK, ITV, Fox, and DirecTV, has developed prefetch architecture specifically because just-in-time ad resolution alone is insufficient at scale during live events. Without pre-fetching, 90 percent of ad requests to the ad server occur within approximately 1.5 times a video segment duration of the break point. That is not the 200-millisecond auction window of display programmatic. It is a timeline that requires creative to be available at edge before the auction runs, which is the fundamental reason guaranteed buys and programmatic guaranteed structures dominate premium CTV rather than open real-time bidding.

The implication is significant. An open exchange auction for a live television ad break with the creative fetched post-auction works for on-demand content where pre-loading is feasible. For live sports, news, and premium linear programming, the inventory is either pre-sold with committed creatives, traded in programmatic guaranteed structures where the creative is submitted in advance, or it runs a fallback. This is precisely what FreeWheel’s Video Marketplace Report found in its 2025 data: over 70 percent of streaming ad views are still delivered directly, outside of programmatic channels. The 90-percent-programmatic figures that circulate in industry presentations include programmatic guaranteed deals, which are advance negotiations for reserved inventory at fixed prices with creatives submitted upfront, executed through programmatic infrastructure to automate delivery and reporting. They are insertion orders with better plumbing.

The OS layer is directly relevant because the OS is where format-level decisions can be enforced across all content sources on a device without depending on individual streaming applications. If a smart TV OS implemented squeezeback as a platform service with native rendering support and standardized SCTE signal interpretation, any streaming application on that OS could offer squeezeback inventory without building it separately. The creative would be rendered at the OS level. The format would be consistent across publishers. The buyer could purchase it programmatically because the creative specification would be uniform. No OS manufacturer has built this cross-application capability at scale, which is why the upfront model persists and why 49 percent of CTV impressions still run the same 30-second spot it has been running since television advertising was invented.

Why Private Marketplaces Are the Native Structure for CTV

The open exchange in display advertising succeeded because display impressions are cheap, fungible, abundant, and tolerant of post-auction creative fetching. None of these conditions apply to premium CTV advertising, which is why private marketplaces are structurally better suited to CTV than open auctions, and why the industry’s gradual migration toward PMPs reflects genuine market logic rather than advertiser timidity.

Pixalate’s Q2 2024 research documented that invalid traffic rates are 110 percent higher when SSAI is used in programmatic CTV compared to non-SSAI environments, and Jounce Media has found that a meaningful portion of the CTV devices appearing in the open bidstream are fabricated or misrepresented. For an advertiser spending 50 dollars per thousand impressions on premium CTV, a 20 percent fraud rate is not an acceptable rounding error. PMPs eliminate this exposure by restricting access to a curated, vetted group of buyers bidding on verified publisher inventory. The publisher knows who is buying. The buyer knows whose inventory they are buying. The fraud vectors that exploit the open exchange’s anonymity have no equivalent mechanism in a private deal.

The creative availability constraint also favors PMPs. Because premium video ad breaks require creative to be pre-transcoded and available at CDN edge before the SCTE-35 break point fires, the post-auction creative fetch that works in display programmatic creates a technical problem in live video. PMPs solve this because they include advance creative submission as part of the deal structure. The buyer commits a creative. The publisher quality-normalizes it and caches it. The SSAI system can confidently serve it when the break fires. eMarketer data shows that 76.2 percent of US programmatic digital display ad spending is expected to flow through programmatic direct channels, including PMPs, by 2026. In CTV, the proportion for premium inventory is substantially higher already.

PMPs also enable the data-sharing that makes CTV targeting genuinely superior to linear television. In a PMP, a publisher can share first-party audience signals, including content genre, program rating, viewing behavior, and household demographic segments, directly with the invited buyers. This level of signal richness does not travel through the open exchange bidstream, where data is often stripped, obfuscated, or sold back as a separate fee. The AMY Leifer at DirecTV has noted that direct buys, including programmatic guaranteed deals, hinge on a single publisher’s data, while PMPs allow buyers to layer specific audience targeting across a range of publishers, combining the data richness of a direct relationship with the targeting flexibility of a biddable environment. This is precisely the combination that premium CTV advertisers need, and it is the combination that the open exchange cannot reliably provide when the inventory involved is finite, pre-sold, and contextually sensitive.

SSAI Cannot Have a Client SDK and Why That Makes SGAI the More Interesting Conversation

An important distinction that gets lost in industry conversations about server-side ad insertion is definitional. SSAI, server-side ad insertion, is by definition a server-side operation. The advertisement is stitched into the video manifest by a server before the stream reaches the player. The player, whether running on Samsung Tizen, LG webOS, Amazon Fire TV, Roku OS, or any other OS, receives a continuous video stream and has no visibility into which segments are content and which are advertisements. There is no player SDK in traditional SSAI because the player is not involved in the ad logic. Yospace, which has served Tubi, DirecTV, ITV, Channel 4, and Fox as SSAI customers for more than a decade, runs its ad stitching entirely server-side. AWS MediaTailor, Google DAI, Amagi, and FreeWheel all operate the same way. SSAI vendors do not ship client SDKs as part of the core SSAI architecture because the architecture by design operates upstream of the client.

This matters for advanced ad formats. A pure SSAI approach to delivering formats like L-bands, squeezebacks, and overlays requires the server to composite the final video frame before delivery. The server resizes the content window, generates the L-shaped advertising space, inserts the creative into that space, and encodes the composite frame into the stream. The player receives and displays a stream that already contains the formatted output. InTheGame has demonstrated L-band delivery via server-side compositing in this way. The approach works on any device regardless of player capability, because the format logic is entirely server-side. The cost is computational: compositing frames server-side at scale, particularly during live high-demand events, requires substantially more processing infrastructure than standard ad stitching. But it is technically valid and commercially deployed.

The more promising architectural evolution for advanced formats is SGAI, server-guided ad insertion, which Yospace demonstrated at IBC 2025 as a world-first using the MPEG-DASH Events standard, achieving less than a second of advertising latency with glass-to-glass latency under five seconds for live streams. SGAI is a hybrid model: the server identifies ad break opportunities and guides the client player via standardized signals, while the client player handles ad fetching, format rendering, and measurement. Because the client player is involved in format execution, SGAI can support advanced formats that require rendering within the player’s layer, including squeezebacks, overlays, and pause ads, without requiring the server to composite the entire video frame. The tradeoff is that SGAI requires a compatible player implementation. A video player that does not support HLS Interstitials or MPEG-DASH Events cannot participate in SGAI, which means the format capability depends on the publisher’s player stack, not just the server-side infrastructure.

Yospace supports both SSAI and SGAI for compatible players, recommending SSAI combined with client-side measurement as current best practice, and SGAI as an evolution for publishers whose player environments support it. The emerging Common Media Client Data version 2 standard, which Yospace is contributing to, allows any compliant player to report playback position using metadata, providing measurement capability even without custom player code. This matters for advanced formats: measurement of a pause ad or squeezeback requires knowing not just that the ad was stitched, but that a real person on a real device viewed it in the specific format presented. SGAI combined with CMCDv2 is the current leading path to that combination at scale, but it remains an early-stage capability in production at relatively few publishers.

What SSAI and SGAI cannot do, regardless of sophistication, is replace the OS layer for the home screen, cross-application frequency capping, ACR data collection, or device-level attribution across all inputs. Those capabilities remain structurally owned by the OS operator. The server-side stack and the OS layer are complementary, operating at different layers of the same opportunity. The publishers who build SGAI capability into their players will have meaningful format differentiation. They will not have the cross-application data and inventory control that the OS operator holds by virtue of being the device’s operating system.

The OS Is Now the Product. The Television Is How You Deliver It.

By September 2024, LG’s webOS Summit in Incheon, South Korea, disclosed that advertising and services revenue on the webOS platform was projected to exceed KRW 1 trillion (approximately 760 million US dollars) in 2024, a fourfold increase from 2021. LG announced plans to invest a further KRW 1 trillion (approximately 762 million US dollars) in webOS development through 2027, with a stated target of growing platform-based services revenue fivefold by 2030, at which point the company projects platform revenue at 20 percent of LG Electronics’ total operating earnings. More than 600 global brands had adopted webOS Hub, LG’s commercial display platform. LG Channels distributes content across more than 4,000 channels in 33 countries.

Samsung TV Plus reached 88 million monthly active users globally in 2024, growing more than 30 percent year-over-year in Q1 2025. Samsung Ads, entering its tenth year of operation in 2025, renewed an exclusive global CTV partnership with Publica by IAS in September of that year. Samsung launched its Creative Canvas interactive ad format service in April 2025, becoming the first TV manufacturer to offer an in-house interactive ad format capability, and added home screen advertising inventory to Amazon Publisher Services in an integration LG Ad Solutions also established with APS in April 2025, enabling expanded programmatic demand for home screen placements through Amazon’s advertising infrastructure alongside their own channels.

Vizio’s CastOS, acquired by Walmart for approximately 2.3 billion US dollars in February 2024, grew 40.4 percent year-over-year in North American TV shipments in Q1 2025 according to Omdia. Walmart subsequently announced it would replace Roku’s operating system on its Onn private-label television brand with CastOS, ending Roku’s largest OEM partnership in a single commercial decision. Walmart’s interest in the television screen is retail attribution: connecting television ad exposure to Walmart.com purchases and in-store transactions through first-party purchase data. Amazon Fire TV sets grew 21.5 percent year-over-year in North American shipments in Q1 2025 in the same Omdia data. Amazon moved all Prime Video households to an ad-supported model in January 2024, immediately adding an estimated 50 billion additional impressions to the US market and forcing CPM corrections across competing streaming platforms.

TCL and Hisense, both targeting 30 million unit shipments each in 2025 according to Omdia, distribute their advertising monetization differently. TCL ships Google TV as its primary OS on most devices. Hisense ships at least seven different operating systems across its global lineup depending on market, including V (formerly VIDAA) where that platform has commercial scale. V’s 35 million US dollar exclusive deal with Nexxen, signed in August 2025 for ACR data access and exclusive North American ad monetization rights through at least 2029, represents a decision to outsource advertising capability rather than build it internally. Nielsen estimates that viewers spend an average of 10.5 minutes browsing the home screen before deciding what to watch, making it the most commercially valuable placement on any device. At the 2025 NewFronts, LG Ads CMO Tony Marlow described TV as becoming the launch pad to proactive, lean-forward experiences.

Ventura’s Correct Diagnosis, and the Architecture Question Still Ahead of It

The Trade Desk began building Ventura in late 2021, three years before Jeff Green announced it publicly in November 2024. The project was developed with a team that included former senior Roku employees and was built on the Android Open Source Project, the uncontrolled fork of Android that does not require Google certification. The platform was announced with support from Disney Advertising, Paramount Advertising, Tubi Media Group, and Sonos. The Sonos partnership collapsed in March 2025 when Sonos cancelled plans to produce a video streaming device. The DIRECTV UI collaboration announced in October 2025 is designed for third-party TV manufacturers and hospitality partners. The Ventura Ecosystem launched in February 2026 with V and Nexxen as founding collaborators. No public technical roadmap exists for Ventura beyond the programmatic stack: all publicly stated features are OpenPath, UID2, OpenAds, OpenPass, cross-platform content discovery, and subscription management.

The Ventura Ecosystem model, which connects existing OS operators to The Trade Desk’s programmatic infrastructure without requiring them to replace their own platform, is strategically more coherent than a full OS replacement pitch would be. It separates the supply-chain transparency question from the hardware distribution question. V retains its UI and its brand. The Trade Desk gains supply-path influence over 50 million Hisense and Toshiba devices it does not manufacture. The commercial incentive for OS operators is a share of incremental programmatic revenue that The Trade Desk’s buy-side relationships can unlock, without the OS operator having to build the programmatic infrastructure themselves. This is a genuine value exchange, not a theoretical one.

The supply-chain transparency work that Ventura is doing addresses a documented and serious problem. The ANA’s Programmatic Transparency Benchmark has found that significant portions of programmatic spend are absorbed by intermediary costs, fraud, and non-viewable inventory before reaching a real viewer. OpenPath creates direct supply connections that reduce those hops. OpenAds was developed specifically to address auction manipulation and improve fairness following Prebid’s controversial transaction ID changes in 2024. UID2 gives publishers and advertisers a privacy-preserving identity standard that works without third-party cookies. These are real solutions to real problems. Dentsu and WPP exited OpenPath in February 2026 citing transparency concerns and hidden fees, a characterization The Trade Desk disputed. This is a credibility challenge for a platform whose central proposition is supply-chain transparency, and it is one that Ventura will need to resolve if the Ecosystem model is to grow beyond its current founding members.

The more constructive observation is about what Ventura could add to its current architecture. The advanced creative format problem, which is the most significant unsolved technical and commercial problem in CTV advertising and the one most directly requiring OS-level participation at scale, is not currently addressed in any public Ventura feature or roadmap. The Ventura Ecosystem’s OS partners could, in principle, offer SGAI as a native monetization service through the Ecosystem, since Yospace and InTheGame both support SGAI and both have relationships across the broadcast and streaming ecosystem. A Ventura Ecosystem that brokered SGAI capability alongside UID2, OpenPath, and OpenAds would give its OS partners the infrastructure to offer publishers advanced format delivery without each publisher rebuilding it separately. Separately, if Ventura built server-side video compositing support for the six IAB-standardized formats as a Ventura platform service, OS partners could offer L-band, squeezeback, and overlay inventory programmatically through a single integration point, creating premium format inventory that commands higher CPMs than standard in-stream video, which would increase the value of every programmatic transaction flowing through OpenPath. The business case for adding this layer is internally coherent with Ventura’s stated mission. The public roadmap does not yet reflect it.

What the CTV industry needs from Ventura is not simply better auction mechanics on the same inventory. It needs the format expansion that the auction mechanics would price more efficiently if the formats actually existed at scale. The two problems are connected: supply-path transparency without format innovation produces a cleaner market for the same product. Format innovation without supply-path transparency produces better experiences that the market cannot efficiently price. Ventura is building one leg of the table. The other leg is still waiting for someone to build it.

Ventura correctly diagnosed that the OS is the leverage point in CTV advertising and built a programmatic infrastructure to act on that insight. Its next challenge is whether it will also invest in the video technology layer that would give its OS partners something genuinely new to sell — not just a cleaner version of the same 30-second spot.

Europe Moved First. The Industry Largely Ignored It.

The Hybrid Broadcast Broadband TV standard, HbbTV, is a pan-European open specification that allows broadcasters to blend traditional linear broadcast with internet-delivered interactive services. By the end of 2024, approximately 97 million European households were reachable on HbbTV interfaces according to Dataxis research, growing at a compound annual rate of 17 percent since 2020 and projected to reach 120 million by 2026. Germany’s SevenOne Media and RTL Ad Alliance have run addressable advertising over HbbTV since 2016, progressing from L-shaped banner overlays to seamless dynamic ad insertion in live linear streams. Austria’s AGTT, as described by Thomas Gruber of ProSiebenSat.1 Puls 4 Austria at the 2025 Connected TV World Summit in London, integrated HbbTV return-path data into linear campaign optimization that reduced zero-rated advertising impressions for the 12-to-49 demographic by 97 percent. Italy’s Mediaset implemented HbbTV’s addressable advertising extension with frame-accurate ad replacement guaranteed to millisecond precision.

Warner Bros. Discovery’s Italian deployment, announced December 2025 and developed with Amazon Web Services, combined HbbTV as the trigger layer with AWS MediaTailor for server-side ad insertion in live linear broadcast, producing what the company described as a tenfold increase in average ad-viewing session length. WBD is extending the model to Poland, Spain, and Germany. In the UK, Freely, launched in 2024 by Everyone TV, a joint venture between BBC, ITV, Channel 4, and Channel 5, uses the HbbTV Operator Application specification to deliver a seamless IP-based viewing experience that unifies live and on-demand channels via a common guide. The UK market also continues to have Sky AdSmart, which remains the most advanced addressable linear advertising system in the English-speaking world but operates specifically on Sky set-top boxes rather than open-standard smart TVs. The lesson from Europe is not that HbbTV solves everything uniformly. The WBD case study itself notes that each new country requires extensive device testing because HbbTV implementations vary across TV manufacturers. It is that when a device-level standard achieves sufficient scale, the advertising innovation enabled is genuinely different in kind from what server-side approaches or isolated OS proprietary implementations can deliver alone.

The Closing Argument, Delivered From a Hotel Ballroom in May

This month in New York, the 2026 upfront season is unfolding exactly as it has for seven decades, with the addition of better audiovisual production values and a significantly larger bar tab. NBCU is at Radio City Music Hall on May 11, marking NBC’s 100th anniversary. YouTube’s Brandcast is at Lincoln Center on May 13 with Trevor Noah hosting and Chappell Roan performing. Between the events, programmatic companies are in the hallways and sponsored breakfasts, representing the claim that CTV has finally made the transition to digital-quality buying. The claim is partly true and largely a rhetorical maneuver to participate in budget decisions that have historically belonged to linear television buyers.

The CTV advertising market reached approximately 33.35 billion US dollars in 2025 according to eMarketer, growing at 16 percent year-over-year, projected to reach 46.89 billion by 2028, at which point it will surpass traditional television advertising for the first time. These are the numbers everyone in those hotel ballrooms is citing. What the numbers do not reveal is that over 70 percent of the streaming ad views underlying that spend are still delivered directly outside programmatic channels according to FreeWheel’s VMR, that 49 percent of the creative formats are still 30-second spots, that supply concentration in premium CTV means 90 percent of impressions flow through ten publishers, and that the formats capable of genuinely differentiating CTV from linear television have been standardized by the IAB but not deployed at any meaningful scale.

The operating system is why all of these things are true simultaneously. The OS controls which inventory is classified as premium. The OS controls which data signals flow into the bidstream. The OS controls the home screen moment that is the highest-attention placement on the device. The OS is where cross-application advanced format delivery would need to be built to work consistently, and where no operator has built it. Samsung and LG have built the most complete advertising businesses of any OEM, with Samsung TV Plus at 88 million users and LG’s webOS advertising revenue crossing KRW 1 trillion (approximately 760 million US dollars) in 2024. Roku is simultaneously a supply partner to The Trade Desk, a data partner to Amazon, and an OS licensor to the same Chinese manufacturers that The Trade Desk’s Ventura Ecosystem is also pursuing. Walmart and Amazon have converted their retail infrastructure into OS businesses because the television screen, for them, is a point of sale connected to a purchase history dataset that no pure-play media company can match.

The Trade Desk’s Ventura has correctly identified that the OS is the leverage point. Its Ecosystem model is a pragmatic and architecturally sound approach to building programmatic infrastructure at the OS layer without requiring hardware manufacturing relationships. It now faces a strategic choice about whether to add the video technology and format delivery layer that would give its OS partners not just cleaner programmatic pipes but genuinely new advertising products to sell. SGAI, server-side video compositing, IAB format standardization as a platform service rather than a specification document: these are the capabilities that would convert Ventura from a better version of the existing market into something that could actually shift it. The OpenPath credibility challenge with Dentsu and WPP requires resolution. The technical roadmap requires expansion. And both of these things are achievable for a company with Ventura’s programmatic expertise and The Trade Desk’s balance sheet.

The set-top box operators of the 1990s knew that signal-chain control was the source of advertising power. They built formats, addressability, and dynamic insertion around that control. The smart TV OS operators of the 2010s and 2020s learned the same lesson and applied it at a different layer: device identity, home screen real estate, and cross-input data. The industry is in the middle of a third transition, one in which the publisher layer, the OS layer, and the server-side infrastructure layer are all capable of solving parts of the advanced format problem but none is solving it completely or consistently. The May 2026 upfronts will be sold and bought. The 30-second spots will run. The programmatic companies will claim their wins. And somewhere in a product roadmap that has not yet been published, or a partnership that has not yet been announced, or a standards working group that is still arguing about implementation details, is the beginning of something that will make all of this look like the set-top box era. The only question is who gets there first, and whether the industry will recognize it when they do.

In 1876, John Wanamaker, who had built one of America’s first department stores in Philadelphia, made the observation that would follow the advertising industry for the next 150 years: half of his advertising spend was wasted, and he could not figure out which half. The line became the most quoted complaint in the history of marketing not because it was clever but because it remained accurate for so long. You placed an ad in a newspaper and you hoped. You bought a television commercial and trusted a panel of 5,000 households to represent 330 million viewers. You ran a billboard and assumed that location, repeated exposure, and human memory were doing something useful. The whole expensive enterprise rested on inference.

Digital advertising was supposed to end all of that. The internet created, for the first time, a mechanism that could theoretically follow a consumer from the moment they encountered an advertisement to the moment of purchase, producing the causal proof that print and television could never provide. The industry built impressions and click-through rates and last-click attribution models, and spent a decade discovering that what it had built was good at measuring what happened after an ad and poor at measuring whether the ad had actually caused anything. A consumer who sees a sponsored link for shoes they already planned to buy and then buys them is not evidence that the ad sold the shoes. But the model said it was, and the budgets grew accordingly.

Retail media was the industry’s most compelling answer to this problem, and the argument for it deserves to be stated in its strongest form before being examined. A retailer occupies a position that no other advertising medium has ever occupied: it owns the transaction. When a shopper buys a product at Kroger, Kroger knows the shopper’s identity, purchase history, basket composition, frequency, price sensitivity, and the precise moment of conversion. That is not an inference about consumer intent. That is a receipt. If you can connect the ad to the receipt, you have finally given John Wanamaker his answer. The loop, the industry announced, was closed.

US advertisers spent $60.32 billion on retail media in 2025, according to EMARKETER’s December 2025 forecast, and will spend $71.09 billion in 2026, the fastest-growing major advertising channel in the world, expanding at roughly 18 percent annually against a broader market growing at 4.3 percent. The promise is on every slide deck and in every conference keynote: first-party data, purchase-intent audiences, closed-loop attribution, proof of ROI. The channel has attracted more than 200 retail media networks globally, with over 80 operating in the United States alone, ranging from Amazon Ads and Walmart Connect at the top to grocery chains including Albertsons Media Collective, Kroger Precision Marketing, and CVS Media Exchange; specialty retailers like Home Depot’s Orange Apron Media, Best Buy Ads, Lowe’s One Roof Media Network, and Sephora Media Collective; and smaller but notable networks including Sam’s Club’s Member Access Platform, Walgreens Advertising Group, Macy’s Media Network, Kohl’s Media Network, 7-Eleven’s Gulp Media, Ace Hardware’s RedVest Media launched in 2025, and Petco’s emerging network. Every retailer with a loyalty programme and a website is building one, and the commercial logic is obvious: advertising generates 50 to 70 percent operating margins against the 3 percent that the underlying retail business produces.

This Tuesday, April 14, the Interactive Advertising Bureau (IAB) convenes its Connected Commerce Summit in New York, the industry’s most focused annual gathering on the specific business of retail media; and the session titles that appear on the agenda are more honest than anything that tends to appear in a keynote at a retail media launch event. Executives from Colgate-Palmolive, Mondelēz, Bayer, Monster Energy, and Mars Wrigley will appear alongside platform leaders from Target Roundel, Best Buy Ads, Grocery TV, Dollar General, 7-Eleven, and Ace Hardware’s newly launched RedVest Media network to discuss sessions titled, with a directness that the industry has usually reserved for private conversations: ‘Making Omnichannel Measurement Real’, ‘Building the Transparent Commerce Media Tech Stack’, and ‘Connecting Off-Site Media Signals to Sales Outcomes’. The final panel of the day is titled ‘The Great Debate: Will Retail Media Be the Casualty of AI-Driven Commerce?’ — a question that, as this article will attempt to demonstrate, would have been considered professionally hazardous to ask in public as recently as two years ago. The right questions are finally being asked. The fact that they are being asked at the IAB’s industry summit rather than answered by the retailers selling the product is the point.

Bain and Company surveyed advertisers to determine how many trust the measurement numbers they receive from retail media networks. The answer is 6 percent. Not 60. Not 16. Six. The industry running on the most data-rich advertising proposition in the history of the medium is being trusted, by the people spending the money, at about the rate you would trust a stranger you met ten minutes ago. Wanamaker’s problem has not been solved. It has been monetised at scale.

“Only 6% of advertisers fully trust retailers’ reported media metrics.” — Bain and Company, 2025

· · ·

The measurement crisis begins with a methodology called last-touch attribution, and understanding it is the key to understanding everything that follows. An analysis of the top sixteen ad types across the six largest retail media networks in the US (Amazon Ads, Target Roundel, Walmart Connect, Kroger Precision Marketing, Criteo’s commerce media platform, and Instacart Ads) found that 80 percent of them assign 100 percent of the credit for a purchase to the final advertisement the buyer encountered before completing the transaction. Every other touchpoint in that buyer’s journey, the television commercial they saw six months ago that introduced the brand, the social media post that kept the brand salient, the display ad that appeared when they were considering switching to a competitor, is credited with zero.

The critical flaw in this methodology is not that it measures the wrong ad. It is that it frequently credits the ad for a sale that was already happening. Consider a consumer who has bought the same protein bars every three weeks for two years. On Thursday, Kroger Precision Marketing serves them a sponsored product for those bars. On Saturday, they buy the bars, as they were always going to. KPM’s attribution model records a successful campaign. The actual question, did the ad change this consumer’s behaviour in any way, was never asked, because answering it would require comparing the consumer’s behaviour against a control group that never saw the ad. That methodology has a name: incrementality. And it is precisely what the industry has been avoiding.

This avoidance is not accidental. Incrementality testing produces lower ROAS numbers than last-touch attribution, because it strips away the organic demand that the last-touch model was claiming as advertising success. If a brand’s loyal customers would have repurchased without any advertising at all, which many of them would, then the ROAS attributable to genuinely incremental sales is substantially lower than what the platform dashboard shows. Skai and Stratably surveyed 166 retail media advertisers for their 2026 State of Retail Media report and found that while 71 percent of brands now consider incrementality their most important KPI, only 20 percent are good at both measuring it and acting on the results. The remaining 80 percent are making budget decisions against numbers they cannot independently verify and privately do not trust.

The distinction between retention and acquisition campaigns makes this problem even starker, and it is a distinction that retail media’s standard measurement frameworks have almost entirely ignored. Retention campaigns reaching existing loyal buyers to maintain their purchase frequency look outstanding on a ROAS metric because the buyer was going to purchase anyway, and the ad gets full credit. Acquisition campaigns reaching consumers who have never bought the brand, or who are actively considering switching from a competitor require proving genuinely incremental demand, because there is no baseline purchase pattern to misattribute. They look worse on ROAS not because they are worse advertising but because they are doing harder work. The 2025 ANA study on retail media found that 62 percent of retail media budgets go to audiences who are already brand buyers. The brands spending $71 billion on retail media are using more than half of it to advertise to people who were already their customers, measuring the effectiveness of that spend with a methodology that guarantees a favourable result, and calling the combination a closed loop.

PepsiCo’s test with Skai capabilities across Amazon DSP, where the company specifically optimised for new-to-brand customers rather than existing buyers and measured results using incremental ROAS, unlocked over 80 percent new-to-brand ROAS, a metric that looks different from, and less flattering than, overall ROAS precisely because it is honest. That test is instructive not because PepsiCo discovered some breakthrough but because it demonstrated how different the numbers look when the measurement methodology is designed to find the truth rather than confirm the budget allocation. Jason Wescott of WPP Media has said the overreliance on ROAS as the benchmark of value is over, and independent, transparent measurement is the baseline. CJ Pendleton, Chief Strategy Officer at Matrixx CPG, has named the commercial consequence without diplomatic softening: the platforms that solve for incrementality will earn the lion’s share of CPG investment while those that don’t will likely die on the vine.

· · ·

Understanding why the measurement crisis persists requires understanding what retail media’s first-party data actually is, and what it is not, across the spectrum of networks that are selling it. At the top of the market, the data proposition is real. Amazon’s $68.63 billion in global advertising revenue in 2025, confirmed by the company’s Q4 earnings release, with $21.32 billion in that quarter alone is built on direct transaction data from hundreds of millions of consumers who have bought on Amazon’s platform. Amazon knows your purchase frequency, your basket composition, your price sensitivity, your device preferences, and the moment your repurchase window opens. Walmart, which generated $6.4 billion in global advertising revenue in 2025, up 46 percent year-over-year per its February 2026 earnings, is building toward comparable depth: CFO John David Rainey noted that advertising and membership together accounted for a full third of Q4 2025 operating income, describing a business where advertising has become structurally important to profitability rather than incidental to it.

But Amazon and Walmart together hold approximately 87.7 percent of US digital retail media advertising spend. Amazon at roughly 79.7 percent and Walmart at about 8 percent, per EMARKETER’s December 2025 platform analysis. EMARKETER further projects that approximately 89 percent of the $10.77 billion in incremental retail media growth in 2026 will go to those same two companies. That leaves the other 80-plus US retail media networks competing for a slice of the market that, in incremental terms, is barely growing for anyone else. They are competing for smaller budgets with weaker data infrastructure, using the same first-party data marketing language as Amazon, and delivering a product that Georgia-Pacific, the consumer goods company, described in terms that should give every CPG brand media buyer pause.

Georgia-Pacific evaluated approximately 40 retail media networks over several years as it shifted more than 20 percent of its total media budget to the channel. Its Digital Marketing Director’s conclusion, reported in 2024, was precise and devastating: they told you they had first-party data, but they were not always giving first-party data to activate against them. The company was not describing an experience with a single bad actor. It was describing a pattern across a systematic evaluation of 25 networks it trialled. The data being sold as first-party was, in a material number of cases, something else: modelled segments, third-party augmentation, or loyalty data with coverage gaps so significant that the targeting precision implied in the pitch bore little relation to what was actually available.

EMARKETER identified the structural reason in February 2026: even genuinely robust first-party datasets reflect only shoppers who have engaged with the retailer’s loyalty or login system. That excludes irregular buyers, lapsed customers who still influence category dynamics, high-value prospects who are in-market but have never used that retailer’s card, and entirely new-to-category consumers who represent the actual growth opportunity for most CPG brands. A brand trying to find new customers which is the primary purpose of brand marketing investment is using targeting data that systematically excludes the people it most needs to find. The closed loop that retail media promises is real for the customers who are already inside it. For everyone outside it, retail media is programmatic display with better demographic labels.

The way retailers build audience segments compounds this problem in ways that rarely surface in a sales meeting. The standard taxonomy available across most retail media networks (heavy category buyers, lapsed purchasers, competitive brand switchers) consists of demographic buckets with transaction labels attached, and those constructs were designed for programmatic display targeting in 2010. A retailer with years of weekly purchase history across tens of millions of loyalty members has, in principle, the raw material to build something genuinely different: individual-level purchase cycle models that predict when a specific household is entering its consideration window, price elasticity signals that identify the moment a loyal customer is vulnerable to a competitor’s promotion, declining frequency patterns that flag at-risk customers before they switch. The most sophisticated networks like Kroger Precision Marketing through its 84.51 analytics arm, which works with purchase data from over 60 million loyalty households, are beginning to build in this direction. Most networks are selling the same five segments to every brand that will buy them.

· · ·

The audience segmentation problem becomes more severe, not less, when retail data is used to target audiences outside the retailer’s own properties. Off-site retail media, the practice of using a retailer’s first-party data to serve ads on external publisher websites, connected television, programmatic display, and audio is the fastest-growing component of the channel. In Q4 2025, 60 percent of Walmart Connect’s self-serve display spend went to offsite inventory, per Tinuiti’s Digital Ads Benchmark Report. Target’s Roundel reports that more than 30 percent of partner media spend now happens off its owned platforms. The retailers’ audiences are being used to buy ads everywhere. The question of whether the measurement travels with them has a troubling answer.

CTV is where the audience segmentation problem is most visible, and most consequential. Retail media’s extension into connected television is one of the industry’s most celebrated developments: Amazon’s Prime Video ad inventory, which delivered ads to an average of 315 million viewers globally in Q4 2025; Walmart Connect’s CTV capability through the $2.3 billion Vizio acquisition; Kroger Precision Marketing’s partnership with Magnite to extend into CTV. The proposition is genuinely compelling. Use purchase-based audience data to reach household-level audiences in the most engaging media environment available. The measurement reality is described, with characteristic bluntness, by Advertising Week’s 2026 analysis of CTV performance: measurement is still stuck two or three years behind the channel’s growth, and most CTV campaign reporting is comparing apples and oranges.

Comscore’s 2026 State of Programmatic Report, based on more than 200 media buyer respondents, found that 87 percent say cross-channel performance metrics inside programmatic platforms are critical or valuable for decision-making, and simultaneously that CTV measurement challenges like inconsistent reporting windows, limited cross-device visibility, and fragmented clean room integrations, remain significant barriers. The problem is not that CTV is unmeasurable. It is that the audience segment that entered the CTV buy as a retail-data-defined group of heavy category buyers cannot be tracked through the living room, through the consideration period, and back to a verified purchase at a retail location, without a data infrastructure connecting all three environments that most networks have not yet built. Kroger’s partnership with Magnite is a step toward it. The measurement framework that closes that loop does not yet exist at scale.

Audio faces the same fundamental challenge, with an additional layer: the format is inherently harder to connect to a purchase outcome because a consumer listening to a podcast while cooking dinner is not in a purchase moment the way a consumer on a search results page is. Programmatic audio is growing as Comscore’s 2026 report projects audio will capture 10 percent of programmatic budgets on average, with 21 percent of marketers reallocating from linear radio. Comscore itself launched audio targeting and measurement capabilities with The Trade Desk in January 2026, providing contextual targeting across 4.6 million podcasts and campaign measurement without relying on identity signals. The measurement progress is real. The gap between a retail-defined audience segment applied to a streaming audio buy and a verified purchase outcome at a physical or digital retail location remains wide enough that most brands treating audio as a performance channel are doing so on attribution windows that cannot distinguish advertising causation from coincidence.

Programmatic display, the channel that retail media was supposed to supersede with its superior first-party data, has quietly absorbed the segmentation problem rather than solved it. When a retailer’s audience segments are activated through The Trade Desk, DV360, or any other major DSP and served as display impressions across the open web, the impression appears on a publisher’s page, the cookie or identifier connects to the retail audience profile, and the attribution model waits for a subsequent purchase on the retailer’s platform. If the purchase occurs on a different retailer like the consumer saw Kroger’s display ad, considered the product, and bought it on Amazon because the price was lower, Kroger’s model counts a miss while the sale happened regardless. If the consumer was already planning to buy, Kroger’s model potentially counts a hit for a behaviour the ad did not influence. The closed loop that retail media promised was closed at the transaction point. It remains open everywhere the ad ran before the transaction.

Topsort, the commerce media platform that partnered with Skai in September 2025 to enable unified access to retail media networks across 40 countries through a single API integration, is building toward the cross-channel measurement consolidation that these problems require. Its architecture treats the retailer audience as a portable targeting signal rather than a siloed network property, which is the structural shift that cross-channel retail media measurement needs. Pentaleap, which in July 2025 launched real-time bidding for sponsored product ads through its Teads partnership enabling, for the first time, fully programmatic sponsored products on retailer sites is creating the supply-side transparency that makes cross-channel attribution auditable. These are meaningful technical contributions. They are building components of the infrastructure that the measurement problem requires. The full infrastructure does not yet exist anywhere except at Amazon, and only partially at Walmart.

· · ·

The most interesting and least honestly discussed extension of retail media is the physical store, and the timing of its emergence as an industry priority reveals something important about the online channel’s maturity. When onsite inventory runs out, when every available sponsored product slot on a search results page is sold and the user experience of adding more would drive shoppers away, the channel’s next move is either offline or inside someone else’s inventory. Both have happened simultaneously. The offsite expansion into CTV and programmatic display has been widely discussed. The in-store screen expansion has received less scrutiny, which is unfortunate, because it is where the measurement crisis is most acute and where the parallel with a $200 million failure in a Dallas warehouse is most directly relevant.

Retail stores were advertising’s original closed-loop environment. The shopper arrives with a purchase intent, encounters a display, makes a decision, and the transaction happens within the same four walls. Brand managers have understood this since the 1950s, and the physical store, the endcap, the in-aisle display, the checkout lane has always been the most valuable last-mile advertising surface available to CPG brands. What retail media has now added is the possibility of making that surface programmatic: digital screens in the aisle that can be targeted, measured, and optimised in near-real time, connected to the same loyalty data infrastructure that powers online retail media. This is the promise of Digital Out-of-Home within the retail environment, a $7.4 billion market in North America in 2026, according to Fortune Business Insights, and it represents a genuine opportunity that traditional DOOH advertising has historically left uncaptured because it was not connected to transaction data.

Kroger committed in June 2025 to a nationwide expansion of in-store digital screens through its partnership with Barrows Connected Stores, with Christine Foster, SVP of Kroger Precision Marketing, noting that the grocer sees the physical store as one of the most underutilised platforms for brand storytelling. Albertsons launched a digital display network in 80 stores in July 2025 through Stratacache. Hy-Vee deployed more than 10,000 screens across 400-plus locations with Grocery TV. Best Buy announced full-store takeover packages. Even Save-a-Lot is plotting an in-store rollout across 900 locations. Paul Brenner, SVP of retail media at In-Store Marketplace, reported in January 2026 that his firm had 20 active RFPs from retailers for 2026 screen deployments, some requesting thousands of units per retailer. The investment thesis is clear: over 91 percent of food and beverage purchases still happen in physical stores, yet 99 percent of retail media advertising has been online. The physical store is the single largest measurement gap in the entire retail media landscape.

The DOOH connection matters here because it is the technical framework that makes in-store screens programmable and measurable at scale. A screen in the frozen food aisle, running on DOOH infrastructure, can in principle be served to a shopper whose loyalty profile indicates they have not purchased a particular brand for six weeks — a lapsed buyer, targeted at the precise moment they are standing in front of the product. That is a targeting capability that outdoor DOOH on a highway billboard can never match, and it is the reason that the smart money in advertising infrastructure is watching in-store retail media very carefully. When a screen at the Kroger dairy case can trigger a sponsored display based on the household loyalty profile of the shopper scanning their app at the same time, and when the subsequent basket scan closes the attribution loop at the same transaction, the closed-loop measurement that online retail media has been claiming becomes something it actually can deliver.

The reason it has not delivered this yet is the Cooler Screens lesson, made permanent in a Dallas warehouse. Cooler Screens installed 10,300 smart refrigerator door screens across 700 Walgreens locations, claiming 100 million monthly impressions from sensors that turned out to be counting bathroom trips, shopping carts, and shadows. The screens generated $215 per door annually against a contractual minimum of roughly double that. When CEO Arsen Avakian cut data feeds to 100 stores during the December 2023 holiday season in a contract dispute, screens went white across refrigerator aisles at the industry’s most commercially sensitive moment. The $200 million lawsuit followed. The hardware, nearly $50 million worth, is sitting in Texas. The measurement system had been counting the wrong things, calling them impressions, and reporting the result as evidence of a closed loop. In-store retail media, run correctly, could finally deliver what the DOOH industry has always promised and what grocery advertising has always needed. Run without genuine measurement infrastructure, it produces lawsuits and warehouses.

Christine Foster’s comment about not wanting the screens to feel like Times Square is the tension that the industry has not resolved. A screen designed to be invisible is not a high-attention advertising placement. A screen designed to attract attention is a disruption to the shopping experience the retailer depends on. These two goals are not reconcilable without measurement that tells you which placement intensity produces the best combination of advertising effectiveness and shopper retention which is precisely the measurement that the IAB’s December 2025 in-store standards framework, which defines store zones and impression metrics but does not address individual-level attribution through loyalty data or cross-channel purchase journeys, does not yet require.

· · ·

Criteo, the independent commerce media company that is probably the industry’s most instructive case study, is not a retailer. It is a technology platform operating both as a demand-side platform helping brands buy across retail media inventory and as a supply-side platform helping retailers monetise their audiences and it built the technical infrastructure on which a significant share of the industry runs. Criteo is not Amazon or Walmart. It is the middleware. That distinction matters enormously for understanding its measurement position, because Criteo sits in the measurement gap itself: it processes transactions between retailers and brands, which means it sees both the ad delivery and the attributed purchase signal, but it does not own either the retailer’s closed-loop data or the brand’s independent measurement. In September 2025, Criteo became Google’s first onsite retail media partner through Search Ads 360. The announcement positioned this as a significant expansion. It also reveals that an independent infrastructure company, after building integrations with more than 200 retailers globally and completing a $380 million acquisition of IPONWEB in 2022, needed Google’s advertiser demand to make its retail media infrastructure commercially viable. That is not a failure. It is a structural reality about where the leverage in retail media actually sits.

The agencies that manage the advertising investment of the brands spending these billions are navigating the same structural terrain, with varying degrees of conflict. Publicis Groupe, which reported 5.9 percent organic growth in Q2 2025 and maintained its position as the industry’s growth leader, owns both CitrusAd, a retail media technology platform powering multiple retailer networks and Epsilon’s data infrastructure, which powers audience segmentation for clients including Kroger. Publicis is the agency advising brands on which retail media networks to trust, while simultaneously operating the technology those networks run on and the data infrastructure those networks use. CEO Arthur Sadoun’s framing of the company’s 2026 strategy as building agentic solutions that deliver business outcomes at a moment when 95 percent of AI projects fail is directionally accurate. The structural position his company occupies in the retail media ecosystem is worth noting alongside it. WPP’s $60 billion annual media investment operation, operating under the WPP Media brand after retiring GroupM in 2025, is in a different position: managing at massive scale without the same proprietary data infrastructure, which creates both a cleaner measurement posture and a more exposed commercial position in a world where data ownership is becoming the primary competitive variable.

The $60 million FTC settlement that Instacart reached in December 2025 is, on the surface, a consumer protection story: the commission alleged that the company’s advertising and billing practices misled consumers with promises of free delivery that masked mandatory fees adding up to roughly 15 percent of order value. But it connects to the retail media measurement story in a specific way. Instacart is one of the larger retail media networks outside the Amazon-Walmart duopoly, generating close to $1 billion in annual US advertising revenue according to EMARKETER projections. Its advertising proposition to CPG brands rests on the same closed-loop claim as every other network: reach high-intent grocery shoppers at the moment of purchase, measure results through Instacart’s transaction data. The FTC’s finding that Instacart’s representations to consumers were not accurate about its own fees, the basic facts of how much the platform costs, raises a question about whose methodology is being used to measure everything else the platform claims about the effectiveness of the advertising it sells. A platform that was found to have misrepresented its cost structure to consumers was simultaneously providing CPG brands with closed-loop performance data about the advertising those same consumers were shown. The FTC settlement is about consumer protection. The measurement implication is about advertiser protection.

The Instacart case is the Cooler Screens pattern at a different level of the ecosystem. Both companies built their retail media products on a specific measurement claim. Cooler Screens on impression counting that was measuring bathroom trips; Instacart on a consumer transparency standard that the FTC determined was not being met. And both cases ultimately revealed that the numbers being used to justify advertising investment were produced by parties with competing interests in what those numbers showed. This is the structural condition that produces the 6 percent trust figure. It is not that retailers and platforms are deliberately falsifying data. It is that the entity producing the measurement has a financial interest in the measurement being favourable, and the entity receiving the measurement has no independent mechanism to verify it.

This is precisely where demand-side and supply-side platforms, the DSPs and SSPs that sit between advertisers and retail media networks, have an underutilized and under-appreciated role to play. Because independent DSPs are not retailers and do not own the inventory they are buying, they occupy a structurally different measurement position than the retail networks themselves. When The Trade Desk runs a campaign across Walmart Connect, Kroger Precision Marketing, and Instacart simultaneously, it has cross-network impression data that none of those three networks can see independently, data that, in principle, allows a brand to detect frequency overlap, audience duplication, and attribution double-counting that individual network dashboards systematically obscure. The problem is that The Trade Desk’s Retail Index marketed as a neutral benchmark for cross-network performance is produced by the same DSP routing spend through those networks, with financial agreements in place with the data providers whose segments are being scored. Grading your own homework is a problem at the retailer level. Outsourcing the grading to your DSP, whose revenue depends on routing your spend efficiently, does not fully solve it. Andrew Casale of Index Exchange has said publicly that some supply-side vendors extract more value than they create is a notable observation from the CEO of an SSP, and one that applies with equal force to some DSPs. What the independent DSP ecosystem could provide, if it chose to build the capability, is cross-network attribution that no single retail media network can produce: a unified view of where an advertiser’s spend went, what audiences it actually reached in deduplicated terms, and which network attributed the same consumer’s purchase as its own success. Rajeev Goel of PubMatic, whose AgenticOS platform is building toward agent-driven media buying, has described the value chain compression that agentic AI enables as an opportunity to create more transparent, direct connections between buyer and seller. The measurement opportunity in that compression is real. The DSPs and SSPs that build toward it, rather than routing around the transparency question, will earn a position in the measurement ecosystem that the retailers and their networks cannot credibly occupy.

· · ·

Adobe Analytics reported that AI-driven traffic to US retail sites increased 805 percent year-over-year on Black Friday 2025 and 670 percent on Cyber Monday. ChatGPT now drives more than 20 percent of referral traffic to Walmart.com, approximately 15 percent to Target, and 10 percent to eBay, per a December 2025 MetaRouter analysis. Salesforce reports that 39 percent of consumers, and more than half of Gen Z, already use AI for product discovery. These numbers are still small in absolute terms relative to human-browsed retail traffic. They are not small in directional terms.

The sponsored product being the foundational revenue unit of retail media, projected to account for $38 billion in advertiser spend in 2025 is a format built for human browsing. A consumer loads a search results page, sees a placement, processes the creative, and decides. An AI agent managing a household’s weekly grocery order does not load a search results page. It queries the inventory API, checks price and availability, cross-references reviews and nutritional data against stated preferences, and completes the transaction without ever rendering a sponsored product or registering an impression in any attribution model. The ad never ran. The attribution window opened and closed around a transaction it could not see.

This is where Generative Engine Optimisation connects directly to the measurement problem, and the connection is not a trend analogy. It is a causal relationship. GEO as the practice of ensuring that product data is complete, structured, and machine-readable so that AI agents can confidently discover and recommend it addresses exactly the data quality gap that has made retail media measurement unreliable from the beginning. The reason last-touch attribution dominates retail media is partly technical inertia, but it is also a symptom of insufficient data discipline: if the signal connecting an ad impression to a purchase outcome were as clean and structured as a product API response, incrementality testing would be computationally trivial. GEO forces brands to build the data infrastructure that makes products findable by agents. That same infrastructure, applied to the advertising side, makes ad exposures linkable to purchase outcomes with the kind of precision that closed-loop measurement has promised and not delivered. The brands building GEO discipline are, almost without realising it, building the foundation for measurement that actually works.

Walmart’s Sparky, the agentic shopping assistant that reached half of all Walmart app users and generated 35 percent higher average order values than standard search, has already begun testing sponsored prompts inside its conversational interface that an advertiser can pay to be the recommendation Sparky surfaces when a consumer asks for a product category. Walmart CEO Dave Guggina said on the Q4 2025 earnings call that the company is learning as it goes on how advertising works in agentic commerce. That honesty is more valuable than any measurement dashboard Walmart could have published instead. Amazon’s Rufus assistant, with 300 million users and $12 billion in attributed sales in 2025, is facing the same question. The retailer that invented retail media and the retailer that followed it most closely are both trying to figure out what an ad looks like when the shopper cannot see it.

· · ·

The measurement problem is solvable. This is the critical point, because the failure to solve it is commercial rather than technical, and commercial barriers move when the financial incentives change. The infrastructure that would genuinely close the loop has been described accurately by the industry’s own standards bodies, the IAB and MRC published retail media measurement guidelines in January 2024 but guidelines describe what should happen, not what does. What does happen is that the entity producing the measurement has a financial interest in that measurement being large, and the entity receiving the measurement has no systematic mechanism for independent verification. There is a phrase for this in education: grading your own homework. In advertising, it has been the norm for thirty years. In retail media, which was supposed to solve the problem, it has become the structural foundation.

A small ecosystem of independent measurement companies has built the tools that could break this cycle, and the most interesting thing about them is how small they remain relative to the problem they are trying to solve. Haus, founded by former Google and Facebook data scientists and backed by meaningful venture capital, runs incrementality measurement through geo-based holdout experiments: it divides a brand’s geographic markets into exposed and unexposed groups, measures the difference in sales outcomes, and produces an estimate of genuinely incremental lift that is not filtered through the retailer’s attribution model. Measured, which raised $21 million in Series B funding and works with hundreds of brands across their full media portfolio, offers multi-touch attribution combined with incrementality testing, allowing brands to evaluate not just whether retail media is working but how it compares against other channels competing for the same budget dollar. Northbeam applies machine learning to attribution modeling across channels, helping brands understand which combination of touchpoints actually contributed to a sale rather than which single touchpoint the last-touch model credited. NCSolutions, which connects advertiser campaign data to actual purchase outcomes using purchase-based measurement drawn from over 150 million US households via retail loyalty programmes, operates at the intersection of panel-based measurement and first-party retail data — providing a verification layer that does not require the retailer’s own attribution system to be trusted. Neustar, now a TransUnion company, offers campaign measurement and media mix modelling that treats retail media as one input among many rather than a closed system that defines its own success.

These companies exist. They work. The brands that use them describe, consistently, a gap between platform-reported ROAS and independently measured incrementality that is large enough to change budget allocation decisions. Ryan Verklin, Retail Media and Paid Media Senior Lead at Bayer — who will participate in the IAB Connected Commerce Summit’s ‘Great Debate’ panel next week — won AdExchanger’s Best In-House Media Operation award specifically by pioneering self-serve retail media with incremental measurement at the centre of the strategy rather than the margin. The capability exists. It is not scaling. And the reason it is not scaling is a cascade of commercial barriers that are worth naming precisely.

The first barrier is data access. An independent measurement firm running a holdout experiment needs impression-level exposure data from the retail media network, specifically, a list of which consumers were served the ad and when, linked to purchase outcomes after the campaign. Most retail media networks do not provide this data to third parties. Amazon Marketing Cloud offers impression-level data within its clean room environment, which means Haus or Measured can run holdout analysis there but only within AMC’s infrastructure, under AMC’s terms, using AMC’s identity graph. This is meaningful progress and simultaneously a measurement system that still relies, at the data layer, on the entity being measured. Kroger Precision Marketing, to its credit, has been more forthcoming with clean room access for external measurement partners than most networks its size. Most networks below the top tier provide aggregate reporting with limited granularity and no access to individual impression data that would support true holdout analysis.

The second barrier is commercial inertia. A brand that is receiving ROAS numbers of 5 or 6 from a retail media network dashboard, and has not run an independent incrementality test, has no empirical reason to doubt those numbers except for the intuition, increasingly widespread, that they seem too good to be consistently true. Running an incrementality test requires budget, time, internal analytical resources, and the willingness to accept a result that might require going back to the CFO and explaining why the advertising that was reporting excellent returns has been producing lower incremental sales than the dashboard suggested. The brands that have run these tests tend not to publicise the results, because the results frequently require awkward conversations with retail media network account teams. The brands that have not run these tests tend to remain comfortable with the dashboard numbers, because the alternative is uncomfortable. This is not irrational behaviour. It is rational behaviour in an environment where the measurement system was designed by a party with an interest in large numbers, and where the cost of discovering the truth includes the cost of revising a budget allocation that current stakeholders have committed to.

The third barrier is scale. Haus, Measured, and Northbeam are formidable technical products serving sophisticated brands. They are not The Trade Desk, which reported nearly $3 billion in annual revenue. They are not IAS or DoubleVerify, which have achieved the market position and MRC accreditation to be included in standard campaign specifications at major agencies. IAS earned its first MRC accreditation for retail media measurement specifically viewability within Amazon’s ecosystem in November 2025, nearly a decade after IAS was founded. MRC accreditation for incrementality measurement within retail media networks, which is the certification that would actually address the 6 percent trust problem, does not exist. The MRC accreditation process takes approximately two years and costs hundreds of thousands of dollars per network. An independent measurement firm that wanted to achieve MRC-accredited incrementality certification across the ten largest US retail media networks would be looking at a multi-year, multi-million-dollar credentialing exercise before any agency could include it in a standard measurement specification. The economics of that path are not attractive for a venture-backed startup competing with platforms that have unlimited marketing budgets and brand recognition built over decades. The measurement ecosystem needs these companies to be significantly larger than they are, and the path to being significantly larger runs through advertiser mandates that do not yet exist, which requires advertiser sophistication that is still being built, which requires campaign results that are still too often treated as proprietary, which means the ecosystem stays smaller than the problem requires.

Closing the loop on this cascade requires, first, genuine holdout testing: a portion of the target audience is held back from seeing the ad, and the difference in purchase rates between exposed and unexposed groups is measured. Second, attribution windows must be set empirically based on documented purchase cycles for specific product categories, not selected by the platform to maximise attributed conversions, and disclosed before the campaign runs. Third, and this is the requirement that the industry’s most powerful players have the least incentive to adopt, the measurement itself must be certified by parties with no financial relationship to the platforms being measured. Integral Ad Science’s November 2025 MRC accreditation for viewability measurement within Amazon’s ecosystem is genuine progress. It measures viewability, which is the floor of advertising standards: it tells you the ad was technically capable of being seen by a human. It does not measure incrementality, which is the ceiling: it does not tell you whether seeing the ad changed what that human did. The gap between viewability and incrementality is the gap between the measurement the industry has accredited and the measurement it actually needs.

Fourth, and most foundationally, the audience data being sold as first-party must actually be first-party. This means standardised data quality audits making the kind of systematic evaluation Georgia-Pacific conducted over years available to every brand without requiring a multi-year self-funded programme. The startups building the next generation of retail media infrastructure, Pentaleap with RTB for sponsored products, Koddi with unified commerce media, Kevel with API-first ad serving built for retail constraints rather than adapted from display advertising are creating technical conditions under which data provenance becomes auditable as a structural feature of the platform rather than a voluntary disclosure. The question is whether the incumbent networks follow. The answer will depend on whether the brands writing the cheques start requiring data quality audits the way they started requiring brand safety controls a decade ago. They started requiring brand safety controls when a YouTube adjacency scandal made not requiring them commercially embarrassing. The retail media equivalent of that moment has not yet arrived. The Instacart FTC settlement and the Cooler Screens warehouse in Texas are early drafts of it.

86 percent of commerce media decision-makers in North America and Europe identified strengthening measurement and attribution as a high or critical priority for 2026, according to the November 2025 Koddi and Forrester survey. Only 12 percent had reached an advanced state of full-funnel measurement capability. The gap between the priority and the capability is 74 percentage points. That gap is the measurement crisis expressed in the industry’s own numbers about itself, and it is the most honest data point that the $71 billion market has produced.

· · ·

Possible 2026 in Miami this month, and every industry conference before and after it, will feature a version of the retail media keynote that describes a closed-loop advertising channel with unmatched first-party data and measurable ROI. The presentation will be compelling, and some of it will be accurate. At Amazon, the loop genuinely approaches something close. At Walmart, it is being built with intention and honesty about the gaps. At the other 80-plus US retail media networks competing for a market that is barely growing for anyone outside the top two, the closed loop is a marketing claim running on measurement infrastructure that a Fortune 500 consumer goods company, after years of systematic evaluation, found did not consistently deliver what was promised.

The corridor conversations at Possible 2026, between sessions and over dinner, are more instructive than the keynotes. Brand marketers who have run holdout experiments describe a gap between platform-reported ROAS and independently measured incrementality that, once seen, cannot be unseen. Media buyers who have managed campaigns across seven or eight networks simultaneously describe the operational overhead of managing seven or eight different attribution windows, seven or eight different data definitions, and seven or eight different dashboards, none of which speaks to any other, as something that makes budget allocation feel more like administrative management than strategic investment.

The measurement crisis is not a secret. Everyone in the industry knows it exists. The 6 percent trust figure is not a surprise to the people spending the $71 billion. It is a number they recognise. The crisis persists because the current system, whatever its limitations, produces ROAS numbers that are large enough to justify the next budget request, and because the retailers providing those numbers have structured their measurement methodology to ensure that outcome. The change will come not from a technical breakthrough but from brands refusing to accept measurement that the platform producing it also defines and certifies, and from AI agents routing around the sponsored product in ways that force the industry to build data infrastructure that makes products findable by machines and outcomes attributable to the advertising that preceded them.

John Wanamaker complained in 1876 that half of his advertising was wasted and he could not tell which half. Retail media raised $71 billion on the promise that it had finally answered his question. Six percent of the people spending that $71 billion believe the answer. The other 94 percent are still waiting, not for better technology, but for measurement that belongs to them rather than to the platform selling them the inventory.

Somewhere in a warehouse outside Dallas, $50 million worth of refrigerator door screens sits in the dark, counting nothing, measuring nothing, attributing nothing. The sensors that were supposed to close the loop were counting shadows. In retail media, they still are.

THE OBLIGATORY DISCLAIMER

Written in a personal capacity. All data comes from publicly available, verified sources. No employer input. No confidential sources. Claude assisted with research and structure. The analysis, the frustrations, and the sarcasm are entirely mine. If you find a factual error, say so in the comments and I will correct it.

On Thursday, March 26, 2026, OpenAI’s communications team sent out a press release that was, depending on your point of view, either a milestone or an indictment. The company’s advertising pilot, which had launched on February 9 with a careful announcement and explicit promises of restraint, had crossed $100 million in annualized revenue in under six weeks. More than 600 advertisers and agencies had joined the program. Target, Ford, Adobe, WPP Media, Omnicom and Dentsu were among them. Criteo had become the first formal advertising technology partner, announced on March 2. David Dugan, who had spent the previous decade helping build Meta’s advertising infrastructure into one of the most profitable machines in commercial history, had been named to lead OpenAI’s global advertising solutions team earlier that same week. Self-serve advertiser tools were scheduled for April. International expansion to Canada, Australia and New Zealand was already underway.

The reaction in the publishing industry was muted in the way that the reaction to bad news often is when the bad news has been expected for so long that the actual arrival of it produces mainly silence. The OpenAI advertising business was not a surprise. The timing and scale of it were.

Six weeks. One hundred million dollars in annualized revenue. From a standing start, reaching fewer than twenty percent of eligible users daily, with a product that places ads at the bottom of responses rather than integrated into the conversation. What that number implies about the eventual ceiling, when self-serve tools open the market in April, when international expansion adds hundreds of millions of additional users, when ad load increases from its deliberately restrained pilot level, is a question the publishing industry is now collectively trying not to think about too carefully.

To understand why the publishing industry finds this particular press release so difficult to absorb, you need to understand a chain of causation that stretches back at least three years, and possibly much longer. It begins with a simple fact: the content that powers OpenAI’s models, the reporting, the analysis, the commentary, the wikis, the forums, the reference works, the decades of archived journalism and scholarship and creative writing that gave the models the vocabulary and knowledge and context to produce fluent, accurate, useful answers, was produced by publishers. Not by OpenAI. Not by any technology company. By the journalists, editors, researchers, bloggers, academics and writers who built the open internet’s information infrastructure, piece by piece, over the past three decades.

The models consumed that infrastructure. The platforms that trained on it are now running advertising businesses. The publishers who produced it are, in many cases, fighting for survival. This article is about how that situation came to exist, what it is specifically costing the industry, what a very small number of publishers are doing about it that actually works, and what the publishing economy will look like if the structural problems are not addressed in time.

The Extraction, and How It Actually Worked

In August 2025, a document leaked that the internet’s AI economy would have preferred to keep private. Drop Site News published what appeared to be an internal Meta list of websites that the company had been systematically scraping to train its AI models. The list covered approximately six million unique websites. It included publishers of every kind: national newspapers, specialist magazines, academic journals, forums, recipe sites, legal databases, fan wikis, and personal blogs. It included content published behind robots.txt restrictions that Meta’s crawlers had bypassed. The scraping was not incidental. It was systematic, ongoing, and apparently indifferent to the commercial or legal preferences of the people who had produced what was being taken.

Meta is not an outlier here. It is, if anything, the only company for whom the documentation became public. OpenAI, Google, Anthropic, Mistral and virtually every other major AI developer built their models on the same open internet, using largely the same approach. The distinction between the companies lies in their legal strategies afterward, not in whether they scraped. Some scraped and then offered licensing deals to defuse legal risk. Some scraped and insisted it constituted fair use. Some scraped, got sued, and started negotiating. None of them asked first. The data was there. The legal framework was unclear. The commercial opportunity was enormous. The decision to wait for permission was not made.

The mechanism by which AI companies converted that scraped content into a threat to publishers operates in three stages, each of which would be damaging in isolation. Together they constitute something genuinely new in the history of platform disruption.

The first stage is training. When a large language model is trained on the text of the New York Times, the Financial Times, Reuters, Wikipedia and several hundred thousand other publications, it absorbs not just factual information but editorial judgment, contextual awareness, analytical frameworks, and writing style. The model does not merely learn facts. It learns how to think about and explain facts in ways that readers find credible and useful. That learned credibility is not the model’s. It was lent to the model by the institutions whose work formed its training data.

The second stage is inference. When a user asks ChatGPT a question, the model does not retrieve an article from a publisher’s database. It generates an answer by drawing on patterns it learned during training, patterns that were derived from publisher content. The publisher who produced the underlying reporting receives no payment for this use, no attribution in the answer, and no traffic in the form of a click. The Cloudflare data from June 2025 makes the asymmetry visible with unusual precision: Google, the search engine that at least sends some users back to publishers, generates fourteen crawls for every one referral it eventually delivers. OpenAI, which trained on publishers’ content and now answers questions based on that training, generates 1,700 crawls per single referral sent back. Anthropic, whose Claude product is built on the same content economy, generates 73,000 crawls for every referral. These numbers are not fees for content licensing. They are the ratio at which AI systems extract content versus the rate at which they return traffic value to its producers.

The third stage is advertising. This is the stage that the March 26 announcement represents. Having trained on publisher content, having built a user base of ~900 million weekly active users as of February 2026, OpenAI is now running advertisements against the audience that gravitates to its product partly because that product absorbed the knowledge and credibility of the institutions that produced the content it was trained on. The advertisers buying those ads are, in many cases, the same advertisers who used to buy from publishers. The budget that flows to OpenAI’s ad inventory is budget that does not flow to a publisher’s ad inventory. The loop is closed.

Kint’s data from the nineteen publishers in the DCN network, spanning national newsrooms and global entertainment brands, showed median Google Search referral traffic down almost every week across eight weeks in May and June 2025, with losses outpacing gains two to one. The worst single weeks were particularly stark: news publishers plunged sixteen percent in the week of May 25th. Non-news publishers fell seventeen percent in the week of June 22nd. These are not minor fluctuations or seasonal patterns. They are sustained, structural losses from content categories that Google’s AI specifically summarises at the top of the search results page.

What makes Kint’s framing more penetrating than the typical publisher complaint is that he identifies the mechanism precisely. Google is not simply building a better product that happens to reduce traffic. It is taking publisher content, using it to build an AI answer system, placing that system at the top of the search results it controls through a monopoly position that Judge Amit Mehta is currently adjudicating in the US Department of Justice antitrust case, and thereby channelling the advertising revenue that publisher traffic would have generated into Google’s own products. The content is still producing value. It is just that Google is capturing that value rather than the publisher.

What the Numbers Actually Show

The problem with how the publishing industry discusses its AI-related decline is that it frames everything as a traffic problem. Traffic is a symptom. The disease runs deeper, and it is attacking publisher revenue from multiple directions at the same time, each through a different mechanism, each at a different speed, and each compounding the others in ways that make the aggregate impact considerably worse than any individual measure suggests.

Start with the traffic. Google search traffic to publishers declined globally by thirty-three percent in the twelve months to November 2025, according to Chartbeat data covering 2,500 publisher websites. The US figure was thirty-eight percent. Google Discover, which feeds mobile users through Android and Google’s own apps and which publishers had been building supplementary strategies around, fell twenty-one percent over the same period. Pew Research Center’s study of sixty-eight thousand real user search queries, published in July 2025, found that users clicked on results eight percent of the time when AI Overviews were present, compared to fifteen percent without them. That is a forty-seven percent relative reduction in click-through rate from the same ranking position, simply because Google placed its AI summary above the publisher’s result. Advanced Web Ranking research found that AI Overviews average around 169 words and include approximately seven links when expanded. Once expanded, the first organic result often appears around 1,674 pixels down the page. The publisher is still technically present in the search results. They are just considerably less visible, considerably less likely to receive a click, and considerably less able to do anything about it.

The financial arithmetic that connects traffic decline to revenue collapse is where most external commentary misses the severity of what is happening. A twenty-five percent drop in traffic does not produce a twenty-five percent drop in programmatic revenue. It produces a forty to fifty percent drop, because publisher cost structures are largely fixed while revenue is variable. AdMonsters’s analysis from October 2025 made this precise: the infrastructure, the editorial staff, the technology stack, the licensing fees, the operational overhead, all of these continue at approximately the same level whether the site receives ten million monthly visitors or seven million. When advertising revenue falls faster than costs, the math becomes existential at a speed that quarterly earnings reports can disguise for only so long.

The BuzzFeed number in that table deserves a moment of separate attention. BuzzFeed, which once represented the future of digital media with sufficient conviction that it filed for a billion-dollar IPO, disclosed in its March 12, 2026 earnings release that it lost fifty-seven million dollars in revenue in 2025, that it does not currently have sufficient assets to fund its 2026 cash obligations, and that executives have substantial doubts about the company’s ability to continue operating as a going concern. This is a company that employed Pulitzer Prize-winning journalists, that genuinely pioneered the study of how content spreads on the internet, that at various points in its history was studied as a model for what a modern media company could be. It did not fail because it produced bad journalism. It failed because the business model underpinning its journalism depended on a discovery and distribution layer that has been systematically redirected by platforms that used its content to train the systems that replaced it.

BuzzFeed is the most visible casualty but not the only one. The Washington Post cut approximately one hundred staff in January 2026, with seventy-three positions in the advertising department alone. Dotdash Meredith, now renamed People Inc., cut 226 positions in October 2025 following 143 layoffs in January. HuffPost eliminated its editor-in-chief position. Travel blog The Planet D, founded in 2008, lost half its traffic when Google’s AI Overviews launched in May 2024, laid off staff in response, then lost ninety percent more traffic following those layoffs, and ceased publication entirely. Music blog Stereogum lost seventy percent of its advertising revenue in 2025, its founder Scott Lapatine attributing most of it to AI Overviews and the rest to Facebook and X deprioritising external links. Press Gazette tracked more than three thousand journalism job cuts across the UK and United States during 2025, a figure that does not include the smaller publications that closed entirely without filing public notices.

The CPM numbers in the table require equal attention, because they are the mechanism connecting traffic decline to financial distress in publishers that have not yet collapsed. Display CPMs fell thirty-three percent year-on-year in January 2025. Video CPMs fell thirty-nine percent. These are not modest corrections. They are the market pricing the fact that there are fewer verified audiences to reach, that cookie deprecation is degrading targeting precision, and that programmatic buyers increasingly prefer the authenticated, identifiable audiences in walled gardens and private marketplace deals over the anonymous, algorithmically-assembled audiences in the open programmatic exchange. A publisher who lost thirty percent of their traffic and also saw their CPMs fall thirty-three percent did not suffer a combined thirty percent revenue reduction. They suffered something approaching a fifty to sixty percent revenue reduction, before accounting for the fixed costs that did not fall with either number.

The Deals, the Lawsuits, and the Distance Between Them

Publishers have pursued two distinct strategies in response to the situation described above, and the tension between them tells you more about how the industry sees itself than any amount of conference panel discussion. Some have negotiated. Some have litigated. Most have done both, sequentially or simultaneously, with varying degrees of enthusiasm. Neither approach has solved the underlying structural problem, and both have produced outcomes whose adequacy is a matter of legitimate disagreement.

The licensing deals represent the clearest acknowledgment that publisher content has quantifiable commercial value in the AI economy. News Corp’s five-year agreement with OpenAI, reportedly worth more than $250 million according to the Wall Street Journal’s reporting from May 2024, covers the Wall Street Journal, Barron’s, MarketWatch, the New York Post, The Times and The Sunday Times of London, and a range of Australian titles. Robert Thomson, News Corp’s CEO, described it as historic and said it acknowledged that there is a premium for premium journalism. The Financial Times struck a deal worth between five and ten million dollars annually. Dotdash Meredith received at least sixteen million dollars, confirmed by IAC’s chief operating officer Chris Halpin on a Q3 2024 earnings call, who said the lion’s share of a $4.1 million year-on-year licensing revenue increase was driven by the OpenAI deal. Axel Springer agreed to a deal worth twenty-five million dollars plus variable fees. John Wiley and Sons received a one-time twenty-three million dollar payment for its academic book archive, confirmed by interim CEO Matthew Kissner on an earnings call. Curiosity Stream projected $19.6 million in AI licensing revenue from its 210,000-hour factual video library for 2025, revenue that contributed to the company’s first profitable quarter.

These deals are real money. They are also strikingly modest when measured against what the companies paying for them are spending to build the infrastructure that requires them. OpenAI committed to infrastructure partnerships with Microsoft and Amazon that total nearly three hundred billion dollars. The entire market for AI content licensing, across all publishers, all deals, all platforms, represents a fraction of a percent of what the AI companies are investing in the technology that consumes publisher content. The publishers are the essential raw material supplier for an industry that is spending hundreds of billions of dollars on the factory, and they are negotiating with considerably less leverage than that description would suggest they should have.

The Axios deal is the one that points most clearly toward what a well-structured licensing arrangement could look like. Axios signed a three-year agreement with OpenAI in 2025 under which ChatGPT answers user questions using Axios content, with clear attribution and links back to Axios’s site. What made the deal structurally different from a simple content license was that OpenAI also provided funding for Axios to open four new local newsrooms, in Pittsburgh, Kansas City, Boulder and Huntsville. The deal was not just a revenue payment for past content. It was an investment in future content production, structured around the recognition that OpenAI’s models need fresh, authoritative, locally-reported material to remain genuinely useful, and that this material requires funded journalism organizations to produce it. This is not philanthropy. It is supply chain management.

The litigants have pursued a different calculation. The New York Times filed suit against Microsoft and OpenAI in December 2023, alleging copyright infringement in the training of the models. The suit remains active. Penske Media Corporation, the owner of Rolling Stone, Variety, Deadline and The Hollywood Reporter, sued Google in September 2025, specifically arguing that AI Overviews had reduced affiliate revenue from shopping links by more than a third compared to the end of 2024, and that the summaries appear on approximately fifteen to twenty percent of search results related to PMC’s content. This suit is important because it tests a legal theory that the licensing deals cannot: that the harm occurs not in training but in inference, in the moment when Google’s AI system generates a summary that substitutes for a click that would otherwise have reached the publisher. Chegg filed an antitrust complaint in February 2025 after its non-subscriber traffic fell forty-nine percent in a single year, alleging that Google used Chegg’s educational content to train systems that now compete directly with Chegg’s core product. The argument is not just about compensation for use. It is about the use of publisher content to build a competing service.

The distance between the deals and the lawsuits is the distance between two readings of the same situation. In one reading, the AI companies are building new products that benefit users and happen to reduce publisher traffic as a side effect, and the correct response is to negotiate fair compensation for the content that enables those products. In the other reading, the AI companies are using publisher content to build products that directly substitute for what publishers provide, redistributing publisher revenue to themselves, and the correct response is to treat this as the anticompetitive conduct it arguably is. Both readings can be simultaneously true. The industry is living with that ambiguity while the revenue declines continue at a pace that is not waiting for the courts to resolve it.

The Publishers Who Are Not Dying, and the Specific Decisions That Explain Why

The publishing industry’s failures have been better documented than its successes, partly because failure is more dramatic and partly because the publishers who are navigating this period well tend not to discuss their strategies in public. There is, however, a clear and consistent pattern across the case studies that are available, and the pattern does not support the conclusion that AI disruption is uniformly destroying the industry. It supports the conclusion that publishers who built direct, authenticated relationships with their audiences are surviving, and publishers who built their economics on anonymous, algorithmically-discovered traffic are not.

The Wall Street Journal sells ninety percent of its inventory direct, against the authenticated first-party data of its paid subscriber base. Advertisers who use WSJ’s first-party data in their targeting experience a thirty-seven percent higher renewal rate than those who do not, according to a Google Ad Manager case study. That renewal rate premium is the commercial expression of something the advertising industry has always known in principle but rarely quantified: reaching the right audience in the right context produces results that justify the premium, and the publisher who can verify the audience has the leverage to price accordingly. BCG’s global digital marketing maturity survey confirms the structural advantage: first-party data delivers twice the advertising revenue per impression of non-first-party inventory. Newsweek implemented a first-party data strategy that produced average eCPM increases of fifty-two percent, with peaks reaching two hundred and twenty-four percent. These are not marginal improvements. They are the outcome of a single foundational decision: treat the subscriber relationship as a data asset, not just a revenue line.

That quote, from the person responsible for revenue at the Wall Street Journal’s parent company, is the most compressed description available of what actually works. Logged-in consumption. Knowledge of the impression. Precision targeting that produces outcomes advertisers pay a premium for and renew because they can verify the outcomes. Everything in that sentence depends on one prior decision: requiring the audience to identify itself in exchange for access to the content. The publisher who has not made that decision is not merely leaving a premium on the table. They are operating without the foundational data asset that makes premium advertising possible.

Apartment Therapy Media, the parent company of the lifestyle brand Kitchn, provides a useful case study of what the transition looks like in practice. The company’s president Riva Syrop told Digiday that after traffic declines in the first half of 2025, the team spent the following three months intensively building newsletter offerings and improving its membership and retention programs. The specific goal was to increase the quantity of monetizable known visitors relative to anonymous ones. The Kitchn brand was up fifteen to twenty-five percent year-on-year every month in the period that followed. The mechanism Syrop described is worth quoting: the purpose of getting people as members is first-party data. Enabling them to have a personal experience. And then bringing that data back to the advertisers. This is not a sophisticated technological intervention. It is a commercial decision to invest in a direct audience relationship, and it is producing measurable returns in a period when publishers whose equivalent strategy is “wait for Google traffic to come back” are watching their numbers continue to deteriorate.

The WAN-IFRA sixth AI report, based on a survey of more than one hundred media leaders conducted in the second quarter of 2025, found that seventy-five percent of publishers report efficiency improvements from AI in their operations, sixty-four percent report better content production, and fifty-five percent report faster publishing times. Only nine percent can point to direct revenue gains from AI at this stage. That gap, between operational savings and commercial improvement, is the space where the decisions being made right now will determine which publishers are viable in 2028. The operational savings are real and immediate. AsiaOne in Singapore deployed an AI classification model for advertising targeting and saw revenue rise twenty percent while cutting sales staff forty percent, freeing the remaining team to focus on higher-value commercial activity. The South China Morning Post saves more than three hundred hours monthly through AI-assisted summarization, translation and editorial workflow tools. Legit.ng in Nigeria used AI to halve translation times, expanding Hausa-language coverage without proportional staff increases. Oliver Wyman’s 2025 analysis projects production cost reductions of twenty to thirty percent and labour cost savings of up to forty percent for publishers that implement AI at the operational level without using it to replace the editorial judgment that makes content worth producing.

The publishers using those savings to hire more reporters, assign more original investigations, and produce the kind of journalism that AI cannot summarize because AI cannot conduct it are the ones building the content advantage that will matter in a world where commodity content is free and distinguished content commands the premium. The publishers using those savings to produce more commodity content faster are accelerating toward a market where they will be competing against systems that produce equivalent output at literally zero marginal cost. This is a strategic decision, not a technical one, and it needs to be made consciously rather than by default.

The SSP’s Last Chance

The supply-side platform was a genuine innovation in 2008. Publishers with unsold inventory and no way to efficiently access multiple buyers simultaneously needed exactly what the SSP provided: aggregated demand, real-time auction mechanics, and the technological infrastructure to monetize remnant inventory that was otherwise worthless. That problem was real, and the companies that solved it created genuine value and were appropriately rewarded for doing so. The problem is that fifteen years later, the SSP market is largely still selling the same solution to a problem that has been transformed beyond recognition.

The Jounce Media 2025 State of the Open Internet report establishes the operational reality with unusual clarity. Publishers currently average 24.5 SSP partner relationships. Those 24.5 relationships generate bid requests of which nine out of ten go unprocessed because the system is congested with duplicate signals. Rebroadcasting, the practice of routing the same impression through multiple supply paths to generate the appearance of additional scale, accounts for thirty-seven percent of all display auctions and thirty-two percent of all video auctions. The publisher maintaining 24.5 SSP relationships is paying the operational, latency and complexity costs of those relationships in exchange for a bidstream that is mostly noise. The model that was designed to maximize demand for publisher inventory has, through proliferation and duplication, created a system where the majority of that demand is theoretical rather than real.

In August 2025, The Trade Desk made a change to its Kokai buying platform that received considerably less attention than it warranted. The company reclassified supply-side platforms from direct supply chain participants to resellers. Under Kokai’s scoring and prioritization system, SSP bid requests now receive lower priority than The Trade Desk’s own OpenPath direct connections. The implication, which Jeff Green had forecast in an earnings call months earlier, is that publishers who connect directly to The Trade Desk through OpenPath are delivering their inventory through a higher-priority path than publishers who route through even the largest SSP. Green said in that call that he expected the biggest content owners to integrate directly through OpenPath and essentially be their own SSP, particularly in connected television and audio. OpenPath’s volume grew what Green described as “many hundreds of percentage points” in 2025. OpenAds launched in January 2026 with nine publishers including the Guardian, Hearst Magazines, Newsweek and BuzzFeed, creating a transparent auction environment where publishers see bid-level data that has historically been visible only to the SSP.

The Ventura Ecosystem, announced in February 2026, integrates OpenPath, UID2 and EUID, OpenAds and OpenPass into a unified architecture that allows premium publishers to connect authenticated audience data directly to advertiser demand without SSP intermediation. OpenTTD, launched in March 2026 as a unified developer portal, is the infrastructure layer that makes this architecture accessible to publishers who want to build on it. What The Trade Desk is assembling, piece by piece and with considerable tactical patience, is a direct-to-demand infrastructure for publishers who have authenticated audiences and the commercial sophistication to operate within it.

The SSPs that are growing in this environment share a common characteristic: they have moved from being inventory aggregators to being audience intelligence platforms. PubMatic launched its AI-powered monetization platform in September 2025, integrating natural language deal setup through PubMatic Assistant, which allows publishers to configure PMP and programmatic guaranteed deals by describing them in ordinary language rather than navigating a complex interface. PubMatic Connect helps publishers package and activate their first-party audience data for buyers at the deal level. CEO Rajeev Goel’s statement at launch, that the myth of the passive publisher is over, describes the direction correctly: the SSP’s value proposition for the next decade is helping publishers communicate the specific, differentiated value of their authenticated audiences to buyers who will pay a premium for them, not aggregating anonymous inventory through an automated auction.

Magnite’s April 2025 unification of the SpringServe ad server with its programmatic SSP capabilities, deployed initially with Disney Advertising, Paramount, Roku, Samsung and Warner Bros. Discovery as clients, represents the other viable architectural direction: combining ad serving and yield management in a single system so that the publisher controls every dimension of their monetization strategy in one place. The Jounce Media March 2025 Supply Path Benchmarking Report verified that the unified platform connects buyers to ninety-nine percent of US streaming supply on a dollar-weighted basis. A publisher using this system knows what their inventory is worth, controls how it is accessed, and does not route their audience intelligence through a third party that takes a fee in exchange for providing less information than the publisher could access directly.

Onetag, the European exchange and curation platform, took a different approach when it acquired Aryel in March 2026. Aryel specializes in immersive and interactive advertising formats, including augmented reality and rich interactive experiences, that generate creative performance data alongside placement decisions. By integrating Aryel’s creative technology with Onetag’s global exchange connecting more than 2,000 publishers and 5,000 advertisers, the combined entity is building something that neither a standard SSP nor a standard creative platform can offer: a system where the quality of the creative experience, the quality of the editorial environment, and the AI-driven optimization of campaign outcomes are integrated at the infrastructure level. This matters because it addresses the question that the SSP market has never adequately answered, which is: beyond demand aggregation, what specific and measurable value does the intermediary add to the transaction? Onetag’s answer is: better outcomes, verified by creative performance data and audience quality signals, optimized in real time by AI decisioning that has access to all three elements simultaneously.

The publisher choosing SSP partners in 2026 is making a decision with long-term structural consequences. The partners who are building intelligence layers around authenticated publisher audiences, who are investing in tools that help publishers communicate audience value rather than aggregate anonymous inventory, who can demonstrate in log-level data that their participation improves yield rather than merely adding a fee to a transaction that would occur anyway, those are the relationships worth maintaining. The partners who cannot answer the question of what specific value they add to a publisher’s authenticated first-party audience inventory are the ones that The Trade Desk’s Kokai algorithm is already beginning to route around, and that AI-powered supply path optimization will route around with increasing efficiency over the next two years.

The Infrastructure Being Built Around the Problem

While the publishing industry debates its response to AI, a small number of companies has been quietly building the payment infrastructure for a world that the existing advertising economy has no mechanism to address: a world where AI systems are significant consumers of publisher content, but consume it in ways that generate no page views, no sessions, no impressions, and no advertising revenue. The problem these companies are solving is new in kind, not just in scale. The advertiser-publisher-audience triangle that has underpinned digital advertising for twenty years has a structural blind spot: it has no mechanism to capture value when the “reader” is not a human being.

TollBit, co-founded by Toshit Panigrahi and Olivia Joslin, has raised thirty-one million dollars and serves more than three thousand publisher clients. The company starts from a fact that most publishers have not fully processed: they are visited by AI crawlers at a scale that dwarfs their human readership, by systems that extract content value without generating any of the commercial signals on which advertising revenue depends. According to Cloudflare’s June 2025 data, Anthropic’s crawlers generated 73,000 crawls per single referral sent back to publishers, OpenAI’s crawlers generated 1,700 crawls per referral, and Google’s crawlers generated fourteen per referral. The publisher with a million monthly human visitors and typical crawler ratios is receiving something in the range of fifty million AI crawler visits monthly. Not one of those visits generates an impression, a session, or a commercial signal. TollBit’s bot paywall creates a payment infrastructure for exactly this category of non-human visitor, allowing publishers to set differentiated rates by purpose: the price for AI training access is different from real-time inference access, which is different from search enrichment access, because the commercial value generated by each use case is different.

TollBit’s recent addition of per-query pricing, using Microsoft’s NLWeb protocol, addresses a point that IAB Tech Lab CEO Anthony Katsur made explicitly: one crawl can feed ten thousand queries, fifty thousand queries, and the publisher has been paid only once for the crawl. Per-query pricing converts the relationship between a publisher’s content and an AI system’s output from a one-time transaction to an ongoing royalty structure. It is, in commercial principle, the same model that music rights holders use when they licence recordings for streaming: not a one-time payment for a file, but a per-play fee that accumulates over time relative to actual usage.

ProRata took a different route to the same destination. The company, which raised forty million dollars in Series B funding in September 2025 and has more than five hundred publisher partners, built Gist.ai, an AI search engine that uses only licensed publisher content and distributes fifty percent of all revenue generated by that search engine to publishers based on how often their content powers the responses. The Atlantic, Time, Fortune, ADWEEK, BuzzFeed and Lee Enterprises are among the participants. The model’s structural advantage over Cloudflare’s and TollBit’s bot-paywall approaches is that it does not require AI companies to opt into payment before the model generates revenue. ProRata creates a separate commercial ecosystem where licensed content is the only content, advertising runs against the answers that licensed content powers, and revenue flows monthly to the publishers whose content was cited. It is a parallel economy rather than a reformed version of the existing one.

Cloudflare’s Pay Per Crawl marketplace, launched on July 1, 2025, operates at a different scale than either TollBit or ProRata because Cloudflare sits in front of approximately sixteen percent of all global internet traffic. The company blocks AI crawlers by default for new websites and has created a market where publishers can set micropayment rates per page access. The publisher participants at launch read like a partial inventory of what the open web’s premium content looks like: Conde Nast, Dotdash Meredith, the Associated Press, The Atlantic, TIME, Fortune, BuzzFeed, Gannett, Reddit and Universal Music Group. Conde Nast CEO Roger Lynch: when AI companies can no longer take anything they want for free, it opens the door to sustainable innovation built on permission and partnership. Neil Vogel, CEO of Dotdash Meredith: we can now limit access to our content to those AI partners willing to engage in fair arrangements. These statements describe a commercial reality that is only beginning to materialize. The Cloudflare marketplace does not yet represent the majority of publisher-to-AI company transactions. It represents the architecture for what those transactions could become once the legal and commercial frameworks catch up with the technical infrastructure.

Microsoft’s pay-per-use AI content marketplace, which signed both the Associated Press and USA Today in 2025, is the first attempt by a major AI company itself to build systematic market infrastructure rather than negotiating individual bilateral deals. The significance of this is not merely commercial. It is a signal that at least one major AI platform has concluded that the sustainable model for content access involves a functioning market rather than a series of separately negotiated agreements at different prices, with different terms, signed by publishers who have no way to benchmark what anyone else received. A market is better for publishers than bilateral negotiations, because a market creates pricing transparency and allows any publisher to understand the fair value of their content relative to comparable titles. Markets are generally better for buyers too, because markets are more efficient than individual negotiations. The fact that Microsoft is building one suggests that the current state of bilateral dealing is recognized as unsatisfactory even by the companies that have benefited from it.

The Dependency Nobody Wants to Discuss

There is a structural dependency embedded in the AI economy that the AI companies prefer not to emphasize and the publishing industry has been slow to recognize as leverage. The large language models that power every major AI product on the market were trained on human-produced content. Not on synthetic data. Not on AI-generated text. On the journalism, analysis, fiction, scholarship, instruction, commentary and conversation that human beings produced and published over several decades. The models learned to write well because they read writing that was good. They learned to reason about complex topics because they absorbed reasoning about complex topics. They learned which facts are reliable because they trained on sources with editorial standards and accountability structures. When the training data degrades, which it will if the institutions producing it can no longer afford to operate, so do the models.

This dependency is already visible in the way AI companies are competing for content. News Corp’s $250 million deal with OpenAI was not the price of a favour. It was the market rate for an archive that provides exactly the kind of verified, sourced, dated, contextually-rich content that makes models more accurate and trustworthy. Google’s first AI content deal, struck with the Associated Press in 2025, gives Gemini access to AP’s real-time news production. Washington Post joined OpenAI in April 2025. The New York Times struck a deal with Amazon for Alexa and Amazon’s AI products. These companies are not paying for content because Sam Altman woke up one morning with a conscience about journalism. They are paying because the models need it and the alternative is litigation that could set precedents substantially more expensive than the licensing fees.

Axios understood this dependency more clearly than most when it negotiated not just payment but investment: OpenAI funding four new local newsrooms in exchange for content access is a recognition that the model’s value depends on the continued production of authoritative local reporting, and that this reporting requires funded organizations to produce it. The deal creates a direct commercial incentive for OpenAI to ensure that Axios has the resources to keep generating the content the model needs. This is a structural arrangement, not a charitable gesture, and it is the model that the rest of the licensing market should be working toward.

The publishers who have recognized this dependency as leverage are the ones building the most durable commercial positions. The Financial Times’s Storyfinding team uses AI to surface patterns in large datasets, which human reporters then investigate. The model’s ability to spot patterns in financial data only creates value if a reporter with decades of markets expertise can assess which patterns are significant and pursue the story. That reporting is what makes the FT worth licensing, worth subscribing to, and worth citing in AI answers. The AI improves the reporter’s efficiency. The reporter validates and extends what the AI found. The content that results is both more efficiently produced and more exclusively valuable than it would have been without either component. This is the correct division of labour, and it produces content that AI platforms will continue to need and pay for precisely because it cannot be replicated without the human component.

The publisher who has not made this calculation is still producing content in the old way, at the old cost, for an audience that the AI has redirected. The window in which they can use the operational savings that AI tools make possible to reinvest in the irreplaceable reporting that makes their content worth licensing is not infinite. The models are already training on a substantial volume of AI-generated text, and research suggests this degrades model quality over successive generations of training. The premium that authentic, human-reported, editorially-verified content commands in the AI training market will increase as synthetic content proliferates and the scarcity of genuine human expertise becomes more visible. The publishers who are still producing it in 2028 will be in a stronger negotiating position than those who have already cut the editorial functions that make their content distinctive.

The argument that publishers have genuine leverage in this situation is not naive optimism. It is grounded in the specific dependency the AI companies cannot route around. They need original, verified, recently-produced, contextually-rich human content to remain credible and useful. Without it, the models degrade, hallucinate more frequently, and lose the currency that makes them commercially valuable. Every major AI company knows this, which is why they are paying for licensing deals, funding newsrooms, and building compensation infrastructure rather than simply continuing to scrape freely and defend the practice in court. The leverage exists. The question is whether publishers are willing to price it like the asset it is, rather than continuing to treat access to their content as a default that any technology company is entitled to use.

What Needs to Happen, in Order

The publishing industry is at the point in a structural disruption where the companies that adapt early enough to rebuild their foundations before the revenue collapse reaches critical mass will survive, and those that delay will not. This is not a new observation about disrupted industries. It is, however, an observation with a specific urgency in this case because the pace of the structural change, as measured by the traffic and CPM numbers, is faster than the timeline on which most publishing companies make strategic decisions.

The first decision is architectural and cannot be deferred: publishers need to know who their audience is. Not the monthly uniques number in the analytics dashboard, which measures anonymous visits that have no commercial value in the world being built around them. The number of verified email addresses from people who have actively provided them. The count of logged-in sessions in the past thirty days. The active subscriber total. A publisher who cannot answer these questions with precision does not have an audience in the commercially meaningful sense. They have traffic, and traffic is what they are losing.

The second decision is about content investment. AI tools make it genuinely possible to produce the operational work of a newsroom, the formatting, the metadata, the headline testing, the distribution scheduling, the translation, the summarization, at significantly lower cost than before. The question every editorial leadership team needs to answer is whether those savings are being reinvested in the work that AI cannot do: the investigations that require a reporter in the room, the analysis that requires expertise built over years, the accountability journalism that requires the institutional weight to withstand legal and political pressure. If the answer is no, then the publication is using AI to accelerate its own commoditization.

The third decision is about the supply chain. A publisher running 24.5 SSP relationships while nine in ten bid requests go unprocessed is not operating a monetization strategy. They are operating the simulation of one, at the cost of latency, complexity and the fee that each intermediary extracts from each transaction that does complete. The publishers ahead of this curve are consolidating their SSP relationships to partners who can demonstrate, in log-level data and verifiable yield analysis, that their participation adds margin to the publisher’s inventory rather than extracting it. They are exploring direct connections through OpenPath and OpenAds for their premium authenticated inventory, where the buyer’s demand can reach the publisher’s audience without an intermediary taking a percentage of a transaction that the intermediary did not originate.

The fourth decision is about AI access pricing. Publishers who have not established a commercial position with AI crawlers, through Cloudflare’s marketplace, TollBit’s bot paywall, ProRata’s licensing pool, or bilateral negotiations with AI companies, have implicitly priced their content at zero for AI access. Zero is not a business decision. It is the absence of one. The price of content in the AI training market is in the process of being established right now, through the deals being signed and the court cases being argued. Publishers who participate in establishing that market price, even at modest initial levels, are building commercial relationships and legal frameworks that will matter considerably more once the regulatory environment catches up with the commercial one.

The fifth decision is one of framing, and it is the one the industry finds hardest to make: publishers need to stop thinking of their problem as a traffic problem and start thinking of it as an asset pricing problem. The content is an asset. The authenticated subscriber relationship is a more valuable asset. The editorial authority that makes a publication worth citing in an AI answer is an asset. The twenty-year archive of verified, dated, sourced, human-produced reporting is an asset that AI companies have demonstrated they will pay for when the alternative is litigation. None of these assets is worth its full potential value if the publisher has not built the commercial infrastructure to price and sell them. Building that infrastructure is the work of the next eighteen months, and for the publishers who do not start it, the next eighteen months may be the last period in which starting is an option.

The content that made the open internet’s information economy possible was produced by people who took professional and financial risks to report things that were difficult to report, to explain things that were difficult to explain, to hold accountable people and institutions that would have preferred not to be held accountable. The AI companies consumed that content to build products that are now generating hundreds of billions of dollars in enterprise value. The deal they are currently offering to publishers, modest per-article licensing fees and the promise of citation-with-attribution, is the deal that the only party with leverage in the negotiation should be able to improve. The leverage is real. The dependency is structural. The window is closing, but it has not closed yet.

Nobody, and I genuinely mean nobody on the planet, woke up this morning hoping to see an advertisement. Not one person set their alarm earlier than necessary to catch a pre-roll video. No one booked bandwidth for a programmatic display unit. The person reading the New York Times did not pay thirty dollars a month for a digital subscription because they wanted to be followed around the internet for six weeks by a mattress brand.

Advertising is a tax. Specifically, it is a tax that consumers pay in the currency of attention and time, in exchange for content they actually want. That is the deal. It is a reasonable deal. Premium journalism, streaming entertainment, newsletters, podcasts, sports coverage, specialist blogs across 190 countries, users get all of this either free or below production cost because advertisers pick up part of the bill in exchange for showing their products to the people consuming that content. The user tolerates the ad. The publisher gets paid. The advertiser reaches an audience. Everybody goes home, if not exactly thrilled, then at least functional.

Before we go any further, stop. Stop and ask yourself a genuinely simple question: what is adtech, really? Not what the industry calls it. Not what the conference circuit describes it as. What is it, from first principles? Strip away the acronyms. Strip away the trade press vocabulary. Strip away fifteen years of white papers and working groups. What you are left with is this: supply and demand. Publishers have content and audiences. Advertisers want to reach those audiences. That is the whole thing. Everything that exists between those two parties is, in plain old-fashioned business language, a middleman. That is not an insult. Middlemen exist in every industry, every market, every supply chain in human history. They can add genuine value — aggregating supply, reducing friction, creating liquidity. The question has never been whether middlemen exist. The question is whether they earn their position in the chain, or simply occupy it. That question is what this article is about. And it is a question the adtech industry has been remarkably reluctant to answer from the ground up.

The problem is not the deal between the publisher and advertiser. The problem is the supply chain executing the deal. Between the advertiser who hands over the money and the publisher who is supposed to receive it, there are approximately forty different companies, each with their own technology stack, their own fee structure, their own auction mechanics, and their own sincere belief that their layer of this arrangement is entirely indispensable. By the time the money completes its journey, the publisher receives about forty-four cents of every dollar the advertiser spent. In 2020, the ISBA and PwC tracked one hundred million pounds in UK advertising spend and found fifteen percent of it attributed to nobody at all. Not to fraud. Not to a specific fee. Just to a gap in the universe.

The user is watching a show on Paramount Plus. The advertiser is spending a hundred thousand dollars to reach them. The publisher is receiving forty-four cents of that dollar. The other fifty-six cents are funding the most elaborate toll road in the history of commerce. Nobody commissioned the toll road. Everybody is using it. Everybody in it is getting paid. Including, as it turns out, the people nobody specifically hired to be there.

This article is about how that toll road was built, how AI is about to install cameras on every stretch of it, why publishers outside the walled gardens have more power than they are exercising, and why the open internet, the part that has been writing transparency pledges since 2016 while running practices that made transparency commercially inconvenient, is heading toward a reckoning it cannot write its way out of.

But first, we need to go to India at five in the morning. And then to a trading floor in New York in 2007. Both are relevant. Both are the same story.

01

The Farmer, the Bond Trader, and the Programmatic Supply Chain

A farmer in India loads his harvest onto a truck. Five months of work: seeds, fertilizer, water, labour, and the specific anxiety of someone whose entire year depends on the weather making two correct decisions in a row. He knows his production cost to the rupee. What he cannot know, because the market structure prevents him from knowing, is the price his grain will fetch at the regulated marketplace down the road. That information belongs to the licensed traders who operate there. They know the prices. They know each other. And they are not in the habit of sharing either.

Academic researchers studying agricultural supply chains in India found that for every one dollar a consumer pays more for food, the farmer receives two extra cents. Two. Out of a hundred. The Reserve Bank of India confirmed in its 2024 research bulletin that farmer income as a share of consumer price ranges from 28 percent to 78 percent depending on the crop. The farmers at the 28 percent end have learned to live with this mathematics. They do not have much choice. The information that determines the price sits with someone else.

Now, picture a trading floor in New York in 2007. American banks are issuing mortgages to people who will not be able to repay them. These mortgages are rated BBB. Then the banks package bundles of BBB mortgages together, hire ratings agencies to certify the resulting product as AAA, and sell it to pension funds as a safe instrument. The pension fund cannot see inside the package. The ratings agency certifying the quality is being paid by the bank that built the package.

“It’s a big pile of bonds bundled together. The genius part is that the banks thought they could package them together and spread the risk. But actually they were just piling up the risk.”

-- Ryan Gosling as Jared Vennett, The Big Short, 2015

Michael Lewis documented what happened next in his 2010 book The Big Short. Adam McKay made it into a film five years later. The key scene is a Jenga tower. Each block is a mortgage. Stack enough uncertain blocks the right way, apply the right label, and what was individually BBB becomes collectively AAA. The ratings agency says so. The pension fund buys it. The bank collects its fee.

Now, here is the programmatic advertising supply chain.

An agency trading desk buys digital advertising inventory in bulk. It buys some premium inventory: genuine quality publishers, real engaged audiences, brand-safe environments. It also buys made-for-advertising inventory, which is the polite industry term for automatically generated websites that exist solely to host ads, with traffic assembled to satisfy a targeting algorithm rather than to actually read anything. This mixed inventory gets packaged together as ‘proprietary media’ and sold to clients at a uniform premium price. The client cannot see inside the package any more than the pension fund could see inside the CDO. The entity certifying the quality of the bundle is the same entity selling it.

The ANA found in its 2024 benchmark that 6.2 percent of US programmatic spend, roughly $3.2 billion at $52 billion total market size, still landed on made-for-advertising sites, with some marketers directing more than 25 percent of their budgets there. The BBB inventory, dressed as AAA, sold at the AAA price. The publisher who created genuine premium content receives forty-four cents of the advertiser’s dollar. The farmer gets two cents.

The farmer in India, the 2007 pension fund manager, and the 2026 brand manager approving a programmatic budget are all living inside the same structural problem. Someone in the middle controls the information about what the product is actually worth and what price it traded at. The intermediaries are different in every story. The mechanism is identical in all three.

What Happens When Someone Actually Measures the Gap -- ISBA/PwC 2020 vs 2022

2020: ISBA and PwC tracked approximately 100 million pounds in UK ad spend across 15 major advertisers including Tesco, Unilever, and BT. Publishers received approximately 51 percent of advertiser spend. A further 15 percent (the ‘unknown delta’) could not be attributed to any known supply chain participant. Not to fraud. Not to a stated fee. To nobody. The Guardian received 30 pence per brand pound in its own controlled test.

2022 follow-up: publishers’ share rose to 65 percent. Unknown delta fell to 3 percent. The 14 percentage point improvement in publisher share came entirely from measurement and sustained industry pressure. Nobody improved their practices out of virtue. They improved because the data made their margin visible and clients started asking questions. That is the most important data point in the transparency debate: accountability works when it is quantified. Sources: ISBA/AOP/PwC Programmatic Supply Chain Transparency Studies, 2020 and 2022.

From 51 cents to 65 cents in two years, just from making the measurement public. No regulation. No litigation. No conference panel. Just data, published, visible to the clients paying for all of it. This is also, if you think about it, a fairly devastating commentary on what fifteen years of transparency white papers without measurement achieved.

02

Nobody Cares About Open Internet vs. Walled Gardens, Except the People Whose Margins Depend on You Caring

Here is a confession the advertising technology industry finds uncomfortable: advertisers do not have a religious preference for where their ads run. They are not emotionally attached to Google or philosophically committed to the open internet. They want to reach the people they need to reach, in contexts where those people are paying attention, at a price that produces a return. That is the entire brief. Full stop.

A chief marketing officer at a consumer goods company does not lie awake agonizing over whether their campaign runs on YouTube or the Washington Post or a podcast or a Substack newsletter. They lie awake wondering whether the campaign sold the product. The CMO who tells their board they achieved ‘strong reach and frequency metrics’ while declining to report revenue contribution is not long for their job in 2026, which is a recent development in the profession and a healthy one.

The walled gardens have been winning the budget argument not because advertisers love them but because they made ROAS measurement relatively easy to report, even if the methodology is controlled entirely by the platforms measuring themselves. Meta says your ad produced these conversions. Google says this search campaign drove this revenue. Amazon says these sponsored products generated these purchases. The open internet’s equivalent for many years was: trust us, it worked, here is a third-party viewability report and an audience segment that may or may not have seen the ad. One of these is a harder sell to a CFO who is now, for the first time, actively present in the marketing budget conversation.

The users are not in walled gardens because they love Meta’s advertising infrastructure. They are on Instagram because their friends are on Instagram. They are watching Prime Video because Amazon spent $8.5 billion acquiring MGM in 2022 and has been commissioning premium content ever since. They are on YouTube because YouTube has more video content than any human could watch in several lifetimes. The content and the community are there. The advertiser follows the user. The money follows the advertiser. The walled garden wins not because it invented better advertising technology but because it controls where attention goes, and then built an advertising business on top of that control.

Here is a first-principles point the open internet conversation consistently refuses to make clearly: walled gardens are not a violation of natural order. From basic business logic, they make complete sense. A walled garden owns the technology. It owns the user base. It provides services those users actively choose. It bears the cost of that relationship — the infrastructure, the content, the moderation, the product development. In exchange, it reserves the right to monetise that relationship as it sees fit. That is not opacity. That is a vertically integrated business doing exactly what vertically integrated businesses do. The term “open internet” was not coined because it was philosophically superior. It was coined because not everyone can be a walled garden. Can Paramount? In theory, yes. In practice, it requires billions in technology investment, a global user base of sufficient scale, and the organisational appetite to bear the full cost of building and operating all of it. The open internet exists because most publishers cannot clear that bar — not because the open model is inherently more virtuous. Once you understand that from first principles, the transparency argument changes shape entirely. The question is not why walled gardens are opaque. The question is why the open internet built a supply chain so complex that even its own participants could not account for where fifteen percent of the money went.

Also here is what the walled garden debate obscures: the audiences exist everywhere. The New York Times crossed 10 million digital subscribers in February 2023. Spotify has approximately 600 million monthly active users across 180 markets. Paramount Plus, Disney Plus, Warner Bros. Discovery’s streaming services, Bloomberg, the Financial Times, The Economist, and thousands of specialist publishers across 190 countries, these are not leftover audiences that advertisers have to settle for. In many categories they are better audiences than the walled garden alternatives: more engaged, more logged in, more willing to pay for what they are consuming, which tells you something specific and useful about their disposable income and the quality of their attention.

The advertiser who needs to reach financially active adults in their thirties and forties has a stronger argument for Bloomberg or the Financial Times than for Instagram Reels adjacent to a recipe video. The advertiser trying to reach sports fans in connected television has a better case for ESPN or a league-owned streaming service than for generic YouTube pre-rolls. Advertisers know this. The problem is that the open internet’s supply chain has been too busy collecting its own toll to demonstrate the value of the inventory it sits on top of.

The Statistical Sleight of Hand: Both Numbers Are Real and They Contradict Each Other

The open internet received more advertising money in absolute dollar terms in 2024 than in 2016. Total global digital ad spend grew to approximately $650 billion in 2024. In absolute terms, the open internet is a larger business than it was. In share terms, it went from approximately 48 percent of global digital advertising in 2008 to approximately 20 to 21 percent in 2024. New walled gardens are being created constantly: TikTok built one of the fastest-growing ad platforms in history from essentially zero after 2019 and now has approximately one billion monthly active users. Retailers, from Walmart to Target to Kroger to Tesco, have turned their customer data into closed advertising ecosystems. The market is expanding and the walled garden share of that expanding market is growing faster than the open internet’s share. In the long race, the open internet is losing ground every year even as the cheque it receives gets slightly larger. Focusing on absolute growth to avoid discussing relative decline is precisely the kind of statistical framing that has allowed the industry to avoid the harder conversation for a decade.

03

How the Supply Chain Got Built, and Why It Stayed Complicated

Take a step back further. Adtech as a distinct industry started because the internet happened. With the internet came cookies — a mechanism for remembering who had visited what, which advertisers and publishers latched onto as the infrastructure for targeting and tracking. The rest of the tech world moved on from third-party cookies relatively quickly. Adtech did not. The entire programmatic ecosystem was built on, and remained dependent on, cookie-based identity for roughly thirty years. When Google announced its Privacy Sandbox project to phase out third-party cookies in Chrome, the industry went into a sustained state of anxiety. When Google eventually abandoned the project in 2024, the collective sigh of relief across the industry was audible. Think about what that tells you from first principles: an industry that had three decades to develop a durable, privacy-respecting identity infrastructure had instead spent those three decades hoping nobody would take away the crutch. That is not a technology problem. That is a structural incentive problem. Real-time bidding, when it arrived, was a genuine innovation. But if you look at it honestly, RTB in practice has never been true bidding in the classical economic sense. True bidding produces price discovery — a clearing price that reflects what the inventory is genuinely worth to the buyer who values it most. What RTB produced in practice was closer to a waterfall with extra steps: a structured sequence in which certain buyers get preferential access, floor prices are set opaquely, and the “winning bid” often reflects who had the best positioning in the queue rather than who had the highest genuine valuation. Header bidding emerged specifically to address this — to give publishers a mechanism to run a more genuinely simultaneous auction across multiple demand sources. It was the market’s own correction for the structural distortion that RTB had introduced. All of this, cookies, RTB, header bidding, the proliferation of SSPs and DSPs, happened not because someone designed a rational architecture from first principles, but because each layer solved the immediate problem created by the previous layer, while generating the conditions for the next layer to extract its own margin.

Programmatic advertising’s origin was legitimate. In 2007, publishers had unsold remnant inventory and no efficient way to reach multiple buyers simultaneously. DoubleClick was acquired by Google in 2007 for $3.1 billion because it had built infrastructure the market genuinely needed. The supply-side platform, the ad exchange, and the demand-side platform were real solutions to real market problems. The companies building them created genuine value and were correctly rewarded for it.

Then the market did what markets do when the original problem is solved and the money is still flowing. New entrants arrived. By 2014 the IAB had counted more than 1,000 advertising technology companies operating across the stack. Supply-side platforms proliferated. Supply path optimisation, which became a significant industry practice from 2019 onward, was the market’s first honest acknowledgment that the supply chain had developed a redundancy problem. When buyers started asking which of these hundred intermediary paths was actually necessary, the ISBA/PwC answer was: we cannot tell you where 15 percent of your money went, and the publisher at the end of the chain is receiving 51 cents of your pound.

Here is the core of the issue, stated from first principles: the entire adtech supply chain grew the way it did because publishers did not know how to price their own inventory and advertisers did not know their return on investment in any quantitative, verifiable sense. That is it. That is the whole explanation. When the seller does not know the value of what they are selling and the buyer cannot measure what they are getting, the middle fills with people who are very confident about the value of their own contribution. Advertisers burnt money on the advice of agencies, routed through DSPs, exchanges, SSPs, ad servers, chains of intermediaries stacked on top of each other, each extracting a fee at a point in the chain where neither the publisher upstream nor the advertiser downstream had full visibility. The result was that more than fifty cents of every programmatic dollar got absorbed before it reached the person who had created the content in the first place. The transparency debate — the white papers, the industry coalitions, the conference panels — was a downstream symptom of this upstream failure of price discovery. The Trade Desk built its entire commercial identity around transparency as a differentiator, and it worked, because nobody else was willing to make the same bet. That tells you everything about how the rest of the industry had chosen to compete.

Pritchard said that representing approximately $10 billion in annual advertising spend. The industry agreed, published white papers, and largely continued as before. Three years later, ISBA/PwC found 15 percent unattributable. The muck, it turned out, was load-bearing.

At $52 billion in US programmatic spend in 2024, with $439 per thousand reaching publishers, approximately $22.8 billion reached content creators. The other $29.2 billion went to fees, margins, MFA inventory, and a residual unknown category. In December 2024, the AdExchanger Transparency Benchmark found that 61 percent of media buyers said they did not fully trust reported auction fairness. After fifteen years of programmatic advertising, the majority of the people buying it do not trust how it works. That is not a measurement problem. That is a supply chain that has never been required to be fully honest about itself.

04

The 78 Percent Problem: The Open Internet Is Losing the Long Race

In 2008, walled gardens held approximately 52 percent of US digital advertising revenue. By 2020, per Skai’s analysis of eMarketer data, they held over 82 percent in the United States. Globally, Statista reported that walled gardens reached 78 percent of digital advertising revenue in 2022. The projected share for 2027 is 83 percent. Analyst consensus places the 2030 figure above 85 percent if the retail media expansion continues at its current pace.

Alphabet, Meta, and Amazon generated approximately $422 billion in advertising revenue in 2024, roughly 65 percent of total global digital ad spend outside China. Advertisers in the United States spent approximately $2.50 with walled gardens for every $1 on the open web. And the model is replicating. Amazon, Walmart, Target, Kroger, Tesco, JD.com, and dozens of other retailers have built closed advertising ecosystems around their own customer purchase data. TikTok reached approximately one billion monthly active users and built one of the fastest-growing ad platforms in history from essentially zero after 2019. Every new platform that achieves scale defaults to the walled garden model because the walled garden model is profitable, relatively defensible, and relatively easy to operate.

The open internet did not lose to walled gardens because it ran out of audiences, publishers, or content. It lost because a significant portion of its own intermediaries spent fifteen years competing on margin extraction rather than on demonstrable value creation. Google, Meta, and Amazon are opaque about their auction mechanics, targeting logic, and attribution methodology. The open internet’s claimed advantage over this was transparency. In practice, its own supply chain generated ‘unknown deltas’ of 15 percent, bundled made-for-advertising inventory with premium inventory, collected non-product-related income from client spend without client knowledge, and fought the audit tools that would have demonstrated its value. An industry that cannot account for 15 percent of a client’s spend does not win the transparency argument by stating it. Advertisers made a rational choice.

In Europe, the EU Digital Markets Act designated Google and Meta as gatekeepers in September 2023, imposing data sharing and interoperability obligations that created the first serious legal framework for challenging walled garden dominance. Early implementation has not changed market share but established the regulatory infrastructure that could eventually do so. In China, Alibaba, Tencent, ByteDance, and Baidu operate a parallel walled garden ecosystem entirely closed to Western programmatic infrastructure. In Southeast Asia, Grab and Shopee are building super-app media environments. The open programmatic internet as practised in the United States and Western Europe barely exists in most of Asia Pacific.

05

Why 2025 and 2026 Are Different: Three Reasons the Old Playbook Stops Working

The transparency debate is not new. The ANA and K2 Intelligence documented agency principal-media markups of 30 to 90 percent in 2016. The ISBA/PwC unknown delta was mapped in 2020. What is different now is that three forces are operating simultaneously, each making opacity more expensive to maintain than it was the year before.

The CFO Has Entered the Building

Between 2020 and 2022, cheap money and pandemic-driven digital growth made return-on-advertising-spend conversations aspirational rather than mandatory. US interest rates then rose to their highest level since 2007. Marketing budgets became the most visible discretionary cost on the P&L. CFOs who had been passive observers started requiring ROAS data rather than impressions reports. The question ‘where did the money go’ became a CFO question, and the answers that satisfied media operations teams for a decade did not satisfy CFOs.

AI Is Making the Invisible Visible

When a bidding algorithm manages campaign optimisation in real time at a fraction of the cost of a human trading desk, the question of what the trading desk was charging for becomes impossible to avoid. A 2025 Bannerflow survey found 83 percent of senior brand marketers already use AI to target digital ads. Basis Technologies found 92 percent of advertising agencies use AI in some capacity. The ANA’s 2024 benchmark found that private marketplace spend, where algorithmic curation replaced manual supply chain navigation, rose to 59 percent of programmatic, up from 41 percent in 2023. The manual work that was billed at margin is being automated. The billing structures have not changed at the same speed. That gap is where the new friction lives.

AI is also changing attribution measurement in ways that systematically disfavour opacity. Machine learning attribution models can now produce ROAS estimates with a precision that turns ‘did this campaign work’ from a narrative argument into a numbers argument. Numbers arguments are harder to win when the numbers show 43.9 cents reaching publishers and 6.2 percent going to made-for-advertising sites.

The Courts Produced What the Industry Would Not Volunteer

The US Department of Justice antitrust case against Google’s advertising technology, which went to trial in 2024, placed supply chain economics into a public court record with the specificity and authority that fifteen years of industry conferences could not. The WPP whistleblower case, which became public through WPP’s own court filings in November 2025, placed approximately $1 billion annually in what WPP’s internal documents called ‘non-product related income’ into evidence, with internal data showing 97.4 percent of the relevant clients were not using the inventory that generated that income. Neither disclosure was volunteered. Both were extracted by legal process.

Three Converging Forces: Why the Old Playbook Has a Shelf Life

CFO pressure: marketing must justify itself in revenue terms, not impressions. AI displacement: the manual work of trading desks is being automated, making the cost-for-value question impossible to ignore. Legal exposure: the DOJ antitrust trial and the WPP whistleblower filing put specific supply chain economics into public court records for the first time. Running an opaque supply chain was always expensive for the advertisers paying for it. In 2026, it is also becoming expensive for the intermediaries running it, because the information that justified the opacity is no longer available only to them.

06

Publishers Have More Power Than They Think, But Only If They Control the Technology

Here is the argument the publishing industry needs to make to itself, out loud, in 2026. The content is yours. The audience is yours. The first-party data, the logged-in user behaviour, the reading patterns, the subscription history, the content consumption signals, all of it is yours. The only thing that is not yours, and the only thing that determines what any of it is worth to an advertiser, is the technology sitting between your inventory and the advertiser’s demand. That technology, for most publishers outside the very largest, was built by someone else to serve someone else’s interests.

Consider the specific assets in play. The New York Times crossed 10 million digital subscribers in February 2023. Bloomberg has roughly 400,000 paid subscribers who are among the most financially active people on the planet. Spotify has approximately 600 million monthly active users across 180 markets. The Economist, the Financial Times, Paramount Plus, Disney Plus, Warner Bros. Discovery’s streaming services and thousands of specialist publishers in dozens of languages, these are not substitute audiences for Meta’s social graph. In many advertising categories they are superior audiences: more engaged, more verifiably identified, more contextually appropriate, and more willing to pay for what they are consuming, which is a reliable signal about disposable income.

The publisher whose logged-in subscriber audience receives a $3 CPM when that audience is worth $15 to the right advertiser does not have an audience problem. It has a supply chain problem. The supply chain between the publisher’s content and the advertiser’s budget has been too focused on its own margin to communicate the audience’s value clearly. The farmer’s wheat is better than the bulk trader claims. The bulk trader has never let the buyer see it without his label on it.

The SP500+: The Trade Desk Does What Nobody Else Would

In February 2024, The Trade Desk began beta testing a product called the Sellers and Publishers 500+, deliberately named to evoke the S&P 500. The concept: a curated, actively maintained list of premium open internet inventory sources, evaluated at the individual ad placement level rather than just the publisher level. Initial participants in the beta included the New York Times, Disney Plus, Hulu, ABC, and the Wall Street Journal. Spotify was added subsequently. The product is globally available across connected television, web, and digital audio.

The Trade Desk’s Marketplace Quality team, which does the thankless ongoing work of keeping made-for-advertising inventory and fraudulent traffic out of the platform, underpins the whole exercise. The SP500+ is essentially that quality work made visible and actionable for buyers: instead of maintaining their own exclusion lists, advertisers can select SP500+ inventory and outsource the quality management to The Trade Desk’s ongoing curation.

The SP500+ directly addresses the BBB-dressed-as-AAA problem. It creates a transparent, auditable definition of what ‘premium open internet’ actually means and makes it actionable with a single product choice. This is something an SSP, the IAB, a publishers’ trade body, or an industry consortium could have built years ago. None of them did, because defining quality means excluding inventory that someone is currently monetising, and nobody wanted to be the one who drew that line.

The legitimate concern the industry has expressed about this is worth stating plainly: if buyers route spend primarily through SP500+ inventory that The Trade Desk curates and controls the criteria for, the DSP has moved from being a neutral buy-side technology to being a gatekeeper of the open internet. Green’s structural argument, that TTD’s buy-side-only model prevents this from becoming a true walled garden because the company has no supply-side inventory to favour, is credible but not complete. The SP500+ is a significant expansion of what a DSP has historically been willing to decide unilaterally. You can acknowledge that the work was necessary and nobody else was doing it while also noting that it transfers considerable power to the entity curating the list.

The Ad Server: Where All the Truth Lives

The ad server is the technology that determines which ad appears on a publisher’s page, at what price, in what order, following what rules. It is the point in the supply chain where information asymmetry is smallest: the publisher knows what they received and, with direct access, the buyer knows what they paid. In display advertising, Google Ad Manager dominates publisher ad serving to a degree that forms a central argument in the DOJ antitrust case. The company selling publishers the technology that manages their inventory also operates the exchange that buys that inventory and the demand-side platform that bids on it. This is not a coincidence. It is a competitive strategy that the DOJ is currently examining in considerable detail.

In connected television, two significant alternatives have emerged. Freewheel, acquired by Comcast in 2014, is the primary ad server for broadcast and cable networks making the transition to streaming. SpringServe, acquired by Magnite in 2021, became the principal independent alternative in the streaming space. And in April 2025, Magnite did something the industry had been debating for years: it actually merged the two.

On April 23, 2025, Magnite combined the SpringServe ad server with the programmatic capabilities of the Magnite Streaming SSP into a single unified CTV and OTT platform, initially in closed beta. Initial clients: Disney Advertising, LG Ad Solutions, Paramount, Roku, Samsung, and Warner Bros. Discovery. Jounce Media’s March 2025 Supply Path Benchmarking Report verified the platform connects buyers to 99 percent of US streaming supply on a dollar-weighted basis. General availability was targeted for early to mid-summer 2025. Sean Buckley, Magnite’s President of Revenue: ‘By unifying the programmatic layer as a complementary step in the buying process, it gives buyers greater transparency, predictability, and control over their ad placements, and lays the foundation for more effective monetisation and yield management for media owners.’

A publisher using Magnite’s unified SpringServe platform controls their yield management, their demand access, their supply path decisions, and their deal management in one system. That is a publisher operating as a technology company rather than as a technology customer. Will Doherty, SVP Inventory Development at The Trade Desk, endorsed the move directly at launch: ‘Magnite helps fuel the premium, open internet. Combined with tools like OpenPath, the next generation of SpringServe is accretive to advertisers and publishers.’ Two sides of the same transaction, for once, pointing in the same direction.

The unified ad server and SSP model is the correct architectural direction. A publisher who controls both the technology managing their inventory and the platform connecting it to demand has closed the information gap that has defined their relationship with the supply chain since programmatic began. They know what their wheat is worth. They can see what it traded for. They can compare the two. That is the foundation on which everything else in this article is built.

07

Agentic AI: The Technology That Enforces What Contracts Cannot

On November 13, 2025, the IAB Tech Lab released the Agentic RTB Framework, version 1.0, for public comment. If you missed the announcement in a busy November news cycle, here is what you missed: the most consequential technical standard in programmatic advertising since OpenRTB standardized the auction protocol in 2010.

The ARTF uses containerized architecture to bring demand-side platforms, supply-side platforms, data vendors, and bidding algorithms into the same virtual processing environment, reducing bid request-response latency by up to 80 percent, from the current 600 to 800 milliseconds to approximately 100 milliseconds. Netflix, Paramount, The Trade Desk, Yahoo, Index Exchange, and Chalice were among the early supporters. Anthony Katsur, CEO of IAB Tech Lab: ‘ARTF establishes a true control plane for an agentic future, where autonomous agents and specialised software enhance the bidstream in real time with rigour, safety, and interoperability.’ Joshua Prismon of Index Exchange called it ‘production-tested’ and ‘transformational’ at the November launch.

What the ARTF actually does, in plain language, is allow AI agents to operate inside the auction process as active verifiers rather than passive rule-followers. Within the microseconds available before a bid is placed, an AI agent can check whether the supply path for this specific impression matches the contracted route, whether the floor price the SSP declared matches what the publisher actually set, whether the inventory type is classified correctly, and whether the fee being applied is what was agreed in the contract. The 80 percent latency reduction is not an engineering efficiency. It is the time budget that makes real-time contract enforcement commercially feasible for the first time.

What ARTF Enables and What It Does Not Do Automatically

What it enables: per-impression supply path verification; real-time floor price confirmation; automated fee discrepancy detection across millions of transactions per hour; MFA and inventory quality evaluation within the auction window; and impression-level transaction records creating an auditable supply chain trail. What it does NOT do automatically: enforce anything. The technology detects discrepancies. It acts on them only if the contract specifies what action to take. An AI agent can halt spend on a non-compliant supply path automatically, but only if the buying contract says it should. The technology is the enabler. The contract is the instrument. The buyer or publisher who writes the rules the agent follows decides what actually changes.

Start on the Publisher Side, Not the Demand Side

There is a strong argument, and it is the correct argument, that AI agents should be introduced on the publisher side before the demand side deploys its own. Here is the logic. A demand-side AI agent built by or for an agency that profits from principal buying will be optimized for outcomes that preserve that profit. An SSP’s AI agent will be optimized for the SSP’s revenue. The only participant whose AI agent has no structural conflict of interest is the publisher. The publisher’s agent, sitting at the ad server, has one job: maximize the value of this impression for this publisher. It has nothing to gain from hiding the price from itself.

A publisher who deploys an ARTF-compatible AI agent at their ad server can verify, at the impression level, whether the floor price declared to buyers matches what the publisher actually set. Whether the winning demand source paid what it said it would. Whether the supply path routing matches the contract. The publisher’s agent is not trying to extract margin from anyone. It is trying to verify that the publisher received what it was owed.

When the publisher publishes those verified transaction records at the campaign and supply path level, they become the independent audit the industry has never produced voluntarily. A buyer’s AI agent can compare the publisher’s transaction record to its own. Discrepancies surface automatically. Contracts that specify consequences for discrepancies become enforceable in real time rather than through a formal audit process that takes weeks and produces a report that gets disputed for months.

The ARTF is not a finished product. Version 1.0 is a foundation. Adoption is in closed beta with initial supporters. But the IAB Tech Lab’s track record with OpenRTB, which became the universal auction standard within three years of its introduction, suggests that adoption will accelerate significantly in 2026 and 2027. The agencies and intermediaries that believe the business model they have run for twenty years will survive this shift are making the same bet that Blockbuster made when Netflix started mailing DVDs. The technology exists. The standard exists. The question is how quickly the contracts require it.

And this is where the first-principles argument becomes urgent rather than merely instructive. With agentic AI entering the adtech stack — with the IAB Tech Lab’s ARTF and the emerging AdCP standards creating the infrastructure for autonomous agents to operate inside the auction process — the industry is at a fork in the road. One path: deploy AI as a smarter version of the existing architecture, automating the same intermediary layers, preserving the same structural information asymmetries, making the same toll road faster and harder to audit. The other path: use AI to actually go back to basics. Supply and demand. Publisher and advertiser. A clear answer to the question “what is this inventory worth, and did the money reach the person who created it?” If the industry does not choose the second path consciously and deliberately, agentic AI will default to optimizing the first one, because that is the architecture it will inherit. The fundamentals do not disappear because the technology becomes more sophisticated. They become more consequential. An industry that cannot explain its own supply chain from first principles to a first-year economics student has no business layering autonomous agents on top of it and calling that progress. Stop. Think from the basics. Then build.

The Publisher-Led Agentic Supply Chain: How It Works in Practice

Step 1: Publisher controls its own ad server. In CTV, that now means Magnite’s unified SpringServe or Freewheel. In display and every other channel, it means an independent alternative to Google Ad Manager.

Step 2: Publisher activates first-party audience data within the ad server to demonstrate audience quality to buyers at the impression level.

Step 3: Publisher deploys an ARTF-compatible AI agent that verifies every transaction against contracted terms in real time and maintains an impression-level transaction record.

Step 4: Publisher negotiates contracts with SSPs and DSPs specifying that floor prices are verifiable in the transaction record, supply path routing is disclosed, and discrepancies trigger automatic consequences rather than letters to lawyers.

Step 5: The publisher’s verified transaction records are accessible to buying partners, creating the real-time audit trail that quarterly formal audits have never consistently produced. Every component of this stack exists today.

08

SSPs: The Data Platform or the Toll Booth

The supply-side platform was built to aggregate demand for publishers who could not independently access multiple buyers simultaneously. In 2015, that function justified twenty-plus independent SSPs. In 2026, direct connectivity tools like the PubMatic’s Activate platform, which grew more than threefold in 2025 and allows advertisers to buy directly within the SSP bypassing DSP intermediation, are all testing the same question: how many SSP layers does the supply chain actually need?

AI supply path optimization is the automated version of the question buyers have been asking manually since 2019. An AI agent routing spend to the best-performing supply paths, verified at the impression level, does not favour an intermediary that cannot demonstrate its contribution. The SSPs that have a future are the ones controlling something that AI cannot route around.

The One Asset Worth Building: Aggregated Publisher First-Party Data

Individual publishers have rich first-party audience data. Logged-in users. Behavioural signals within the publisher’s ecosystem. Content consumption patterns. The problem is that individual publishers outside the very largest cannot make this commercially valuable at scale on their own. A regional news publisher with two million monthly readers has excellent audience data. It cannot sell that data to global advertisers the way Meta can.

An SSP that aggregates first-party data across hundreds of publisher partners, with consent frameworks and publisher agreements that make the arrangement transparent, creates something no DSP can replicate through direct publisher connections: a scaled, consented, cross-publisher audience graph that is genuinely differentiated from walled garden data. The SSP that moves a publisher CPM from $3 to $5 by demonstrating, through aggregated audience data, that an impression reaches a logged-in user whose behaviour matches an advertiser’s target, is adding transparent, measurable, auditable value. The SSP that adds a margin to transactions it did not improve is the one that AI routes around.

Magnite has not posted an annual net profit since 2019 and invests over $200 million annually in R&D. On the Q1 2025 earnings call, CEO Michael Barrett’s primary growth argument was that the Google antitrust ruling ‘could significantly increase our monetization opportunities and market share, possibly as soon as next year.’ That may be true. It is also a strategy whose timeline is entirely outside Magnite’s control. The SpringServe unification is a more durable bet because it is entirely within Magnite’s control and delivers something publishers demonstrably want: a single system for their entire monetization operation. Building on that is the right path forward.

09

DSPs and Agencies: The Clock Is Running on Both

The demand-side platform is structurally the correct model for transparent programmatic buying. Works exclusively for the buyer. Charges a disclosed technology fee. Has no supply-side inventory to secretly favour. This is the commission agent from the grain market: declaring its fee openly, competing on service quality, passing the rest through. The model is right. The metric DSPs have historically competed on is wrong.

Reach, frequency, and cost per thousand impressions are activity metrics. Return on advertising spend is the outcome metric that actually matters. The DSP that consistently delivers better ROAS than its competitors can charge a higher technology fee and justify it with a verifiable track record. The DSP competing on CPM against Amazon, which reportedly offers fees as low as 1 percent for major spending clients because its real margin is on Prime Video and Thursday Night Football, is not going to win that argument on price alone.

StackAdapt’s numbers are worth examining carefully. The Globe and Mail reported in February 2025 that StackAdapt generated nearly $400 million in revenue in FY2024 and was expected to surpass $500 million in FY2025 with more than $125 million in operating earnings, making it one of Canada’s most profitable private technology companies. Ontario Teachers’ Pension Plan led a $235 million funding round. Six consecutive years rated number one DSP on G2 by user satisfaction. Founded by former WPP employees who built the platform they wished existed when they were agency practitioners. That origin story tends to produce user-focused product development rather than billing-optimised product development, and the customer satisfaction ratings reflect it.

For agencies, the financial advisor parallel remains the clearest template. The UK Financial Conduct Authority’s Retail Distribution Review, effective January 2013, banned commission payments on retail investment advice and required advisors to charge explicit fees. The number of independent financial advisors fell from approximately 35,000 in 2011 to approximately 22,000 by 2014. The ones who left were the ones whose income depended on the commission structure rather than the quality of the advice. The ones who survived competed on demonstrable investment outcomes. WPP lost approximately 62 percent of its market value in 2025, saw WPP Media decline 10.8 percent in Q4 on a like-for-like basis, and lost the Mars global media account (approximately $1.7 billion) and the Coca-Cola North America media account (approximately $700 million) to Publicis. The financial advisor parallel does not require a regulator. The market is applying the same pressure on its own schedule.

10

The Supply Chain That Could Exist: A Dollar Comparison

The walled garden delivers 80 cents to the publisher because the publisher IS the walled garden. There is no external supply chain extracting a toll because the platform owns both sides of the transaction. The open internet’s answer to this cannot be another transparency pledge. It has to be 70 cents, delivered verifiably, confirmed by AI agents at the impression level, with ROAS outcomes that an independent firm can audit and compare to walled garden alternatives.

From 51 cents in 2020 to 65 cents in 2022 in the ISBA/PwC premium UK market study. From 36 cents in 2023 to 43.9 cents in 2024 in the broader ANA US benchmark. The trajectory is real. What accelerates it is publishers who control their own technology, advertisers who require log-level verification, SSPs that compete on publisher first-party data rather than inventory control, DSPs that publish ROAS benchmarks rather than reach metrics, agencies that separate advisory fees from execution margin, and AI agents whose rules are written by the people whose money is being spent.

11

The End: Or the End of the Beginning

Back to India for a moment. India’s e-NAM platform, the National Agriculture Market, launched in 2016 to give farmers access to the price information that licensed marketplace traders had held exclusively for decades. A farmer can now compare prices across multiple buyers, sell into distant markets, and verify that the weight recorded at the point of sale matches what was actually loaded. The system is imperfect. Implementation is uneven. Some traders have maintained advantages through side arrangements. But the structural condition that allowed two cents to the farmer and ninety-eight to the intermediary is being disrupted, slowly and unevenly, by the simple act of making price data publicly available.

The advertising technology industry’s e-NAM moment is arriving on multiple fronts simultaneously. The ARTF enables impression-level verification. The Magnite SpringServe unification gives CTV publishers integrated control over their ad serving and SSP in one system. The Trade Desk’s SP500+ creates the first actionable definition of premium open internet inventory. The DOJ antitrust case is extracting from Google the specific economic details of how publisher ad serving dominance translates into programmatic market power. AI attribution tools are making the revenue contribution of advertising spend measurable with a precision that did not exist three years ago. And the CFO is now in the room.

None of this requires virtue. None of it requires anyone to voluntarily give up income. The ISBA/PwC improvement from 51 cents to 65 cents reaching publishers came from measurement pressure, not moral awakening. The SP500+ exists because The Trade Desk calculated that cleaning up the open internet serves its own commercial interests. The Magnite SpringServe unification happened because publishers asked for it. Everything that is working is working because someone calculated that transparency was in their commercial interest. That is fine. That is how markets are supposed to work.

The open internet currently holds approximately 20 percent of global digital advertising. It is heading toward 17 percent by 2027 on current trajectory. Seventeen percent of an expanding market is still a large absolute number. But it is also a market that is losing competitive ground every year to platforms that are less diverse, less editorially independent, and less capable of reaching audiences across the full breadth of human content consumption. That matters for reasons beyond advertising economics. The publisher whose revenue falls cannot maintain the newsroom or the production budget or the creator community that made the audience worth advertising to in the first place.

The user watching a show on Paramount Plus, reading the Washington Post, listening to a podcast, scrolling a newsletter, all of them are tolerating the advertising because the advertising subsidizes the content they actually want. That is the deal. It is still a reasonable deal. It just requires the supply chain executing it to actually deliver the money to the people making the content, rather than treating the journey between advertiser and publisher as an opportunity for forty companies to clip the ticket.

The publishers sitting on those audiences have more leverage than they are currently exercising. The technology to exercise it exists today. The contracts to enforce it can be written this quarter. The AI agents to verify it in real time will be commercially widespread within two years. The only barrier is the institutional inertia of participants who benefit from the current arrangement and would prefer to keep discussing it at conferences rather than changing it at the contract level.

The farmers have smartphones now. The price data is available. The supply chain that built a business model on the farmer not having that information is going to have to find a different reason to exist.

Some of them will. The ones who add genuine value, who help publishers activate their audience data, who verify supply path integrity, who deliver ROAS outcomes that hold up under independent scrutiny, those intermediaries will thrive in the transparent supply chain because transparency confirms their value rather than threatening it.

The ones who got paid because nobody looked closely enough will find the next few years considerably more interesting than the last fifteen.

One last thing, and this is the reason this article exists. Every time adtech faces a genuine structural reckoning — privacy sandbox, the DOJ case, the ISBA/PwC findings, the arrival of AI — the industry’s first instinct is to reach for the vocabulary of the existing architecture. New technical standard? Build a working group. New privacy requirement? Launch a coalition. New technology? Retrofit it onto the current stack and give it a new acronym. That instinct is understandable. It is also the instinct that has kept the farmer at two cents and the pension fund manager holding AAA-rated BBB paper. The only way out is to stop, get off the bandwagon for a moment, and ask the genuinely simple questions. What is this industry, at its most basic level? Supply and demand. Publishers and advertisers. Anything in the middle needs to justify its existence by making that exchange more efficient and more valuable — not by making it more opaque. If your business model requires the publisher not to know what their inventory is worth, and the advertiser not to know where their money went, you are not adding value. You are making hay while the sun shines. And the sun, as this article has attempted to document in considerable empirical detail, is beginning to set on that particular field.

Everybody got paid. Including the people nobody hired. That part is about to get a lot harder to pull off.

Disclaimer: I am writing this to the best of my knowledge of whatever time I have spent in my professional life in the mediatech and adtech industry and whatever I could learn either from experience or from others or by media reports. In no way, it reflects on any insider information or anything that could be confidential. Also, it has no input from my current or any of the previous employers and is written in personal capacity with publicly available information. It is just my attempt to summarize the whole discourse that’s going on in recent times in the trade press about the adtech industry in general. This is my attempt to summarize everything as a knowledge base for people to learn more about adtech industry in general and how everyone has been operating for more than a decade. There are 100s of experts who know more than me and if you are one of them reading this, please comment or let me know about inaccuracies mentioned in any way. Like everyone else, Claude has helped me write this article in a journalistic tone to cover facts without adding my personal opinion. Lastly, if you are offended and are planning to sue me, please don’t. I am not so rich that you can make money of it.

Here is a story about a company that spent fifteen years building an industry on a simple, radical idea: tell the truth about what you charge, show buyers exactly what they are buying, and never work for both sides of a transaction at the same time. In an industry that had made opacity its art form and its business model, this was not just unusual. It was a genuine competitive threat — to everyone who had been making a comfortable living from the opacity.

Here is also a story about the very powerful companies that were built on the opposite principle — the idea that complexity, secrecy, and a creative relationship with client budgets could generate extraordinary income if advertisers never asked too many questions. These companies cheered The Trade Desk when it was useful to them. They championed it, partnered with it, named it their preferred technology. And then, when The Trade Desk’s growth began to illuminate exactly the practices they had been quietly running for years, some of them picked up a flag labelled ‘transparency’ and used it to beat the transparency company over the head.

This is a story with a genuine protagonist, a set of genuine conflicts of interest, and — in the spirit of honest journalism — one chapter where the protagonist also trips over its own shoelaces. In public. Loudly. More than once.

To understand all of it, you have to start not in 2026 but in 2009, in Ventura, California, where a man named Jeff Green decided to build something the advertising industry had never seen — and which certain parts of it were hoping it would never have to.

01

The Honest Broker — How Jeff Green Built the Company the Industry Needed

The programmatic advertising ecosystem that Jeff Green entered in 2009 was, to put it charitably, a confidence game with better branding. Advertisers handed enormous budgets to agencies. Agencies bought media on their behalf — or, increasingly, bought it for themselves at undisclosed prices and sold it to clients at a markup, keeping the difference. Supply-side platforms, ad exchanges, and demand-side platforms formed a chain of intermediaries each collecting a slice that was individually invisible but collectively vast. The entire system ran on a foundational assumption: that clients would not, or could not, look too closely.

Green and his co-founder Dave Pickles had both worked inside this system. They understood where the money went and why it went there. And they made a deliberate, calculated, and at the time quite contrarian bet: build a demand-side platform that was transparent about what it charged, worked exclusively for buyers and never for sellers, maintained no supply-side relationships that could create conflicts of interest, and showed advertisers precisely what they were buying, at what price, in what context.

This was not idealism. It was a business model built on a specific insight: the agencies whose clients were beginning to ask awkward questions about where all the money went would pay a premium to use a platform they could actually defend in a board meeting. The platform that made transparency its product would win in a market that was slowly, reluctantly, being forced to confront its own opacity.

The bet paid off beyond almost anyone’s expectations. The Trade Desk went public on September 21, 2016, priced at $18.00 per share. On that first day, it opened at $28.75 and closed at $30.10 — up 67% in a single session. Over the following years it delivered revenue growth above 25% annually with remarkable consistency. Customer retention stayed above 95% — a figure that speaks to something beyond contractual lock-in. It speaks to a platform that was, year after year, doing what it said it would do.

The company’s all-time high closing price was approximately $139–141, reached on December 4, 2024. At its market peak it was valued at tens of billions of dollars. And in 2021, in what was perhaps the most explicit public endorsement the company had ever received, Publicis Groupe — the world’s most valuable advertising holding company — announced that The Trade Desk would be the exclusive third-party DSP partner for Epsilon’s Core ID, Publicis’s flagship identity solution. The world’s biggest holdco naming TTD as its preferred technology partner. A formal declaration of trust in what Green had built.

File that one away carefully. We will need it in about six chapters.

The Trade Desk’s Structural Foundations — Why They Matter

Buy-side only: TTD has never operated a supply-side platform, never owned a media property, and never taken supply-side money. Its entire business model depends on advertisers getting good results — because that is the only reason they keep paying. No walled gardens: TTD does not own the inventory it trades and cannot secretly favour its own media. Transparent fees: TTD charges a disclosed technology fee on managed spend. No rebates, no principal mark-ups, no ‘non-product related income.’ These are structural facts about how the company is built, not marketing claims.

02

The Fire Alarm Nobody Wanted to Hear — 2016 and the Rebate Reckoning

In March 2015, a man named Jon Mandel stood up at the ANA Media Leadership Conference in Hollywood, Florida, and said the thing the entire room had been pretending was not true. Mandel, the former CEO of Mediacom, one of the world’s largest media agencies, alleged bluntly that media agencies were systematically collecting undisclosed rebates from media companies — payments for directing client budgets toward those media companies — and keeping the money rather than returning it to the clients whose spend had earned it.

The ANA commissioned a seven-month independent investigation by K2 Intelligence, one of the most reputable research firms in the field. The report landed in June 2016. One hundred and forty-three interviews. One hundred and fifty individual sources. The findings were unambiguous.

ANA / K2 Intelligence Report, June 2016 — Verified Public Document

Non-transparent practices including cash rebates were found to be ‘pervasive’ in the U.S. media ecosystem. Senior executives were aware of and had mandated some non-transparent practices. Markups on principal media transactions ranged from approximately 30% to 90%. Media buyers were ‘sometimes pressured or incentivised by their agency holding companies to direct client spend’ to principal media regardless of client best interests. Of 41 sources confirming rebate deals exist in the U.S., 34 confirmed the rebates were undisclosed to clients.

Five of the six major agency holding companies declined K2’s request to make current executives available for interview. The American Association of Advertising Agencies called the report ‘anonymous, inconclusive, and one-sided.’ The irony of an industry transparency report being called opaque by the companies who declined to participate in it was apparently lost on the people issuing the statement.

The industry moved on. Some contracts were updated. Some clients asked better questions for a while. The underlying practices did not disappear. They found new vocabulary — ‘principal-based buying,’ ‘inventory investment,’ ‘non-product related income’ — and in some cases simply grew. The Trade Desk kept building. Kept being transparent. Kept growing.

The 2016 ANA/K2 report is not ancient history. It is the context that makes the 2026 dispute comprehensible. Every executive at every agency holding company who is now raising transparency concerns about The Trade Desk was working in this industry when that report was published. Some of them had been interviewed for it.

03

The China Raid — When Allegation Became Criminal Investigation

In October 2023, Chinese law enforcement officers walked into GroupM’s offices in China. More than three employees were detained as part of a state investigation. The subject of the investigation was the systematic retention of client rebates — precisely the practice Jon Mandel had described at a conference in 2015, and precisely the practice documented in the ANA/K2 report in 2016. Seven years later, in one of the world’s largest advertising markets, employees of what was then the world’s largest media-buying network were being detained by police over it.

GroupM is the media-buying arm of WPP. Campaign Asia-Pacific had spent two years investigating rebate-driven practices in China linked to former GroupM employees. WPP confirmed the detentions publicly.

This Is Not an Allegation

A law enforcement raid is a documented legal event. Chinese authorities raided the offices of GroupM — WPP’s media-buying arm — and detained employees. The subject was the retention of client rebates. WPP confirmed the detentions. This is verified fact, not an accusation from a rival or a whistleblower.

The industry largely moved on. Some reporting, some concern, and then the news cycle continued. But the raid was a very loud data point: the practices documented as pervasive in the United States in 2016 were being treated as criminal violations in China in 2023. The same structural logic. A different jurisdiction’s patience.

04

OpenPath — When Transparency Becomes a Commercial Threat

In 2022, The Trade Desk launched OpenPath, its direct-to-publisher buying programme. The concept was both elegant and threatening to a large number of people who had been making money from the absence of what it offered. OpenPath allowed advertisers to buy publisher inventory directly, bypassing the chain of supply-side platforms and intermediaries that each took a cut before the money reached the publisher. More of the advertiser’s budget reaches actual content. A cleaner, more verifiable supply chain.

Jeff Green positioned it as exactly what it was: a transparency tool. Publishers liked it because they received more money. Advertisers liked it because they received better value. By 2025 it had more than 400 publisher partners integrated, and Green had publicly anticipated that year as the beginning of OpenPath’s ‘steep acceleration phase.’

The agencies did not like it, and the reason is not complicated. OpenPath performs, structurally, the function that agency trading desks sell to clients as a service. When a transparent DSP tool performs supply-path optimisation transparently, at cost, the agency’s ability to charge clients for that optimisation — and, more significantly, to execute those optimisations in ways that also happen to generate proprietary margin — becomes very hard to justify.

In late February 2026, AdWeek exclusively reported that Dentsu had quietly disabled OpenPath entirely, after using it since its 2022 launch. WPP had withdrawn shortly after the programme’s launch and had never used it in markets including Australia. Both agencies cited concerns about fee visibility and uncertainty about ad placement transparency.

The Trade Desk’s CMO Ian Colley responded directly and specifically on LinkedIn: ‘TTD doesn’t push spend to OpenPath. It’s not a marketplace or curated inventory. OpenPath is offered at cost to the ecosystem. We’ve been clear about that. There are no hidden fees beyond that. If OpenPath is selected by an advertiser, it is because it represents the cleanest and most cost-efficient path.’

Worth Noting on Both Sides

The agencies’ stated concerns about OpenPath fee transparency have not been independently verified by any published audit. However, the commercial incentive to raise those concerns — since OpenPath competes directly with services agencies charge for — exists entirely independently of whether the concerns are valid. Both things can be true simultaneously: TTD may have had imperfections in OpenPath’s communication, and the agencies may also have had financial motives for exiting that had nothing to do with their clients’ interests. The timing also matters: OpenPath launched in 2022. The agencies did not exit in 2022. They exited in late 2025 and early 2026, at a moment when TTD was at its most commercially vulnerable.

05

Richard Foster and the Report That Became a $100 Million Lawsuit

Richard Foster spent seventeen years at GroupM, WPP’s media-buying arm — now rebranded as WPP Media. He rose to become the Global CEO of Motion Content Group, GroupM’s entertainment investment division. His division co-produced more than 2,500 television series during his tenure, including Love Island, and managed roughly $500 million in annual GroupM entertainment investment. In his final year, his US operation reportedly posted 140% revenue growth. He established Motion Content Group in May 2017, specifically structured to operate independently from GroupM Trading’s media inventory practices, and focused on compliance with client contractual obligations. He was, by any measure, a successful and long-serving senior executive.

In December 2024, Brian Lesser — the incoming CEO of GroupM, which was in the process of being rebranded as WPP Media — asked Foster to prepare a strategic assessment of the division’s operations and potential. Foster produced what became known internally as Project Claridges: an approximately 35-36 page document that did two things simultaneously. It outlined a proposal for a new consolidated entertainment division projected to generate net sales exceeding $2 billion by 2029 at profit margins above 70%. And it raised serious internal concerns about GroupM’s trading practices, concerns Foster had first raised internally as far back as a 2016 meeting with former GroupM executives.

What the Lawsuit Alleges — With WPP’s Response

The following is what Foster’s lawsuit alleges. WPP has filed a motion to dismiss, disputes the material characterisation of all practices, and the case is before the courts with no ruling yet issued.

According to the complaint, GroupM used its approximately $60 billion in annual client advertising spend to negotiate volume-based discounts and rebates from media vendors. Rather than returning these benefits to clients, the complaint alleges GroupM reclassified the resulting inventory as ‘proprietary media,’ sold it back to clients through opt-in agreements, and booked the spread as what internal documents reportedly called ‘non-product related income.’ An internal presentation Foster submitted as part of the Project Claridges report allegedly estimated GroupM derived nearly $1 billion annually from this income stream, with an internal growth target of 15% per year. Foster estimated that over the period 2019 to 2024, GroupM generated $3 to $4 billion from such deals and improperly retained approximately $1.5 to $2 billion of it.

According to the complaint, when Foster submitted the report, Lesser initially expressed concern and said he would investigate. He then, unbeknownst to Foster, forwarded the original report to Mark Patterson — the executive responsible for GroupM’s trading activities. Patterson is currently WPP Media’s Global President of Markets and Business Operations. Hours after Lesser separately asked Foster via text to produce a ‘sanitised’ version of the report excluding criticism of GroupM, a restructure was announced placing Foster’s division under Patterson’s direct oversight. Foster alleges he was then excluded from key meetings, removed from deals he had built, and gradually isolated from decision-making. On July 10, 2025 — the day after WPP’s stock fell 18% on a trading update disclosing serious deterioration at WPP Media — Foster was fired.

In November 2025, Foster filed the lawsuit in the Supreme Court of New York, seeking more than $100 million in damages. He alleges wrongful termination, retaliation, and violations of whistleblower protection statutes in California and New York. The individual executives named as relevant non-parties in the complaint include Mark Patterson (WPP Media’s Global President of Markets and Business Operations), Andrew Meaden (Global Head of Investment at WPP Media), and Nicola McCormick (Global General Counsel for GroupM). The specific allegations against each are described in the complaint, which WPP disputes in full.

WPP’s Defence — In Full, as Published

WPP has filed a motion to dismiss. Its key arguments, as reported by Digiday based on court filings, are: (1) According to a sworn affirmation from Lesser, Foster’s counsel sent WPP a draft complaint on October 10, 2025 — more than two months before filing — and threatened to go public unless GroupM agreed to a large severance payment within 30 days. WPP argues offering to stay silent for a payout is incompatible with being a whistleblower. (2) WPP contends Project Claridges contains no mention of illegal activity and is a business proposal tied to Foster’s own ambitions, not a whistleblower document. (3) Foster was among hundreds of U.S. employees let go in a routine organisational restructuring. WPP states: ‘The court has not yet made any findings in relation to the allegations and we will defend them vigorously.’

The Number From WPP’s Own Filing

Among the internal documents that entered the public record through WPP’s own court filings was a piece of data that is striking on its face. Across GroupM’s top 30 U.S. clients — representing $13.4 billion in total billings — 97.4% of the proprietary inventory from which GroupM was allegedly generating income had not been used by the opted-in clients. Google, GroupM’s single largest U.S. client at $2.3 billion in annual billings, had utilised just 0.51% of the proprietary inventory its budget was helping to generate. This data was in WPP’s own filing, submitted as part of its motion to dismiss, and was first reported by The Times and subsequently by Digiday.

WPP’s purpose in submitting these documents was to characterise Project Claridges as a routine business proposal rather than a whistleblower disclosure. The commercial picture the documents created in the public record was not something WPP anticipated becoming the story. Ivan Fernandes, a former WPP executive now advising other groups, described the filing as ‘commercially significant’ in comments reported by The Times.

The case is ongoing. No court has made findings on any of the allegations. Both the specific claims and the defences deserve to be heard in full before conclusions are drawn. What is clear is that the financial picture described in the publicly filed documents has given the advertising industry’s clients a great deal to think about.

06

Agencies Take the High Ground — A Brief and Ironic History

It is worth documenting, clearly and chronologically, how enthusiastically the major agency holding companies supported The Trade Desk — before they found it commercially inconvenient to do so.

When TTD went public in 2016, holding company trading desks were among its most significant clients. GroupM, Publicis’s Starcom, IPG’s Mediabrands, Dentsu — all were buying through TTD, recommending it to clients, and in several cases publicly praising its approach. The model Green had built — powering the agencies rather than competing with them — was working exactly as intended. The agencies loved having a best-in-class transparent DSP they could point to when clients asked awkward questions about programmatic buying.

In 2021, Publicis named TTD the exclusive third-party DSP partner for its Epsilon Core ID. Not a quiet commercial arrangement. A loud, public endorsement from the world’s largest advertising holding company by market capitalisation, telling the market: we trust this platform above all others.

The dynamic began shifting after 2022, as The Trade Desk launched a series of initiatives that were simultaneously transparency tools and competitive threats to specific agency revenue lines. OpenPath (2022) competed with agency supply-path services. Kokai (2023-24) automated functions agencies traditionally charged for. OpenAds (2025) built an alternative auction infrastructure. Ventura (2025) put TTD into the CTV operating system business. Each was framed accurately as advancing the open internet. Each also happened to undermine a specific way agencies were extracting margin from client budgets.

The Progression — From Partner to Auditor: A Verified Timeline

2016-2021: Agencies enthusiastically use and recommend TTD. GroupM, Starcom, Mediabrands, Dentsu all buy through the platform.

2021: Publicis names TTD exclusive DSP partner for Epsilon Core ID — a flagship public endorsement.

2022: TTD launches OpenPath, directly competing with agency supply-path optimisation services.

2023-24: Kokai AI platform automates campaign functions agencies charge for. TTD and agencies begin to experience operational friction.

2025: WPP and Dentsu quietly exit OpenPath, citing fee and transparency concerns. TTD launches OpenAds and Ventura. WPP stock falls 62% in 2025. Foster lawsuit filed November 2025.

Early 2026: Publicis commissions FirmDecisions audit of TTD. Issues memo advising clients to stop using it. The relationship publicly celebrated in 2021 becomes, five years later, the subject of a client advisory.

Green’s frustration is understandable and his characterisation is, on the available facts, substantially supported. The companies now auditing TTD’s fee structure are the same companies whose practices were documented in the 2016 K2 report, whose Chinese offices were raided in 2023, and whose internal documents — submitted in their own court filings — describe a financial model whose beneficiaries, according to those same documents, were not primarily the clients whose budgets funded it.

The question is not whether the agencies are hypocritical. The question is whether their specific concerns about TTD are valid. Those are different questions, and conflating them is a mistake both sides have made.

07

The Epsilon SSP — Transparency Concerns, Apply Elsewhere

In September 2025, AdWeek exclusively reported that over the prior 18 months, the media-buying arms of WPP, IPG, Dentsu, Havas, and at least four independent agencies had been purchasing ad inventory indirectly through Publicis’s own Epsilon supply-side platform. In some cases, apparently without full visibility into where their supply chain was routing client money.

Publicis, through its 2019 acquisition of Epsilon for $4.4 billion, had built an SSP that was operating as a competitor to other agency holding companies. The agencies buying through it were, at least in some cases, routing client budgets through a direct rival’s infrastructure, potentially giving that rival access to buying patterns, audience data, and campaign intelligence.

No one commissioned an independent audit of this situation. No holding company issued a client memo advising its clients to pause using the relevant supply chain until the conflict was reviewed. The trade press covered it, the industry discussed it, and life continued. The contrast with the formal audit process subsequently applied to The Trade Desk is a question worth sitting with.

08

The Trade Desk’s Difficult Year — In the Spirit of Honest Reporting

A story told fairly requires honesty about the protagonist’s stumbles alongside its strengths. The Trade Desk had a difficult 2025 that was not entirely other people’s fault, and deserves to be reported as such.

The First Earnings Miss in Company History

In the fourth quarter of 2024, The Trade Desk missed its quarterly earnings expectations for the first time in its history as a public company. Bank of America’s analysts cited one primary factor in their note: poor execution on the rollout of Kokai, TTD’s AI-powered campaign management platform. Agencies that had used TTD’s familiar interface for years were asked to adopt a new system, and TTD’s onboarding support was not adequate to the scale of the transition. Clients experienced frustration. Campaigns ran less efficiently during the migration. Revenue guidance came in below expectations. CEO Jeff Green issued a rare public apology to investors. The stock fell approximately 25% on the news.

Two CFOs in Under Twelve Months

In August 2025, TTD announced that CFO Laura Schenkein — a long-serving company veteran — was departing, to be replaced by Alex Kayyal, a board member from Lightspeed Ventures with no prior experience as a public-company CFO. The announcement was paired with a revenue miss and conservative guidance. The stock fell approximately 39% in a single trading session — the worst day in TTD’s history as a public company. Green issued a public apology to investors at the time.

Kayyal was then terminated effective January 24, 2026. He had been CFO for approximately five months. Tahnil Davis, the Chief Accounting Officer and an 11-year company veteran, was named interim CFO. Two CFOs in under twelve months at a company whose brand identity rests substantially on stability and trustworthiness is a self-inflicted wound. The agencies did not cause it. Good governance requires better.

Revenue Deceleration — Context Matters Both Ways

The Trade Desk — Revenue Growth Deceleration, 2025 (Source: TTD Q4 2025 Earnings Release, February 25, 2026)

The deceleration is real and material. Amazon Ads reported 23% year-on-year growth in Q4 2025 — compared to TTD’s 14% — and was reportedly offering DSP fees as low as 1% to major spending clients in direct competition. The broader programmatic market grew; TTD’s relative position weakened. These are real competitive dynamics, not simply the result of agency pressure.

Operational Friction

Digiday’s February 2026 reporting documented operational friction beyond the CFO situation. A media director at one mid-sized agency described cycling through three separate TTD account teams in under a year — three rounds of introductions, three attempts to rebuild working context on a platform complex enough that continuity is operationally critical. A separate executive described an incident where TTD threatened mid-contract rate increases when spending was pacing below an agreed annual target, then backed down when the agency threatened to move spend. Customer retention above 95% over eleven consecutive years means the vast majority of clients do not have these experiences. But the experiences are real, and a platform that claims the transparency high ground has less tolerance for this kind of inconsistency than one that never made that claim.

What TTD Gets Right Despite Everything

Full-year 2025 revenue was $2.9 billion, up 18% from 2024. Q4 2025 revenue was $847 million, up 14% year-on-year. Net income for 2025 was $443 million. Adjusted EBITDA margins were approximately 41%. The company carries no long-term debt and holds approximately $1.5 billion in cash. Customer retention has been above 95% for eleven consecutive years. Between March 2 and 4, 2026, Jeff Green personally purchased 6 million shares of TTD stock at weighted average prices between $23.49 and $25.08, totalling approximately $148 million — the largest insider stock purchase in the company’s history, per insider trading tracker Secform4, confirmed by an SEC Form 4 filing. These are the characteristics of a company with genuine structural strength navigating a difficult transition period.

09

The Allegedly Failed Audit — The Main Event

On March 17 and 18, 2026, Publicis Groupe — the company that had publicly named The Trade Desk its exclusive DSP partner just five years earlier — sent an email to select clients advising them it would no longer recommend TTD as a preferred DSP. The memo was based on an audit conducted by FirmDecisions, a unit of the Ebiquity Group. FirmDecisions is a credible, well-established media compliance auditor that had collaborated with the ANA on the K2 transparency report a decade earlier. Its institutional credibility is not in question.

The audit examined Publicis’s Master Services Agreement with TTD and, according to Publicis’s account of its findings, identified three specific concerns: that TTD had ‘improperly applied their DSP fee to other fees’ charged to Publicis and its clients; that Publicis and its clients had been billed for tools they were automatically opted into without authorisation; and that TTD had not provided the auditor with information necessary to validate that media and data costs were invoiced at cost without mark-up, as the agreement reportedly required.

Publicis stated it had engaged TTD’s senior leadership without reaching a satisfactory resolution. Stifel analyst Mark Kelley confirmed the same day that Publicis represents more than 10% of TTD’s gross billings — making it TTD’s largest holding company client. The stock fell approximately 5.7% on the news. Stifel characterised the move as likely a negotiating tactic, while downgrading the stock to reflect the financial risk.

The Word ‘Allegedly’ Is Doing Real Work Here

TTD ‘failing’ the audit is Publicis’s characterisation of FirmDecisions’ conclusions. The Trade Desk flatly disputes it. No court has ruled. No regulatory body has found a violation. What exists is a contractual dispute between a platform and its largest holding company client, mediated by an auditor whose findings one party accepts and the other specifically rejects. TTD states it has ‘never failed any audit ever’ in its history as a public company. This article uses ‘allegedly’ deliberately and accurately throughout.

The Trade Desk’s denial was specific and direct. In a company statement: ‘Any notion that TTD failed an audit is not true.’ The company argued that the auditor had requested data that would violate customer and partner confidentiality agreements — framing the refusal to provide certain information as a contractual necessity, not concealment.

Jeff Green took to LinkedIn with the broader argument about agency opacity. He is not wrong about the hypocrisy of agencies criticising TTD’s transparency while operating principal-based buying models, collecting rebates, and routing client spend through rival holding companies’ SSPs. He is also not, in addressing those arguments, specifically addressing the audit allegations. Those are two different things. Being more transparent than WPP is factually true, but it is not what resolving a FirmDecisions audit finding looks like.

The specific dispute — fee stacking, auto-opt-ins, refusal to validate cost claims — remains unresolved. TTD denies them. Publicis asserts them. FirmDecisions concluded them. No independent adjudicator has yet ruled. The word ‘failed’ is Publicis’s word, not a settled fact, and should be read as such.

10

Amazon, the Walled Gardens, and the Selective Transparency Standard

There is a third major actor in this story that rarely receives adequate scrutiny: Amazon Ads, which reported 23% year-on-year revenue growth in Q4 2025 while The Trade Desk reported 14%, and which was in 2025 reportedly offering DSP fees as low as 1% for major spending clients in a direct effort to attract agency budgets away from TTD.

Amazon’s structural advantages over any independent DSP are genuine and growing. It owns Prime Video, which launched ad-supported streaming in January 2024. It owns Thursday Night Football. It has the world’s largest e-commerce intent dataset. And it can trace an advertising impression to a purchase on Amazon.com in a way no independent DSP can replicate. At the 2025 Cannes Lions festival, Amazon Ads and Roku announced a partnership giving Amazon DSP exclusive authenticated access to logged-in Roku user data across more than 80 million U.S. households — the largest authenticated CTV footprint in the United States. In September 2025, Netflix announced advertisers could purchase Netflix inventory directly through Amazon DSP.

Green has argued, not entirely without basis, that Amazon’s ad business is approximately 90% Sponsored Listings competing with Google Search, not open-web programmatic display. Several agency buyers confirmed to Digiday in 2025 that Amazon DSP spend was ‘additive’ — from retail media budgets rather than trade desk budgets. This is plausible in the near term. Its durability as Amazon’s CTV ambitions mature is a different question.

The Transparency Standard Applied Selectively

The agencies auditing TTD’s fee structure route enormous budgets to Meta (ad revenue up approximately 21% in 2025), Google, and Amazon — platforms where advertiser transparency into auction mechanics, targeting logic, and fee structures is essentially zero. There are no FirmDecisions audits of Google’s ad exchange. There are no client memos about Meta’s auction opacity. There are no formal reviews of Amazon DSP fees when Amazon also owns the media. The Trade Desk — the one platform that publishes its fee structure and operates no owned media — is the one that received the formal audit. This observation does not resolve the audit findings against TTD. But it describes the environment in which they were commissioned.

11

WPP’s Freefall and the Industry’s Financial Pressures

Understanding why the agencies are fighting so hard right now requires understanding just how severe their financial situation has become. WPP’s story is the most dramatic illustration.

Agency Holding Companies vs. The Trade Desk — March 2026 (Sources: WPP 2025 Preliminary Results Feb 2026; StockAnalysis.com; PitchBook; TTD Q4 Earnings Release)

WPP’s market capitalisation of approximately $3.3 billion as of mid-March 2026 is smaller than the value of the accounts it lost to Publicis alone in 2025. The Mars global media account — approximately $1.7 billion — went to Publicis. The Coca-Cola North America media business — approximately $700 million — also went to Publicis. Its top 25 clients fell 4.1% like-for-like in 2025. WPP Media, its trading arm, declined 5.9% for the full year and 10.8% in Q4 alone. The company was removed from the FTSE 100 in December 2025. New CEO Cindy Rose acknowledged publicly that WPP Media had ‘lost its way.’

The structural pressure across all holdcos comes from the same source: AI is systematically compressing the cost of the functions agencies have historically been paid to perform — media planning, data analysis, creative iteration, audience segmentation. A 2025 Bannerflow report found 83% of senior brand marketers already use AI to target digital ads. Basis Technologies found 92% of advertising agencies use AI in some capacity. The margin that complexity used to justify is under pressure from every direction. WPP launched ‘WPP Open Pro’ in October 2025, a self-service AI marketing product for smaller clients — an acknowledgment in product form that its traditional service model cannot economically serve a large portion of its potential market.

In November 2025, Omnicom completed the acquisition of IPG — transforming the ‘Big Six’ holding companies into effectively a ‘Big Two or Three.’ The new Omnicom includes IPG’s Acxiom data platform, creating first-party data infrastructure competitive with platform data from Google and Meta. Publicis has Marcel AI and Epsilon. WPP has WPP Open. All of them are racing to demonstrate that their AI investments justify the fees they charge. The Trade Desk is watching from a position of financial strength but competitive vulnerability in this race.

12

The Clients Are Finally Noticing

The most striking statistic in the entire industry transparency conversation is from a 2025 World Federation of Advertisers report: 18% of marketers surveyed did not know whether principal-based buying had been part of their media activity in the past year. Not 18% who had concerns about it. 18% who did not know whether it had happened. This is the information asymmetry that makes the practices described in the Foster lawsuit possible at scale.

The ANA’s most recent benchmark found 54% of U.S. brands have an in-house media unit, and 51% have partially moved their programmatic buying function in-house. More than half of agency contracts, per the same WFA report, lacked clear penalties or enforcement mechanisms for principal-buying non-compliance.

Tucker Matheson, co-founder and managing partner of Markacy, told Digiday that his agency had moved spend toward direct buys and other platforms for a reason that is more honest than most: ‘TTD hadn’t done anything egregiously wrong — the alternatives had simply grown up.’ This is a useful corrective to narratives that frame every client departure as a verdict on TTD’s practices. Sometimes the market develops and dominant platforms face more competition. That is a legitimate commercial outcome, not a scandal.

The in-housing trend has genuine limits. Building a real in-house programmatic function requires technical expertise, organisational commitment, and ongoing investment that is not economically viable for most brands. The hybrid model — brand owns the DSP seat and the data relationship, agency manages execution — is the most common outcome. In this model, the agency’s margin on the technology layer disappears. The agency must compete on genuine intellectual value: strategy, creativity, analysis. This is, it turns out, what transparency looks like when it reaches the client relationship. It is not comfortable for business models built on the absence of it.

13

The Verdict — And What Comes Next

The Trade Desk built something that programmatic advertising genuinely needed. In an ecosystem structured around opacity, it built a platform on transparency. In a market where everyone was working both sides, it chose one side and stayed there for fifteen years. It delivered on that promise with sufficient consistency that 95% of its customers renewed, year after year.

The agencies that built their business models on the practices the K2 report documented in 2016 — the practices that led to law enforcement raids in China in 2023, to a $100 million whistleblower lawsuit in 2025, to internal documents describing a financial model whose primary beneficiaries, according to those documents, were not the clients paying for it — these agencies did not suddenly discover transparency concerns in 2026. They discovered that TTD’s growth was making their opacity structurally harder to maintain. And they reached for the language of transparency to address it, from a position whose relationship with that concept is, as the factual record demonstrates, complicated.

Jeff Green is right that the agencies auditing his platform are doing so from glass houses. He is right that an industry that routes vast budgets to Meta, Google, and Amazon without auditing their mechanics is applying transparency standards selectively. These are accurate observations.

Where TTD Must Do Better

The allegedly failed Publicis audit is not fully resolved by pointing at WPP’s court filings. TTD’s refusal to supply certain data to FirmDecisions — however contractually justified — is not what ‘passing an audit’ looks like to the clients observing the situation. The company that has spent fifteen years saying ‘look at how transparent we are’ cannot, at the first genuine audit stress-test of that claim, respond primarily by critiquing the auditor’s clients. The specific allegations — fee stacking, auto-opt-ins without authorisation, refusal to validate costs — deserve a specific, published, detailed response, not just a blanket denial.

Two CFOs in twelve months was a self-inflicted governance failure. The Kokai rollout execution was a genuine operational lapse for which Green publicly apologised. The account team instability reported by Digiday is correctable. These are real issues in a company that holds itself to a higher standard. Holding itself to that standard means addressing them.

The Structural Outlook

TTD’s financial position is stronger than its current narrative would suggest. Full-year 2025 revenue of $2.9 billion, up 18%. Net income $443 million. Adjusted EBITDA margins approximately 41%. $1.5 billion in cash. No long-term debt. Customer retention above 95% for eleven consecutive years. Jeff Green personally investing $148 million in his own company’s stock between March 2 and 4, 2026. These are not the characteristics of a company in existential crisis.

The agency holding companies face a harder structural question. The practices that generated margin in the past are under legal scrutiny through the Foster lawsuit, under client scrutiny through WFA transparency reporting, and under competitive scrutiny because the AI tools compressing their business are also making opacity harder to maintain at scale. Publicis and Omnicom, investing heavily in AI and data infrastructure, may navigate this transition. WPP, at $3.3 billion in market capitalisation with a CEO who has publicly acknowledged the company lost its way, faces an uphill climb.

The clients who fund the entire system have the most to gain from demanding genuine accountability from all sides. The 18% who do not know whether they participated in principal buying last year are subsidising a system that profits from their not knowing. That is an information problem, and the only people who can solve it are the ones writing the cheques.

In programmatic advertising in 2026, everyone is standing in a glass house. The Trade Desk built its out of glass on purpose, because it understood that transparency is not just a principle — it is a competitive advantage, if you are actually committed to it. The question the next twelve months will answer is whether it remains committed to it under pressure. The question the industry needs to answer is whether it will hold all parties — not just the convenient one — to the same standard.

Somebody, at last, is watching. The question is who blinks first.

Read the second part of this article here :

SOURCES — VERIFIED PUBLIC RECORDS ONLY

All facts in this article are drawn from the following verified sources. Every claim traceable to a named, published document, court filing, or analyst note. All allegations from Foster v. WPP are attributed to the complaint. WPP disputes the material characterisation of all practices. No court has ruled. The case is ongoing.

• ANA / K2 Intelligence — Media Transparency Initiative Full Report, June 2016 (public document)

• AdWeek — Exclusive: Dentsu and WPP Quietly Exited The Trade Desk’s OpenPath, February 2026

• AdWeek — Publicis advises clients to stop using The Trade Desk following allegedly failed audit, March 2026

• AdWeek — Publicis-TTD Epsilon Core ID exclusive DSP partnership announcement, 2021

• AdWeek — Publicis Epsilon SSP conflict-of-interest reporting, September 2025

• Ian Colley, CMO, The Trade Desk — LinkedIn post on OpenPath allegations, February 2026 (verified verbatim)

• Jeff Green, CEO, The Trade Desk — LinkedIn post on agency transparency, March 2026 (verified verbatim)

• Jeff Green, CEO, The Trade Desk — AdExchanger interview on agency partnerships, 2016 (verified verbatim)

• Jeff Green, CEO, The Trade Desk — Q4 2025 earnings call quote on competitive complexity, February 2026

• Digiday — In Fighting a Whistleblower Suit, WPP Gave Away the Game, February/March 2026 (WPP motion to dismiss details, Project Claridges, executive descriptions)

• Digiday — The Numbers Behind the WPP Whistleblower Case, March 2026 (97.4%, $13.4B, Google 0.51% stats)

• Digiday — Agency Shopping Around on The Trade Desk, February 2026 (account team friction, rate increase incident)

• Digiday — In-housing programmatic analysis, February 2026 (Tucker Matheson quote)

• B&T Australia — Former WPP Exec Sues Holdco (detailed Foster complaint, executive roles), November 2025

• Campaign Asia-Pacific — GroupM China rebate practices investigation and raid coverage, October 2023

• Campaign US — Former CEO within WPP Media sues WPP (case filing details, verified roles), November 2025

• Storyboard18 — Jeff Green LinkedIn post verbatim analysis, March 2026

• Brewer Attorneys & Counselors — Official press release on Foster v. WPP filing, November 12, 2025

• WPP plc — 2025 Preliminary Results and Strategy Update, February 2026 (revenue figures, account losses)

• WPP plc — Q3 2025 Trading Update, October 2025

• The Trade Desk — Q4 and Full Year 2025 Earnings Release, February 25, 2026 ($2.9B revenue, $847M Q4, net income $443M)

• The Trade Desk — Q1 2026 Guidance ($678M minimum), February 2026

• The Trade Desk — Ventura Ecosystem launch, February 24, 2026

• The Trade Desk — OpenAds launch, October 2025 (first wave of publishing partners)

• The Trade Desk — Official IPO press release, September 20, 2016 (priced at $18.00/share)

• Nasdaq.com — TTD IPO first-day close $30.10, up 67%; September 21, 2016

• MacroTrends — TTD all-time high closing price $139.51 on December 4, 2024; 52-week high $91.45

• TradingView — TTD all-time high $141.53 on December 4, 2024; ATL $2.20 November 2016

• SEC Form 4 filing — Jeff Green insider purchase, 6 million shares, $148M, March 2-4, 2026

• Motley Fool — Jeff Green $148M purchase analysis, March 5, 2026 (confirms $23.49-$25.08 weighted avg)

• Stifel Research — Analyst note: Publicis as >10% of TTD gross billings; likely negotiating tactic, March 2026

• Bank of America — Analyst note citing Kokai execution as factor in Q4 2024 earnings miss

• StockAnalysis.com — WPP market cap $3.28B as of March 18, 2026; TTD stock data

• PitchBook / market data — Omnicom ~$26B market cap; Publicis ~$21B market cap, March 2026

• ANA — In-house media benchmark report, 2023 (54% have in-house unit; 51% partially in-housed programmatic)

• WFA (World Federation of Advertisers) — 2025 report: principal media, audit rights; 18% marketers unaware stat

• W Media Research — Amazon and Roku CTV authenticated partnership, Cannes Lions 2025

• Bannerflow / Basis Technologies — AI adoption statistics in advertising, 2025

• The Times (UK) — Ivan Fernandes quote on WPP court filing commercial significance (reported in Digiday)

• Court filing — Foster v. WPP plc and GroupM Worldwide LLC d/b/a WPP Media, filed November 2025 / December 2025, Supreme Court of New York, New York County / US District Court Southern District of New York

The global advertising industry is not dying; in fact, it is thriving. Global advertising spend is forecast to surpass an unprecedented $1 trillion for the first time in 2026, while the digital advertising sector alone is projected to exceed $350 billion by 2030. AdTech as an industry segment will undeniably remain and continue to expand. However, the rules of the game and perhaps the game itself, are poised for a radical rewrite over the next two years.

We are entering the “AI Acceleration Era” (2026-2028), where Agentic AI is no longer just an experimental tool, but will become the fundamental operating logic of marketing and commerce. As conversational interfaces synthesize information and autonomously guide real-time purchase decisions, entirely new walled gardens are growing by leaps and bounds. Platforms like OpenAI and Anthropic are rapidly replacing traditional search engines as the “front door” to the internet, signaling a structural shift in how brand discovery and economic value are created.

History proves that while entire industry segments rarely disappear overnight, individual companies frequently do when they fail to recognize technological inflection points. As the old guard struggles and new leaders emerge, a complete paradigm shift is not just expected in AdTech, it is strictly necessary for survival.

The Anatomy of Corporate Hubris and Historical Paradigm Shifts

The history of corporate extinction is rarely a story of sudden, unforeseeable catastrophe. It is almost always a story of willful blindness wrapped in the comforting blanket of quarterly earnings. The decline of companies like Kodak, Blockbuster, and BlackBerry is frequently portrayed by business schools as catastrophic, isolated failures. But viewed through the lens of industrial evolution, their demise was a necessary transition, a shedding of obsolete skin that allowed entirely new ecosystems to thrive.

Kodak possessed over 7,000 patents and literally invented the core technology behind the digital camera, yet it filed for bankruptcy in 2012. Why? Because its executive boardroom could not stomach the idea of cannibalizing their highly lucrative, 80% market-share photographic film cash cow. Blockbuster executives famously laughed Netflix out of the room, hopelessly wedded to the predictable, highly margin-accretive cash flow of retail late fees. BlackBerry commanded the enterprise mobile market with an iron fist, only to watch its empire evaporate overnight because its leadership treated the touchscreen app ecosystem as a frivolous consumer toy, failing to anticipate the shift toward app-centric operating systems. These are not just stories of companies that missed the boat; they are stories of massive, entrenched organizations that looked the asteroid dead in the eye and decided they could simply out-earn the impact.

The advertising technology (AdTech) industry, currently navigating the turbulence of 2026, is standing precisely on the precipice of its own evolutionary filter. Driven by a shift toward digital-first marketing strategies, the global online advertising market is projected to surpass $350 billion by 2030. Yet, for the better part of two decades, the digital advertising ecosystem has been a bloated, byzantine machine built on the crumbling foundational physics of third-party cookies, fragmented identity graphs, and a labyrinthine supply chain that extracts a staggering “tech tax” just to move a digital banner from a buyer to a seller. But the asteroid hurtling toward this ecosystem is not merely another privacy regulation out of Brussels or another Apple iOS update. It is the widespread commercialization and deployment of Agentic Artificial Intelligence.

To truly understand the magnitude of this impending disruption, one must look at the historical paradigm shifts that rewired the global economy. The advent of the personal computer decentralized compute power, placing a mainframe on every desk. The mobile era tethered that compute to human geography, birthing the gig economy and the surveillance capitalism model of location tracking. Cloud computing transformed massive capital expenditures into operational expenditures, enabling infinite scale and giving birth to the modern Software-as-a-Service (SaaS) industry.

Today’s AI wave is fundamentally different from the dot-com era, and the AdTech industry misunderstands this at its peril. The current AI infrastructure build-out is not being led by unprofitable startups burning cash on Super Bowl ads. It is being orchestrated by the most profitable technology monopolies in human history, reinvesting billions of dollars of free cash flow into semiconductors, data centers, and power grids. Generative AI requires over 1,000 times the compute of perception AI, and the shift toward Agentic AI—systems capable of autonomous reasoning, context building, and execution—requires an additional 30 to 100-fold increase in compute power.

For AdTech, this means the software paradigm is spontaneously combusting. Since the dawn of programmatic advertising, the industry has relied on isolated SaaS point solutions. There was a tool for dynamic creative, a tool for attribution, a tool for bidding, and a tool for audience segmentation. By early 2026, the State of Martech report charted a “digital Cambrian explosion,” noting that over 1,200 traditional platforms had vanished into the ether. In their place emerged a swarm of AI-native nano-apps and hyper-light agents that flicker into being on command to execute tasks and vanish just as quickly. The center of gravity has violently shifted. Artificial intelligence is no longer a tool within the marketing stack; AI is the marketing stack.

The impending AdTech extinction event will not be evenly distributed. It will obliterate the “long tail” of independent middlemen, decimate the white-collar labor force of global agency holding companies, and force a brutal, zero-sum reckoning for legacy walled gardens that rely on monetizing human search queries. By the end of 2028, the concept of a human media planner logging into a Demand-Side Platform (DSP) to manually tweak cost-per-click (CPC) bids based on deterministic identity graphs will look exactly like a Kodak film development darkroom: an archaic curiosity of a bygone era.

The Macroeconomic Physics of Agentic AI: Efficiency Shocks and the Labor Reset

We are currently transitioning from the “AI Experimentation Era” into the “AI Acceleration Era” (2026-2028), moving rapidly toward the “AI Elevation Era” (2028-2030), where intelligent systems will make autonomous decisions and free human capital to dream up capabilities currently beyond our imagination. To predict the victors and the victims of this timeframe, we must analyze the underlying unit economics of AI compute and inference, as well as the macroeconomic impact of this technology on the global labor force.

In late 2023 and 2024, the narrative was entirely dominated by the sheer, staggering cost of training frontier AI models. Meta’s LLaMA 3 required 16,000 Nvidia H100 GPUs, and OpenAI’s GPT-4 cost an estimated $80 million to $100 million to train. Under those economic conditions, advanced AI would remain the exclusive playground of a few trillion-dollar hyperscalers. However, the release of DeepSeek R1 completely inverted this logic and sent shockwaves through the global tech sector. The Chinese AI startup achieved state-of-the-art reasoning performance by training its model for a mere $6 million using 2,000 older Nvidia H800 GPUs.

DeepSeek achieved this through radical architectural innovations. They utilized a Mixture-of-Experts (MoE) architecture, which activates only 37 billion out of 671 billion parameters for processing each token, drastically reducing computational overhead without sacrificing output quality. Furthermore, their Multi-head Latent Attention (MHLA) mechanism reduced memory usage to between 5% and 13% of previous methods, while optimized distillation techniques allowed reasoning capabilities to be transferred from massive models to smaller, highly efficient ones.

This radical deflation in inference costs is the true catalyst for AdTech disruption. It triggers a massive Jevons Paradox: as the cost of AI inference plummets, the demand and utilization of AI agents explode exponentially. When AI reasoning becomes effectively free to deploy at the edge, the entire premise of “programmatic bidding algorithms” becomes obsolete. We move from an ecosystem where algorithms follow rigid, human-coded rules (e.g., “bid $2.00 if the user is male, 18-34, and visited a checkout page in the last 7 days”) to an ecosystem of autonomous agents negotiating with each other in real-time, operating on semantic understanding rather than deterministic code. Agentic AI turns advertising from a reactive, batch-processed system into continuous, autonomous orchestration.

The macroeconomic fallout of this transition will be severe. Goldman Sachs Research estimates that innovation related to artificial intelligence could displace 6% to 7% of the total US workforce if widely adopted. While they note this impact is likely transitory as new job opportunities emerge, the short-term disruption will be brutal, particularly for white-collar information workers. Bloomberg reports that AI could replace more than 50% of the tasks performed by market research analysts and an astonishing 67% of the tasks executed by sales representatives. In the advertising sector, this translates to a severe “white-collar recession,” where the entry-level roles that historically served as the training ground for the industry—compiling reports, manually pulling data, writing junior copy—are entirely automated out of existence.

However, this transition is not without intense friction. Gartner predicts that by the end of 2027, over 40% of enterprise Agentic AI projects will be canceled due to escalating costs, unclear business value, and fragile governance. The market is currently flooded with “agent washing”—the rebranding of legacy chatbots and basic robotic process automation (RPA) tools as “agents” to appease eager shareholders. The survivors in 2028 will not be the companies that slap an AI chat interface onto a legacy DSP; the survivors will be those that rebuild their infrastructure around unified data intelligence, allowing true agents to see, decide, and act faster than any human team. As Scott Brinker pointed out, the Model Context Protocol (MCP) is emerging as the foundational infrastructure for agentic communication, meaning the new moat is not the software interface, but the underlying data protocol itself.

The Collapse of the Old Gods: Walled Gardens vs. Agentic Search

For the past decade and a half, the digital advertising ecosystem has been a feudal society ruled by an oligopoly of dominant warlords: Google, Meta, and Amazon. These “walled gardens” extracted massive rent from the global economy by monopolizing human attention and hoarding deterministic data. They dictated the rules of engagement, and independent AdTech players were forced to scrape a living off the crumbs that fell outside their walls. But the very architecture of their dominance—the traditional search engine and the scrolling social feed—is under existential threat from a new breed of AI walled gardens like OpenAI and Anthropic.

Generative AI platforms are rapidly becoming the new front doors to the internet. Users are bypassing traditional search engines entirely, instead utilizing conversational interfaces that synthesize vast amounts of information, personalize results, and guide purchase decisions in real-time without the user ever needing to click a blue link. This creates the phenomenon of “Agentic Search,” which is fundamentally hostile to the economic foundations of the ad-funded open web.

When a human searches for a product or an answer, they navigate through web pages, clicking on display ads and generating revenue for the publisher and the ad network. When a Large Language Model (LLM)-based agent searches, it behaves like a swarm of digital locusts. Studies comparing agentic search behavior to human search reveal a terrifying “search explosion.” AI agents generate 10 to 60 times more web traffic than human users for identical queries. An agent might visit 146 distinct web pages to answer a question that a human would have solved with a single click. Yet, this massive traffic generates exactly zero ad revenue, because AI agents do not view display ads, they do not click on affiliate links, and they do not subscribe to paywalls.

This “economic bypass” starves long-tail publishers of oxygen. Without monetization, the open web risks total fragmentation or collapse, as content creators lose all financial incentives to share knowledge. The data ecosystem that trained the very models destroying it will dry up, leaving only vast, enclosed data fortresses.

This dynamic is forcing a brutal realignment among the legacy tech giants. Meta announced a suite of AI-powered advertising tools in early 2025 designed to automate campaign setups and expand AI optimizations. They are desperately trying to prove that their algorithmic “Advantage+” campaigns can drive performance even as underlying signal loss accelerates. Google, fighting a massive multi-front war against antitrust regulators and the existential threat of ChatGPT, is embedding its Gemini models into every facet of its Performance Max (PMax) campaigns. Both Meta and Google are effectively asking advertisers to completely surrender control and blindly trust the machine to find conversions.

Meanwhile, the new AI walled gardens are maneuvering to become the ultimate arbiters of commerce. OpenAI recently appointed Omnicom’s PHD as its global media agency of record, a highly aggressive move signaling a massive push to cement ChatGPT as a ubiquitous consumer brand and transition from in-house marketing to formal, large-scale media buying. Anthropic is aggressively marketing Claude, even launching Super Bowl ads directly targeting OpenAI’s intention to introduce ads into ChatGPT.

The infrastructural decisions made by these AI platforms in 2026 will dictate whether the next era of digital advertising is open and ecosystem-inclusive, or if it simply replicates the walled garden dynamics of the past decade on an even more restrictive scale. If a consumer uses ChatGPT as their primary interface for booking travel, buying insurance, and ordering groceries, OpenAI effectively becomes the ultimate “zero-click” gatekeeper. In this scenario, OpenAI and Anthropic will charge toll fees for agentic access that will make Google’s historic search margins look like pocket change.

The Independent Programmatic Bloodbath: DSPs, SSPs, and the Squeezed Middle

Trapped between the desperate legacy monopolies and the ascendant AI behemoths is the independent AdTech sector. This is a fragile, highly fragmented ecosystem of Supply-Side Platforms (SSPs), Demand-Side Platforms (DSPs), and data brokers that survive by arbitraging the inefficiencies of the open internet. By 2028, this sector will experience a mass extinction event that mirrors the collapse of the over-leveraged telecoms during the dot-com crash.

The fundamental value proposition of a traditional DSP is workflow automation, audience segmentation, and bidding efficiency. But if AI agents become the new integration layer—stitching together workflows across apps, data sources, and cloud warehouses instantly—the traditional software interface becomes entirely irrelevant. As Scott Brinker noted, generative AI and agentic automation are replacing old integration tools, turning AI agents into the new Integration Platform as a Service (iPaaS). Why would a brand pay a 15% to 20% take-rate to a DSP when a highly tuned, locally hosted open-source AI model can autonomously negotiate media buys directly with a publisher’s API?

Let us examine the current titans of the independent space. The Trade Desk (TTD) is widely considered the undisputed king of the independent DSPs. In 2024, TTD reported $2.44 billion in annual revenue, an impressive 26% increase from the previous year, and its Q1 2025 revenue hit $616.02 million, maintaining a 25.4% year-over-year growth rate. Despite maintaining a fortress balance sheet with $1.7 billion in cash and minimal debt, The Trade Desk is inherently vulnerable to the coming paradigm shift. It faces relentless pricing pressure from Amazon’s DSP, and the overarching threat of market saturation as digital advertising matures. More critically, TTD’s entire strategic moat for the privacy era is built around Unified ID 2.0 (UID2). If deterministic identity graphing is rendered obsolete by AI-driven predictive modeling and Agentic search, TTD’s core value proposition evaporates.

The middle tier of independent AdTech is already showing severe signs of strain, divergence, and a desperate reliance on Connected TV (CTV) to mask the terminal decline of standard programmatic display advertising.

Magnite, the largest independent SSP, is clinging to premium video as a life raft. Its Q1 2025 results highlighted strong CTV growth, fueled by massive partnerships with Netflix (now live with programmatic capabilities in the US, EMEA, LATAM, and expanding to APAC) and Disney (expanding integrations with 35 DSPs). Magnite’s integration with the Spotify Ad Exchange (SAX) for omnichannel audio and video is a strategic move to lock up premium, logged-in inventory that AI agents cannot easily spoof. However, the cracks are showing. Magnite’s management recently refused to reaffirm their full-year 2025 expectations, citing tariff-driven economic uncertainty and notable softening in high-risk advertising verticals like auto, retail, and travel.

The broader programmatic display market is effectively becoming a toxic asset class. Display advertising is plagued by infinite synthetic inventory generated by AI content farms, rampant click fraud, and plummeting human attention spans. The “long tail” of independent AdTech companies—StackAdapt, Moloco, and hundreds of localized, niche ad networks—will either be absorbed by larger entities for pennies on the dollar or quietly go dark. Moloco recently released a research study attempting to prove that 88% of mobile app ad spend is irrationally concentrated in Google and Meta, arguing that advertisers could increase financial returns by up to 214% by diversifying into the independent app ecosystem. While their underlying data (highlighting a 25% growth in non-gaming app revenues to $70.5 billion) is accurate, the reality is that brands consolidate spend in walled gardens because of algorithmic simplicity and safety. AI will only accelerate this consolidation; an AI agent optimizing purely for Return on Ad Spend (ROAS) will naturally gravitate toward platforms with the most robust, deterministic feedback loops and the lowest friction, starving the independent web of liquidity.

Even Digital Out-Of-Home (DOOH), which represents a unique physical moat that cannot be ad-blocked, is succumbing to programmatic automation and financial strain. JCDecaux recently expanded its global programmatic DOOH (pDOOH) solution across 30,000 premium digital screens in 35 markets, exclusively utilizing its VIOOH SSP to offer instant, real-time campaign activation globally. Meanwhile, Clear Channel is fighting a brutal war of financial attrition. The company is attempting to grow its Adjusted EBITDA at a 6% to 8% CAGR just to escape a crushing debt load, targeting 7x to 8x net leverage by 2028. While DOOH is immune to digital ad blockers, its programmatic yield relies on the exact same mobile location data signals that are currently being degraded by privacy regulations, making its long-term algorithmic valuation highly susceptible to AI disruption.

The CTV Paradigm Shift: FAST Channels and the OEM Small Language Model Rebellion

Connected TV (CTV) and Free Ad-Supported Streaming TV (FAST) channels were heralded as the ultimate safe harbor for the programmatic industry. But the underlying mechanics of this ecosystem are being hijacked by the hardware manufacturers themselves. TV OEMs like Samsung, LG, and TCL are staging an unprecedented coup: they are no longer content with just selling glass panels; they want to own the entire AI transaction layer.

Smart TV advertising has turned the TV homescreen into the most valuable ad real estate in the world. But the introduction of Agentic AI is fundamentally altering who controls that screen. Samsung is heavily deploying “Vision AI” and integrating Small Language Models (SLMs) directly on-device. Google AI Edge is expanding its Gemma 3 SLMs to run locally on mobile and web hardware. Meanwhile, TCL is aggressively transforming its display interfaces with AI-powered portfolios showcased at CES 2026.

By embedding SLMs directly into the TV’s operating system, OEMs can process user intent locally. When a user asks their TV, “Find me a jacket like the one in this movie and order it,” the on-device SLM negotiates the commerce transaction internally. It never passes that intent data to an external DSP or SSP. The OEM becomes the ultimate gatekeeper, leaving traditional adtech entirely blind to the consumer’s living room behavior. Adtech companies relying purely on CTV bidstreams will find their supply choked off at the hardware level.

Retail Media: Sitting on Data Gold, Tripping Over the Pipes

If traditional programmatic display is dying, Retail Media Networks (RMNs) have been heralded as the undisputed saviors of the industry. Retailers possess the holy grail of modern advertising: closed-loop, deterministic purchase data. They know exactly who bought the diapers, when they bought them, whether they used a coupon, and what flavor of ice cream they bought to cope with the lack of sleep. Consequently, retail media is recognized as the fastest-growing advertising channel globally; it is projected to be a $4.7 billion market in Southeast Asia alone by 2030, and vastly larger in the US and Europe.

However, there is a fundamental structural flaw in this narrative: most retailers are grocers and supply chain experts, not software engineers. They are sitting on a mountain of digital gold but lack the pipelines, the ad servers, and the algorithmic intelligence to extract it efficiently. To bridge this gap, historical holdouts are either forming desperate alliances or pivoting to Agentic AI at breakneck speed.

Walmart is leading the charge, aggressively transforming its massive $4.4 billion ad business from a rudimentary search-bidding platform into a fully autonomous Agentic AI ecosystem. Walmart Connect is shifting from a general retail media arms race into a specific “agentic AI race” aimed squarely at challenging Amazon’s market dominance. Walmart is now placing ads directly within “Sparky,” its AI shopping agent, which 81% of Walmart customers indicate they would use to check product details and availability. Furthermore, Walmart introduced “Marty,” an advertising assistant agent currently in beta for Sponsored Search campaigns that handles the minutiae of billing and bidding, directly challenging Amazon’s AI shopping assistant, Rufus.

As Walmart CEO Doug McMillon presciently stated, “A standard search bar is no longer the fastest path to purchase, rather we must use technology to adapt to customers’ individual preferences”. To execute this, Walmart developed “Wallaby”—a series of proprietary, retail-specific Large Language Models trained on decades of Walmart’s transactional data. This provides Walmart with a contextual data moat that a generic foundational model like ChatGPT cannot easily replicate. Target is following suit, prioritizing AI investments to upgrade their customer-facing assistants and improve employee workflows, though they are still in the early days of realizing top-line sales impacts from these initiatives.

In Australia, the supermarket giant Coles recognized that raw loyalty data is useless without frictionless media activation. They recently folded “Unpacked by Flybuys”—the data division of their massive loyalty program—directly into their retail media business, Coles 360. This transition allows media agencies to directly plan, buy, and measure their investments against Flybuys data, eliminating the severe friction of relying on third-party clean rooms and disjointed DSPs.

In Singapore, NTUC FairPrice Group (FPG) launched FPG ADvantage, an omnichannel network connecting 570 physical touchpoints with 1.7 million app users and two million Link Rewards members. FairPrice is proving that retail media is not just about monetizing digital banners; it is about digitizing the physical store itself. Their “Store of Tomorrow” initiative features AI-powered smart trolleys that guide shoppers through the aisles, display real-time, location-based promotions, and handle automatic checkout using vision AI and eight onboard cameras. The economic results are staggering: the flagship store saw the average basket size explode from $19.64 to $35.36—an 80% surge in shopper spend directly attributable to localized, agent-driven retail media.

Companies like Criteo are desperately trying to pivot to service this retail boom. Once heavily reliant on third-party cookies for cross-site display retargeting, Criteo has entirely repositioned itself as the infrastructural backbone for retail media commerce, citing Gartner research that by 2020, 60% of organizations would use AI for digital commerce. But Criteo faces the classic SaaS dilemma of the 2020s: as AI agents replace core workflows with autonomous execution, the value of the “software in the middle” compresses. If Target or Walmart can spin up hyper-efficient, bespoke agentic infrastructure using open-source models (like DeepSeek or LLaMA), the need for a third-party retail media tech vendor taking a massive percentage of the media spend diminishes to zero.

The Extinction of Dynamic Creative Optimization (DCO)

For over a decade, Dynamic Creative Optimization (DCO) was the shiny, expensive toy of the AdTech world. Companies like Celtra, Flashtalking, and Clinch built businesses around the promise of hyper-personalization at scale. DCO operated on a strictly rules-based, decision-tree paradigm: if the user is in Seattle, and it is raining, and they are a mother, swap the background image of the ad to an umbrella, change the text to “Stay Dry,” and push the ad to the publisher.

This technology, frankly, rarely lived up to the massive hype generated by its sales teams. Marketers routinely complained of doubled ad serving costs, exorbitant audience data fees, disconnected reporting, and a severe degradation of creative quality. As one industry retrospective bluntly noted, “DCO technology doesn’t work, and in-house marketing teams shouldn’t buy it”. It required massive, painful manual effort from design teams to build the thousands of individual assets needed for the permutations.

Generative AI renders traditional DCO completely obsolete. We are moving from the optimization of existing, human-made variants to the real-time generation of bespoke assets. Why build a matrix of 5,000 pre-rendered background images and headlines when an AI agent can dynamically generate a pixel-perfect, contextually relevant image, write the copy, and render the video on the fly in milliseconds based on real-time semantic signals?

Legacy DCO companies are desperately attempting to pivot to “Agentic AI” to survive the slaughter. Clinch recently announced a first-to-market DCO strategy using “Agentic AI” within its Flight Control platform, integrating external media signals from The Trade Desk to optimize dynamic video creative toward post-exposure search intent for brands like Genesis USA. While a valiant effort to stay relevant, they are fighting a losing battle against the hyperscalers.

Massive conglomerates are internalizing creative production. Disney invested $1 billion in Epic Games and partnered with OpenAI, allowing brands to generate high-quality video ads using OpenAI’s Sora model featuring iconic Disney characters. Coca-Cola has made generative tools a central, highly publicized part of its creative operating system, famously producing its holiday campaigns with WPP Open X using AI, signaling a shift toward scalable, user-assisted brand expression.

Furthermore, Adobe is leveraging its massive distribution advantage to embed Generative AI directly into its enterprise suites, allowing organizations to instantly accelerate content output without needing a third-party ad server to piece the creative together. By 2028, the standalone “DCO company” will not exist. It will either be an embedded, free feature of a walled garden’s self-serve platform or a forgotten casualty of the GenAI creative explosion.

The Agency Apocalypse: A White-Collar Recession

If AdTech software platforms face infrastructural collapse, the traditional global advertising agencies face an existential crisis of labor and utility. The business model of the major holding companies—WPP, Omnicom, Publicis, and IPG—has historically relied on the simple arbitrage of human capital. Agencies charge massive corporate clients highly marked-up hourly rates for armies of junior staff to manually build media plans in Excel, execute buys in DSPs, compile weekly PowerPoint reports, and write rudimentary copy.

AI agents do not require salaries, they do not require healthcare, and they do not need sleep. They do not make manual data entry errors on a Friday afternoon. According to the World Economic Forum’s Future of Jobs Report, 40% of employers expect to reduce their workforce where AI can automate tasks, and technology is expected to displace 9 million jobs globally. In the advertising sector, this translates to a severe “white-collar recession”.

The financial markets have already priced in this structural decline, punishing the holding companies severely. WPP, the former undisputed heavyweight champion of the agency world, saw its shares crash by roughly 60% by late 2025, hitting their lowest level since 1998. The company suffered a barrage of high-profile account losses, including the massive Mars global media account. Following a disastrous Q3 2025 where financial performance was publicly deemed “unacceptable,” CEO Mark Read departed, handing the reins to former Microsoft executive Cindy Rose. Rose immediately initiated a ruthless strategic review aimed at aggressively embedding AI and slashing structural inefficiencies, warning that jobs were at risk. WPP’s internal AI platform, Open Intelligence, is a desperate, expensive attempt to prove that the holding company can still add value in a world where brands can just ask an AI agent to buy their media.

Omnicom took an even more aggressive, bloodier path. Following its massive $13 billion acquisition of Interpublic Group (IPG), Omnicom announced the elimination of more than 4,000 roles, primarily targeting administrative and operational functions. The company is brutally consolidating legacy brands that have existed for decades: DDB (founded in 1949) and MullenLowe are being absorbed into TBWA, while the historic FCB (dating back to 1873) is being folded into BBDO. Omnicom’s Chief Technology Officer, Paolo Yuvienco, explicitly stated on an earnings call that this is not just about operational efficiency; it is about building a proprietary network of AI agents embedded across the organization. These agents are simulating focus groups, tweaking campaigns in real-time, and surfacing sales signals without human intervention. As Yuvienco noted, AI is becoming the “creative infrastructure,” and the subtext is impossible to ignore: the more AI becomes infrastructure, the fewer people are needed to build around it.

Omnicom’s media agency, PHD, encapsulates this bizarre pivot perfectly. PHD launched “Ascension,” a generative publishing platform to guide clients through the AI transition, and subsequently won the highly coveted global media AOR account for OpenAI itself. The irony is palpable: the traditional media agency is utilizing the very technology that threatens to destroy its business model to service the creator of that technology.

By 2028, the traditional media agency will be unrecognizable. A modern, AI-native marketing stack orchestrated by 4 to 6 interconnected platforms could run multi-billion-dollar global campaigns with just two or three human operators. The agencies that survive will transform into bespoke software consultancies, focusing on high-level strategy, data governance, and training proprietary AI models for enterprise clients. The era of the “media buyer” arbitraging human misery and spreadsheets is over.

The Funeral for Identity and the Measurement Pivot

Perhaps the most profound shift occurring in the background of the AdTech ecosystem is the total collapse of deterministic identity and the subsequent reinvention of measurement. For a decade, companies like AppsFlyer, Adjust, and Branch built billion-dollar valuations acting as Mobile Measurement Partners (MMPs). They relied heavily on device-level identifiers—Apple’s IDFA and Google’s GAID—to track users precisely across the internet, attributing a specific click on a Facebook ad to a specific in-app purchase.

Privacy legislation (GDPR, CCPA) and platform-level changes (Apple’s App Tracking Transparency and Google’s Privacy Sandbox) have systematically dismantled this surveillance architecture. Signal loss is not a future threat; it is a brutal present reality. McKinsey estimates that signal loss puts up to $10 billion of value at risk in the US market alone. On iOS, the rate of installs with an IDFA dropped to a mere 27%. On Android, the deprecation of GAID in favor of the Privacy Sandbox’s Attribution Reporting API, FLEDGE (for audiences), and TOPICS (for personalization) will finalize the death of device-level tracking. According to the IAB State of Data 2025 report, only 30% of agencies and brands have fully integrated AI into measurement, but those who haven’t are flying completely blind.

If identity is dead, how is advertising measured? The answer is a complete pivot to probabilistic modeling, AI-driven predictive incrementality, and Media Mix Modeling (MMM).

The industry is moving aggressively away from the illusion of “Multi-Touch Attribution” (MTA) toward a “Single Source of Truth” (SSOT) model. This approach completely ignores traditional identity, instead utilizing AI to synthesize highly fragmented data sources—SKAdNetwork postbacks, consenting user data, and aggregate macro data—into a unified view. MMM, once a slow, expensive statistical exercise reserved for massive CPG giants, has been supercharged by AI. Modern MMM uses multi-linear regression analysis on aggregated historical data (incorporating uncontrollable variables like economic conditions, inflation, or competitor activity) to determine the baseline sales versus the incremental sales driven specifically by marketing.

The MMPs are scrambling to adapt or face immediate irrelevance. AppsFlyer, reading the writing on the wall, announced a major pivot, rebranding itself as a “Modern Marketing Cloud” rather than just a mobile attribution provider. In late 2025, AppsFlyer launched a suite of AI-driven products, including “Incrementality for UA” (a causal measurement engine that runs holistic lift tests without manual setup) and “Signal Hub” (a privacy-safe data collaboration clean-room designed for the signal loss era). AppsFlyer understands that if its primary value proposition remains “we hold your deterministic data,” it will die, because enterprise data teams are consolidating data into their own cloud data warehouses (like Databricks or Snowflake).

By 2028, the concept of a “user profile” in the traditional programmatic sense will be a historical relic. AI agents will evaluate the contextual relevance, creative resonance, and historical aggregate performance of a placement in real-time, completely bypassing the need to know the specific identity of the human on the other side of the screen.

Global Regulation and the Geopolitics of AI AdTech

The transformation of AdTech cannot be analyzed in a vacuum; it is inextricably linked to the geopolitical race for AI supremacy and the corresponding regulatory fallout. By early 2026, over 72 countries had proposed more than 1,000 AI-related policy initiatives, creating a fiercely complex global compliance matrix for multinational advertisers and AdTech vendors.

The European Union (EU): The EU continues to act as the global regulatory vanguard, but the weight of its own bureaucracy is causing severe structural fatigue. The landmark EU AI Act faced massive implementation hurdles, leading the European Commission to release a “Digital Omnibus on AI Regulation Proposal” in late 2025. This proposal essentially delayed the enforcement of provisions governing “high-risk AI systems” due to a lack of harmonized standards and competent authorities. While the EU focuses heavily on protecting its citizens from algorithmic harm, its stringent data localization and AI compliance laws make it an increasingly difficult, high-friction theater for independent AdTech to operate profitably.

The United States (US): In stark contrast to Europe, the United States executed a hard, aggressive pivot toward deregulation. Following the change in administration in 2025, President Donald Trump signed executive orders specifically designed to remove barriers to AI development and “unleash prosperity through deregulation”. This included the revocation of the previous administration’s EO 14110 and active efforts by the White House to preempt and eliminate state-level AI laws (such as those in California, Colorado, and Texas). Furthermore, U.S. courts set a critical precedent by ruling that training AI models on copyrighted works likely qualifies as fair use. This deregulatory environment heavily favors the massive American tech conglomerates (OpenAI, Google, Meta), allowing them to aggressively train agents and deploy autonomous AdTech systems with minimal legal friction. The U.S. is also flexing its geopolitical muscle through initiatives like “Stargate UAE,” a massive partnership with OpenAI and the Emirates to build frontier-scale compute capacity.

The Asia-Pacific (APAC) Region: APAC represents a fractured but highly innovative landscape.

1. China: The Chinese government is tightly controlling the output of Generative AI to maintain ideological security. China promulgated strict “AI Labeling Rules,” requiring explicit and implicit labels on all AI-generated content. Despite these restrictions, Chinese firms like DeepSeek are proving that they can vastly out-engineer Western firms in pure cost-efficiency.

2. Australia: Australia updated its Privacy Act to address automated decision-making and released a Voluntary AI Safety Standard containing 10 specific guardrails (including human oversight, rigorous testing, and stakeholder engagement). However, the Productivity Commission deliberately cautioned against heavy-handed mandatory legislation to prevent chilling investment.

3. South Korea & Japan: South Korea finalized its AI Framework Act in 2025, investing heavily in R&D and launching plans for the world’s highest-capacity AI data center. Japan enacted the “light touch” AI Promotion Act and notably amended its Copyright Act to explicitly permit the use of copyrighted works for AI training, positioning itself as a highly attractive safe haven for uninhibited AI development.

4. India & Singapore: India is aggressively pushing the Digital India Act to govern AI-generated content and establish self-regulatory standards for explainability and privacy. Meanwhile, Singapore is acting as the diplomatic nexus of global AI, signing bilateral cooperation agreements on AI safety and governance with the US, Australia, and the EU AI Office.

For the AdTech industry, this fragmented regulatory landscape means that running a truly “global” campaign is becoming technologically and legally perilous. AI agents operating in the US will have vast access to training data and predictive targeting capabilities that would be fundamentally illegal under the EU’s AI Act or Australia’s guardrails. AdTech platforms will be forced to build highly localized, geofenced AI models, vastly increasing compliance and compute costs and further accelerating the death of undercapitalized independent vendors.

Conclusion: The 2028 Horizon

The narrative of catastrophic failure in business is often a misinterpretation of industrial evolution. Kodak, Blockbuster, and BlackBerry did not merely fail because of executive stupidity; they acted as the necessary fertilizer for the subsequent exponential growth of the digital camera, streaming video, and the mobile smartphone ecosystem. Their decline was a brutal but required transition that allowed the broader technology industry to adapt and thrive.

The AdTech industry is currently undergoing this exact evolutionary filtration process. Between 2026 and 2028, the bloated, SaaS-heavy, manual-labor-dependent marketing stack is deflating at an unprecedented rate. It is being replaced by a unified intelligence layer powered by Agentic AI that observes, decides, and executes in real-time without human intervention.

By the end of 2028, the industry will look radically different:

1. The Middleman Extinction: The vast majority of independent DSPs, SSPs, and rules-based DCO providers will be rendered obsolete, outmaneuvered by open-source agentic frameworks that connect buyers directly to premium inventory and retail data without exacting a 20% “tech tax.”

2. The Agency Transformation: Holding companies like WPP and Omnicom will have shed hundreds of thousands of low-level execution jobs. They will survive only if they successfully pivot from media-buying sweatshops into elite AI software consultancies.

3. The Death of Deterministic Identity: Signal loss and privacy legislation have permanently blinded the old cookie-based infrastructure. Measurement is now exclusively the domain of AI-driven Media Mix Modeling and predictive causal engines. MMPs like AppsFlyer will survive only by becoming holistic data collaboration hubs rather than device-ID identity brokers.

4. The Rise of the Agentic Walled Gardens: OpenAI, Anthropic, and the architects of “Agentic Search” will become the new gatekeepers of human commerce, posing an existential threat to the ad-funded open web and traditional publishers.

The companies that survive this paradigm shift will not be those that attempt to bolt a shiny AI chatbot onto a legacy programmatic platform. The survivors will be those that realize the fundamental physics of the industry have changed forever. Identity is an illusion. Manual optimization is a relic. Agentic AI is the operating logic of the future. The asteroid has already hit the atmosphere; most of the dinosaurs just haven’t looked up yet.

The Epilogue to the Rainy Tuesday: 2027 Edition

Let’s fast forward to that same rainy Tuesday in November 2027. The bus shelter hasn’t moved, but the brain inside it has been lobotomized and replaced.

The rain starts falling. The crowd gathers. This time, the screen doesn’t play a perfume ad. It doesn’t play a loop. The screen goes dark for 20 milliseconds—a “Headless” reset triggered by an edge signal. Then, a bright, high-contrast creative appears: “Uber Surge Pricing Waived for Next 15 Mins from this Stop. Scan to Unlock.”

The crowd doesn’t ignore it. They attack it. QR codes are scanned. Rides are booked. The Advertiser (Uber) just paid $60 CPM and smiled about it because their Return on Ad Spend (ROAS) was calculated in real-time. The Media Owner just made more in 15 minutes than they usually make in a day.

The Consumer got home dry.

In my last article, Why DOOH Brought a Knife to the CES 2026 Gunfight: The 12-Month Roadmap to a ‘Minority Report’ Reality for CES 2027. I argued that our industry is failing because we rely on static loops and spreadsheet math. And in this linkedin post about the same, many senior leaders of DOOH and adtech industry commented about loops, measurement and attribution issues in DOOH

Now that CES 2026 is over with all agentic AI discussions and along with the response and messages in context of my previous post, I was itching to write more on this topic and it also seemed like everyone was asking me the same question: “Okay, smart guy. You told us we’re holding a knife. How do we actually build the gun?”

And the first thing that I want to say is: We don’t need more “digital transformation” fluff. We don’t need another panel on “The Future of OOH” where four people agree that “data is important.” We need a systems architecture.

We need to stop building Video Players, CMS Connectors, Creative Management Tools etc and start building Decision Engines.

As Jensen Huang, CEO of NVIDIA, (paraphrasing) said in his keynote just three days ago at CES 2026:

“We have moved past the era of ‘smart’ devices. We are now in the era of ‘Agentic’ devices. If your hardware cannot negotiate its own value in real-time, it is already obsolete.”

Your digital billboard is obsolete. It’s a dumb terminal waiting for orders. Here is the 5-step technical roadmap to lobotomize that dumb screen and give it an Agentic Brain by CES 2027.

Introducing Ad Pod in DOOH: Move Ad Decisioning to the Edge

The Paradigm: From “Winner Takes All” to “Unified Yield.”

The biggest lie in Programmatic DOOH is the “Unified Auction.” In reality, most Media Owners run a “Waterfall.” They check their Direct Sold schedule, then check one SSP, then another. This is inefficient and technically flawed.

The solution is the OpenRTB 2.6 Structured Ad Pod.

The Technical Architecture (Multi-DSP Scenario)

In Connected TV (CTV), a “Pod” is a commercial break. In DOOH, we must treat every availability window (or that ill-conceived “loop”) as a Pod.

The Workflow:

1. The Trigger: The screen signals it has a 60-second avail.

2. The Request: The Supply Side Platform (SSP) sends out a Pod Request to 5 different DSPs (The Trade Desk, DV360, Yahoo, etc.).

3. The Response:

   • DSP A bids $20 for a 15s slot (Nike).

   • DSP B bids $35 for a 10s slot (Uber).

   • DSP C bids $5 for a 30s slot (Local Lawyer).

4. The Stitching (The Edge Logic):
   Here is where the magic happens. The Media Owner’s Ad Server (running on the Edge, not the Cloud) receives these bids. It does not just pick the highest price. It runs a Knapsack Algorithm to maximize Yield per Second.

   • Option 1: Play DSP C (30s @ $5). Total: $5.

   • Option 2: Play DSP A + DSP B (25s @ $55). Total: $55.

The Transparency Gain: Currently, DSPs blindly buy slots and pray they don’t run next to a competitor. In the Ad Pod model, the Media Owner’s Ad Server constructs the pod and sends back a “Pod Composition” signal.

• To DSP A: “You won Slot 1. Slot 2 is Uber. Slot 3 is Weather.”

• To DSP B: “You won Slot 2. Slot 1 was Nike.”

This transparency allows DSPs to bid more confidently. A luxury brand might bid 2x more if they know (cryptographically guaranteed) that the ad playing before them is High-End Fashion and not a Personal Injury Lawyer.

The Playback Shift: Why SSAI is the Only Way Forward

The Paradigm: From “File Download” to “Live Stream.”

Currently, 90% of DOOH failures happen because we ask a $200 Android box to do a $2,000 job. We ask it to download heavy creative files (4K ProRes), manage schedules, parse VAST tags, and handle proof-of-play logs locally. When it runs out of memory, you get a black screen (or worse, the Windows desktop).

Anthony Wood, CEO of Roku, proved this model in the living room years ago:

“The TV operating system is the new battleground. It’s not about the screen; it’s about the software delivering the stream.”

The 2027 Fix: Server-Side Ad Insertion (SSAI)

We must adopt the “Headless” model used by Netflix and Roku.

The Technical Stack

1. The Device is Dumb: The media player on the street is stripped of all logic. It does not “play files.” It opens a single HLS (HTTP Live Streaming) or DASH stream from the cloud. It is a dumb rendering pipe.

2. The Edge is Smart: The decision logic moves to Edge Compute nodes (think Fastly or AWS Wavelength). The cloud stitches the content (weather, news) and the ads into a linear stream in real-time using a Just-in-Time (JIT) Transcoder.

3. The Seamless Stitch: The HLS manifest is updated dynamically. The player just sees a continuous stream of video segments (.ts files).

The Business Case for SSAI

• Cost Saving (CapEx): You no longer need expensive Intel NUCs or high-end media players. You can run a 4K stream on a $35 Raspberry Pi or a Tizen chip embedded in the screen. You save millions in hardware refreshes.

• Revenue Increase (Latency): An Android box downloading a file takes 30-60 seconds to “ingest” a new ad. An SSAI stream can inject a new ad in 200 milliseconds. This unlocks “Real-Time Triggers” (like the Uber rain example) that are impossible with file-based systems.

The Attention Economy: CPAS (Cost Per Attention Second)

The Paradigm: From “Opportunity to See (OTS)” to “Cost Per Attention Second (CPAS).”

This is the most critical section. For decades, DOOH has relied on “fuzzy math”—selling impressions based on historical footfall averages. “We think 500 people passed by, so pay us for 500.”

To move budgets from Meta and Amazon, we must move to Deterministic Science. We need to build a measurement stack that functions like a pharmaceutical trial.

The New Metric: CPAS

We stop counting “bodies.” We start counting Gaze. The industry needs to transition from selling the box to selling the time spent inside the box.

The Tech Stack (Sensor Fusion):
We deploy a sensor suite on the screen, combining Computer Vision (CV) and Depth Sensors (LiDAR).

• LiDAR: Creates a privacy-safe 3D point cloud of the crowd to count bodies and distance.

• Computer Vision: Analyzes the Gaze Vector. It calculates the angle of the head and eyes relative to the screen.

The Financial Algorithm:

The Business Model Shift:

We transition from CPM (Cost Per Thousand) to CPAS (Cost Per Attention Second).

• Scenario A (The Highway Billboard): 1,000 cars pass. 50 drivers glance. Average gaze is 0.5 seconds.

  • Old Model: You charge for 1,000 impressions. (Fraudulent).

  • New Model: You charge for 25 Attention Seconds. (Low Value).

• Scenario B (The Elevator Screen): 10 people ride. 8 people stare. Average gaze is 15 seconds (captive audience).

  • Old Model: You charge for 10 impressions. (Undervalued).

  • New Model: You charge for 120 Attention Seconds. (High Value).

This flips the economics. It allows DOOH to compete with TikTok, which also sells attention. If you are a CFO, wake up. We are giving away our most valuable asset—Sustained Attention—for free, bundled inside a generic “Impression.”

The Trust Shift: Cryptographic Proof of Play

The Paradigm: From “Log Files” to “Digital Signatures.”

The biggest blocker to DOOH spend is trust. Did the ad actually play? Is the screen actually there? Did the player crash halfway through?

Currently, we send “Proof of Play” (PoP) logs. These are text files. They can be edited. They can be faked. They are worthless.

The Fix: Device Attestation & TPM

We need to borrow security protocols from the Fintech world.

The Technical Workflow

1. The Hardware Root of Trust: Every media player must have a Trusted Platform Module (TPM) or a Secure Enclave (like in your iPhone). This chip contains a private cryptographic key that cannot be extracted.

2. The Signed Heartbeat: Every 5 seconds, the player generates a “Heartbeat” signal containing:

   • Timestamp

   • GPS Coordinates

   • Hash of the Frame currently on screen

3. The Signature: The TPM signs this heartbeat with its private key.

4. The Verification: The Advertiser receives this signed packet. They use the Public Key to verify it.

   • If the signature matches: It is mathematically impossible for the log to be faked. The ad must have played on that specific device.

Brian O’Kelley, CEO of Scope3, puts it bluntly regarding supply chain transparency:

“Waste is a design flaw. If we build systems that require redundant processing and inefficient delivery, we are just burning money and carbon.”

Fraud is waste. Cryptography kills fraud. By 2027, if your PoP isn’t cryptographically signed, you aren’t getting paid.

The Attribution Shift: Closed Loop & Clean Rooms

The Paradigm: From “Fuzzy Attribution” to “Synthetic Control Groups.”

This is the endgame. The previous steps ensure the ad played and was seen. This step ensures it worked.

How do we know if the viewer bought the coffee? We bridge the gap between the Screen and the POS without tracking the user.

The Tech: Snowflake, InfoSum, or LiveRamp.

The Workflow:

1. Publisher: Uploads “Exposed Environment” data (Time + Location + Device IDs hashed via Wi-Fi sniffers) to the Clean Room.

2. Retailer (e.g., Walmart Connect): Uploads Point-of-Sale transaction data to the same Clean Room.

3. The Match: The Clean Room runs an encrypted intersection. No PII (Personally Identifiable Information) ever leaves the room.

The Proof (Synthetic Control Groups):

We don’t need to track you. We need to track the difference.

• Methodology: The system automatically builds a “Synthetic Control Group”—a virtual audience that is statistically identical to the exposed audience but was not near the screen.

• The Result: “Viewers of Screen A had a 14% Incremental Lift in purchase probability compared to the Synthetic Control.”

Kelly Gerrard, Director at Marshall (speaking on Retail Media in 2026):

“Some brands use ‘new-to-brand’ as a proxy for incrementality, but it’s not enough. By the end of 2026, if you can’t prove incremental lift via a clean room, you aren’t on the media plan.”

This turns DOOH from a “Brand Awareness” channel into a “Point of Purchase” Performance Channel. We are literally the last screen a consumer sees before they buy. If we can’t attribute that sale via a Clean Room, we deserve to go out of business.

Closure: The Builder’s Manifesto

Let’s go back to that rainy Tuesday in November 2027.

The screen goes dark. The Agent negotiates. The Cloud stitches. The TPM signs. The Clean Room verifies. The screen lights up: “Uber Surge Pricing Waived. Scan Now.”

The Consumer gets a ride. The Advertiser gets a sale. The Media Owner gets rich.

We are standing at a bifurcation point. One path leads to us becoming “Landlords”—renting out glowing rectangles to whoever will pay the rent, slowly losing relevance to Retail Media Networks and Mobile. The other path leads to us becoming “Broadcasters”—owners of the most powerful, privacy-safe, high-attention medium on the planet.

This DOOH future won’t be built by people who are good at real estate or running traditional trading and arbitrage models. It will be built by engineers who understand Edge Compute, HLS Stitching, Clean Rooms, and Cryptographic Attestation.

We have the screens. We have the data. The “Gun” is just a stack of code waiting to be written. Who is ready to build it?

The Pre-CES Hype vs. The Rainy Tuesday Reality

Next week, Las Vegas will turn into a neon shrine for CES 2026. You’re going to see “Agentic AI” that can order groceries for you, holographic displays that float in mid-air, and transparent OLEDs that look like something out of Minority Report.

The press releases will scream about “The Future of Connection.” And then, you will fly home, stand at a digital bus shelter in the rain, and watch a static JPEG of a sunscreen bottle rotate on a 60-second loop.

This is the schizophrenia of our industry. We have Ferrari hardware running on Flintstone software.

Picture this. It’s a Tuesday evening in early December. A sudden, nasty downpour hits the city. At a digital bus shelter, twenty people are huddled together, wet, miserable, and furiously checking their phones for delayed bus times. They are cold. They are stuck. They are arguably the most captive audience on the planet at that precise moment.

On the 75-inch, 4K, 5G-enabled screen next to them, what plays?

A luxury car ad. Then a perfume ad featuring a model frolicking in a sunlit field (mocking them). Then a static poster saying “Visit Phuket this December.”

The people ignore the screen completely. Actually, they don’t just ignore it; they resent it. It’s just light pollution laughing at their misery.

To a normal person, this is just annoying. To me, as a technologist who has spent 25 years building the plumbing for AdTech and streaming, it is a crime scene.

We are witnessing the destruction of value in real-time. That screen had twenty high-intent consumers standing inches away from it. But because the industry is running on logic from 1998, we showed them irrelevant noise.

If that screen were an Uber, the price would have surged 3x because demand spiked. If that screen were a TikTok feed, the content would have read the room and adapted. But because it is Digital Out-of-Home (DOOH), it played a pre-scheduled loop from a playlist uploaded last Tuesday by Dave in Accounting.

This disconnect is what I call the “Digital Billboard Dilemma.” In my previous article, “Why We Are Still Using Floppy Disks in a 5G World,” I ranted about the absurdity of manual content updates. But recently, a piece by Premesh Purayil “Introduction to OOH/DOOH from a Web Publisher’s View” struck a nerve. Premesh correctly identifies that OOH is a “one-to-many” medium trying to survive in a “one-to-one” digital world.

Premesh is right. But let’s be honest: We are using “one-to-many” as an excuse to be lazy.

We are selling Occupancy (filling the slot) when we should be selling Yield (maximizing the value of the moment). We are treating billion-dollar digital networks like patches of drywall.

Here is the brutal reality of why our current model is broken, and exactly how we fix it before the retail giants eat our lunch.

The “Uber Test”: Why Static Multipliers are Financial Suicide

In the ride-sharing world, price is a function of demand. Uber doesn’t ask, “What is the average price of a ride on a Tuesday?” It asks, “How many people need a ride right now?”

In DOOH, we fail this test every single day.

Currently, most “Programmatic” DOOH transactions rely on a Static Impression Multiplier. This is a hard-coded number in the bid request that essentially says, “On average, 8 people see this screen at this hour.”

This is financial suicide. When that rainstorm hit the bus shelter, the audience didn’t just double; the value of that audience tripled. These aren’t just passersby anymore; they are a captive audience stuck in a specific location with a specific need (dryness, comfort, transport updates).

But here is the sheer stupidity of our so-called “Tech” industry: The screen doesn’t know it’s raining. The media player is just a dumb box blindly following a playlist. We pat ourselves on the back for being “Digital,” yet operationally, we are functioning exactly like a paper poster. The only difference is that a paper poster doesn’t crash and reboot in front of a client. We have deployed 5G connections, AI cameras, and edge processors, only to use them to display a static JPEG based on a contract signed three months ago. That isn’t “Technology”; that is just expensive electricity.

The Fix: Dynamic Bid Enrichment

To stop being “Paper with a Plug,” we need to move from Static Averages to Real-Time Census.

• The Tech: Modern screens have cameras or LiDAR sensors (like Quividi or AdMobilize). They count the crowd size in real-time.

• The Mechanism: Instead of a static config file, the media player must inject the live count into the OpenRTB bid request.

  • Old Way: {“impmul”: 5} (Based on historical data from last year)

  • New Way: {“impmul”: 22, “situation”: “crowd_surge”, “weather”: “heavy_rain”}

• The Financial Upside: The SSP instantly raises the Bid Floor. If the CPM is $10, the surge price is $30. Buyers (like Uber or a coffee chain) will happily pay it because they aren’t buying “exposure”; they are buying “verified attention.”

As Jeff Green, CEO of The Trade Desk, famously said:

“Walled gardens aren’t incentivized to make the open web effective. They are incentivized to grade their own homework.”

In DOOH, we are grading our own homework by using static multipliers. We tell buyers “8 people saw this” because the spreadsheet says so. Real-time data forces us to be honest—but it also allows us to charge for the surge.

The Structural Failure: The “Dumb Loop” vs. The “Smart Pod”

So why haven’t we fixed the pricing? Because our plumbing is broken. We are addicted to the Loop.

In the physical world, a landlord wants 100% occupancy. If you have 10 apartments, you want 10 tenants. In DOOH, this translates to the “Loop”—usually a 60-second rotation with six 10-second slots.

If a Media Owner sells 5 slots, they panic. They fill the 6th slot with a “House Ad” or a generic weather widget just to keep the loop full.

• The Business Failure: This signals Inventory Deflation. You are telling the market, “My screen is always available, and I have no better use for this time.”

• The Opportunity Cost: You are diluting the value of your paying advertisers by sandwiching them next to free filler.

Compare this to Connected TV (CTV). In the streaming world, no one buys a “loop.” In fact, no one even buys a whole “pod” (the commercial break). They buy a specific opportunity within that pod, governed by complex, real-time logic.

• The CTV Ad Pod (The Yield Model):

  • Competitive Separation: The ad server knows not to play a BMW ad right after a Mercedes ad. It protects the brand’s value.

  • Frequency Capping: It knows I’ve already seen that insurance ad 4 times this hour, so it swaps it for a CPG brand to avoid ad fatigue.

  • Yield-Per-Second: The first slot in the pod might cost $30 CPM, while the last slot costs $15. The pod is deconstructed to maximize revenue.

• The DOOH Loop (The Real Estate Model):

  • Zero Separation: I see a McDonald’s ad followed immediately by a Burger King ad because two different DSPs bought “Slot 3” and “Slot 4” blindly. The value of both ads is destroyed instantly.

  • Rigidity: If demand drops, we still play the full 60-second loop with filler. If demand spikes (like the rain example), we can’t extend the loop or insert a premium “break.”

The Fix: Kill the Loop.

We need to adopt Dynamic Ad Pods. If you only have 3 paying ads, play 3 ads and then switch to high-value content (news/sports) that actually engages viewers, rather than “filler.” If demand surges, expand the pod. Treat time as a fluid asset, not a rigid bucket.

As Reed Hastings, Co-Founder of Netflix, understood early on:

“Stone Age. Bronze Age. Iron Age. We define our history by the technology we use.”

The Loop is the Stone Age of AdTech. It’s a blunt instrument in a precision world.

The “Content Carrot”: If You Don’t Have a Show, You Must Be a Service

In CTV, the viewer tolerates the ad because they are waiting for The Office or Yellowstone to come back on. The content is the “Carrot.” They hate the ad, but they love the show, so they stay.

In DOOH, there is no carrot. The ad is the content.

Unless you are in Times Square, Piccadilly Circus, or Shibuya Crossing—places I’ve visited where tourists actually stand still to film the 3D whales—no one is “tuning in” to your screen. To a commuter rushing to a train, a generic brand video is invisible. It is “banner blindness” at physical scale.

The Fix: Contextual Utility (Service-as-an-Ad)

If you can’t entertain them, you must help them. You have to earn the right to be looked at.

• The Mechanism: We need to shift from Video Files to HTML5 Templates. Instead of the DSP sending a flat .mp4, it sends a dynamic template.

• The Logic: The media player (or Edge Server) populates the template fields based on real-time triggers.

  • Trigger: Rain > 5mm/hr → Swap Creative A (Iced Coffee) for Creative B (Hot Chocolate).

  • Trigger: Transit Delay > 10m → Insert a live “Wait Time” ticker inside the ad creative.

If your ad doesn’t react to the environment, it is dead pixels.

Marc Pritchard, CBO of P&G, has been screaming about this for years:

“We have a media supply chain that is murky at best and fraudulent at worst. We need to clean it up.”

Showing a sunscreen ad during a rainstorm isn’t fraud, but it is incompetence. It’s a waste. And in 2026, waste is the one thing no CMO will pay for.

The Business Landscape: Evolve or Be Eaten

The market is consolidating, and the “Landlords” (Media Owners who just rent space) are running out of time.

While traditional OOH companies fight over billboards, Retail Media Networks (RMNs) are eating the world. Walmart Connect, Uber Advertising, and Chase Media Solutions are growing at 20%+ while the general ad market crawls at 4%.

Why? Because they own the Loop and the Data.

• Traditional DOOH says: “We think 500 people saw this ad.”

• Retail Media says: “We know 50 people bought the product after seeing this ad.”

The Consolidation Signal

Look at the recent acquisition of Place Exchange by Broadsign. This is the canary in the coal mine. The pipes are merging. The industry is realizing that you cannot have separate silos for “CMS” (Loop management) and “SSP” (Programmatic sales).

The winners of 2026 will be vertically integrated platforms that control the glass and the transaction. If you are a small media owner using a generic CMS and a third-party SSP, your margins are about to get squeezed by the tech giants who can offer “Programmatic Guaranteed” efficiency at scale.

Brutal Reality for CFOs: Stop looking at Occupancy Rate. Occupancy incentivizes you to fill the loop with cheap filler just to say “we are full.” Start looking at Revenue Per Available Second (RPAS). It is better to have a black screen for 50 seconds and play one $100 ad for 10 seconds, than to play six $5 ads just to “fill the loop.”

Brian O’Kelley, CEO of Scope3, puts it bluntly regarding carbon and waste:

“Programmatic advertising creates massive amounts of carbon emissions… mostly from ads that no one sees or that don’t drive value.”

Every time we download a video file to a player that never plays it, or play an ad to an empty room because the “loop” said so, we are burning cash and carbon.

The Enabler: Why SSAI is the “Plumbing” of the Future

None of the above (Dynamic Multipliers, Ad Pods, Contextual DCO) works if we rely on the current hardware-heavy approach. We cannot trust a $200 Android media player to handle complex logic, API calls, and real-time bidding without crashing.

But we already solved this in Streaming.

I spent years building Server-Side Ad Insertion (SSAI) systems for the US streaming market. We realized early on that smart TVs were actually pretty dumb. So we moved the brain to the cloud.

• The Cloud constructs the stream: It checks the weather, checks the crowd size, calls the Transit API, and stitches the perfect ad into the video stream.

• The Screen is dumb: It just plays the video stream it receives. No logic. No crashing. No black screens.

SSAI allows DOOH to function like a FAST Channel (Free Ad-Supported TV). It is the only way to scale “Smart” ads without replacing millions of hardware units on the street. It turns your billboard into a linear TV channel that just happens to be on a sidewalk.

As Anthony Wood, CEO of Roku, proved:

“The future of TV is streaming.”

Well, the future of Billboards is streaming too. We just haven’t admitted it yet.

Closure: The CES Challenge

Next week at CES 2026, you will see a lot of dazzling screens. You will see an 16K resolution. You will see transparent glass. You will see “AI Agents” that promise to run the world.

Don’t be distracted by the shiny objects.

Ask the hard question: “Is this screen smart enough to know I’m standing here, or is it just playing a loop?”

Because if we don’t fix the plumbing underneath these beautiful screens, we aren’t building the future. We are just building the world’s most expensive digital wallpaper.

Let’s replay that Tuesday evening with a new stack in place.

The rain starts.

1. The Sensor detects the crowd surge (20 people) and the weather (Rain).

2. The SSP updates the bid request: multiplier: 20, context: rain.

3. The Yield Engine (killing the loop) triggers a surge price. A ride-share app wins the bid at $45 CPM (up from $15).

4. The DCO Engine builds a creative on the fly: “Don’t stand in the rain. Your ride is 2 mins away. Use code RAIN20 for 20% off.”

5. The Outcome: The audience looks up. They feel understood. They feel helped. Three people book a ride immediately.

The Media Owner made 3x revenue.
The Advertiser got 3x conversions.
The Consumer got a solution, not pollution.
The bus is leaving. Make sure you’re on it.

Let me give you some context before I start ranting. I’ve been in the trenches of media technology for 25 years. I was writing code for Set-Top Boxes (STBs) when “on-demand” meant walking to Blockbuster. I built the architecture for the first DVRs that let you pause live TV. I have worked on engineering the streaming pipes that allow you to binge-watch in 4K without buffering and to figure out how to connect a second screen device to a Satellite Pay TV system.

I know the video. I know data. I know how to move pixels from a server to a screen efficiently.

About a year ago, I moved into Digital Out-of-Home (DOOH) advertising. I thought, “Great! Big screens, programmed logically, just like CTV but outdoors.”

Boy, was I wrong.

Walking into DOOH after working in advanced streaming is like stepping out of an autonomous car and starting driving a stick again. It’s charming, but my god, is it too much effort? Hell Yes!! We are an industry worth nearly $40 billion, yet we are operating with tech habits that might get you in trouble at MAG7 in an afternoon.

Grab a coffee. Let’s talk about why this industry is broken, why we need to be truthful to ourselves, and how—if we actually use some common sense—we might fix it.

1. Landlords with HDMI Ports (The Media Owner Mindset)

Here is the first thing I learned: Media Owners are not tech companies. They are Real Estate companies.

I’ve sat in meetings with owners who control prime inventory in major capitals. When I start talking about “yield maximization” or “latency reduction,” their eyes glaze over. They don’t care about the tech stack; they care about the pole.

They operate with a landlord mindset: “I own this square footage. You pay me rent.”
They prefer to sell monthly packages (or “loops”) because it guarantees revenue. If I tell them, “Hey, if we connect this to an exchange, we can dynamically price your prime hours at 5x the rate,” they look terrified. They prefer a flat $5,000 check from a local divorce lawyer over the potential to make $10,000 from Coca-Cola.

They don’t understand yield. They understand occupancy. It’s the equivalent of a hotel selling every room for $50 regardless of whether it’s a Tuesday in November or New Year’s Eve.

2. The “Programmatic” Lie

If I hear the word “Programmatic” one more time, I’m going to unplug a server.

The industry loves to buzz about how we are “just like the web.” We aren’t. Terence Kawaja, the founder of LUMA Partners (the guy who makes those insane ad tech maps), famously said:

“Programmatic is not a channel, it’s a method of buying.”

Well, in DOOH, our “method” is basically a digital fax machine. Most of what we call “Programmatic” is actually Programmatic Guaranteed (PG).

• Real Programmatic (RTB): An auction happens in 200 milliseconds. The highest bidder wins based on real-time data.

• DOOH Programmatic (PG): Two guys agree on a price over a beer. We input the deal ID into the system. The system delivers the ad exactly as planned.

We built Ferrari engines (DSPs/SSPs) to drive to the grocery store at 10 mph. We want the “cool factor” of saying we are automated, but the control freak nature of the industry refuses to let the algorithms actually do the work.

3. The 30% “Ghost” Inventory

Here is some market research you won’t find in the glossy brochures. Based on what I’ve seen in emerging markets and even parts of Europe, I estimate up to 30% of screens are operating in a legal grey area.

No license. No council permit. Just a guy who bought a cheap LED panel from Shenzhen, bolted it to a wall, and is stealing electricity from the building next door.

When you buy “Run of Network” programmatically, you might think you’re getting premium mall screens. In reality, your brand might be running on a hacked TV screen above a questionable massage parlor. This lack of transparency is why big brands are hesitant. As Brian O’Kelley, CEO of Scope3, points out constantly, the supply chain is full of waste and opacity. In DOOH, the waste isn’t just digital; it’s physical junk inventory clogging the pipes.

4. The 3% Budget & The “Click” Fetish

Advertisers are weird. They will drop $5 million on a TV spot because it “feels right,” but for DOOH? We get the scraps. Usually 3% to 5% of the media budget.

Why? Because digital marketers are addicted to Click-Through Rates (CTR).

I literally had a meeting where a 24-year-old planner asked, “How do we track the click-throughs on that highway billboard?”
I had to take a deep breath. “It is a 40-foot piece of steel. You cannot click it. If you try, you will die.”
But because they can’t measure it easily in Google Analytics, they don’t trust it. They qualitatively know it works, but quantitatively, they prefer the “safety” of social media metrics, even though half those clicks are bots.

The Fix: Mobile Data & Probabilistic Math

(Stop Trusting the “Multipliers”)

This is the biggest scam in the industry that nobody talks about: The Multiplier.
A media owner tells you: “100 people see this screen every hour.”
How do they know? Did they count? No. They guessed. They put a “100x” multiplier on the impression count in the bid request.
The Tech Solution: Mobile Ad IDs (MAIDs) & SDKs.
We need to stop asking the media owner how many people are there and start asking the Data.

1. The Collection: We use third-party data aggregators (like Unacast, Foursquare, etc.) who collect anonymized location data from SDKs embedded in thousands of mobile apps.

2. The Vicinity Match: We draw a geofence (a virtual box) around the screen’s lat/long coordinates.

3. The Time-Stamp Match: We check: “Which Mobile Ad IDs (MAIDs) were inside that box at 10:00 AM when the ad played?”

4. The Probabilistic Reach Algorithm:

   • We know we only capture maybe 5-10% of people (not everyone has location on).

   • We use Probabilistic Modeling to extrapolate. If we see 10 phones, and we know the density model of that area, we can statistically prove that represents 100 actual humans.

This kills the media owner’s inflated numbers. It gives us Independent Verification of reach, frequency, and attribution without relying on a landlord’s best guess.

TECHNICAL DEEP DIVE: The Real Lesson from CTV SSAI

(Making the Player Dumb and the Server Smart)

Okay, let’s get technical. I previously ranted about SSAI (Server-Side Ad Insertion) being “streaming,” but that’s an oversimplification. The real lesson we need to take from CTV isn’t about live streaming video; it’s about Manifest Manipulation.

In the old world (and current DOOH), the Media Player is “smart.” It has complex logic, it talks to ad servers, it tries to make decisions. Smart players are bad. Smart players crash. Smart players require expensive hardware.

The CTV SSAI Model applied to DOOH:
We need to move the brain to the cloud.

1. Server-Side Decisioning (The Brain):
   The server holds the logic. It knows the campaign rules, the frequency caps, the exclusions. It generates the Playlist (Manifest).

   • CTV Parallel: Just like a CTV server builds a .m3u8 playlist for a viewer, the DOOH server builds a precise playlist for the screen.

2. Manifest Manipulation:
   The server modifies this playlist dynamically. It inserts the ad segments into the content schedule.

   • The key difference: In CTV, this is stitched into a linear stream. In DOOH, we generate a manifest of URLs pointing to CDN chunks.

3. The “Dumb” Player (The Fetcher):
   The player on the street becomes a dumb terminal. It receives the Manifest from the server.

   • It sees: Slot 1: Content.mp4, Slot 2: Ad_Nike_Chunk.mp4.

   • The Prefetch & Cache: Instead of streaming these live (risky!), the dumb player simply pre-fetches these files from the CDN and caches them locally.

4. Dynamic Assembly:
   When playback time comes, the player just renders the files it cached.

   • This gives us the flexibility of CTV (changing ads on the fly by updating the manifest on the server).

   • But keeps the reliability of DOOH (local playback, no black screens if internet drops).

Why this is the Holy Grail:

• Thin Client: You don’t need a $2,000 PC behind the billboard. You can use a $50 Android stick because all the heavy lifting (logic, decisioning, playlist building) happened on the server.

• CDN Efficiency: Ads are stored on global CDNs (Content Delivery Networks). The player just grabs the chunks it needs.

• Unified Control: You can change the schedule for 10,000 screens in seconds by just updating the manifests on the server. You don’t need to SSH into every single player to update a config file.

Epilogue: The Meteor is Already Visible

By the Time You Read This, Another Media Plan Was Just Copy-Pasted

So, where does this leave us?

We are standing at the edge of a very expensive cliff. On one side, we have a $40 billion industry that is arguably the most powerful creative canvas in the world. It’s un-skippable, it’s massive, and it exists in the real world where actual humans live, eat, and commute.

On the other side? A pile of burning money fueled by manual spreadsheets, “trust me bro” impression numbers, and a technical infrastructure that looks like it was wired together by a drunk electrician in 2008.

I’ve seen this movie before. I saw it with Cable TV when Netflix showed up. I saw it with Print when Google showed up. The incumbents always say, “We are different. Our relationships matter more than your algorithms.”

They are always wrong. Efficiency always wins. Transparency always wins.

The Roadmap to Salvation (Or How Not to Get Fired)

If you are a CPO, a CTO, or just someone who cares about not vaporizing ad spend, here is your final checklist. Tear it out and tape it to your monitor:

• Kill the “Smart Player”:
  Stop buying expensive PCs to put behind billboards. Stop putting logic on the edge. The player should be a dumb, obedient retriever. It fetches a Manifest, it caches the files from a CDN, and it plays what it’s told. The Server is the brain. The Player is just the muscle. This is the only way to scale without crashing.

• Murder the Multiplier:
  If a media owner hands you a spreadsheet saying a screen gets “50,000 impressions,” throw it in the trash.
  Demand Probabilistic Measurement.

  • Did we see Mobile Ad IDs (MAIDs) in the geofence?

  • Did the timestamps match the play log?

  • Did a third-party data partner verify the reach?
    If the answer is no, you aren’t buying media; you’re making a donation to a landlord.

• Stop “Streaming” to the Street:
  We are not Netflix. We are “Store and Forward.” Use the intelligence of CTV (Manifest Manipulation) to manage the schedule, but use the reliability of Caching to ensure the ad actually plays when the 4G drops.

• Force Transparency:
  If you are buying Programmatic DOOH, ask to see the Supply Path. If there are four logos between your money and the screen, you are being robbed. Demand to know if the screen has a permit. Demand to know if the “Dynamic Creative” actually rendered or if it defaulted to a backup JPEG because the player froze.

A Final Warning

The DOOH industry is currently operating like a casino where the house controls the dice, the cards, and the surveillance cameras. But the gamblers (the advertisers) are getting smarter. They have data now. They can track footfall. They can track app conversions.

If we don’t fix the plumbing—if we don’t modernize the tech stack to be as fluid, transparent, and verifiable as the web—the money will dry up. It will go back to Facebook, back to CTV, back to places where the math adds up.

I built the DVR to let people skip ads. Now I’m trying to build the tech to make sure they actually see them. It’s a funny circle of life.

We have the best screens in the world. Let’s stop powering them with AA batteries and hope.

Now, if you’ll excuse me, I have to go explain to a Board of Directors why “Blockchain” will not solve their latency issues.

If you work in the AdTech domain or are remotely connected to it, I believe you would have surely heard of identity wars and Privacy Sandbox. Safari & Firefox deprecated 3rd party cookies long back and then, Apple also blocked IFAs and now, they released Private Relays to block IP address too. At the same time, you would have heard about Anti-trust lawsuits against Google and Google is planning to deprecate 3rd party cookies from Chrome starting Q1 2024. Although cookies were invented for altogther a diifferent purpose in 1995, online advertising industry latched on to it and it became a multi-billion dollar industry pegged to move towards being a trillion dollar industry. but this is not the time to go into history of cookies. This is time to deep dive into the future. and the best way to do that is to simplify Privacy Sandbox by Google.

Google has been working on Privacy Sandbox initiative for last few years now renaming it multiple times. Many other companies are participating in the discussion and working with them to evolve a solution for the entire industry. Prima Facie, everything looks great except one major thing (or 3 more things that I’ll mention in the last section of the article), it is excrutiatingly difficult for people to understand Privacy Sandbox. Forget the regular people, even seasoned AdTech experts and even engineers are struggling to understand how the whole thing works. Believe me when I say this or actually, don’t believe me but just navigate through hundreds of pages of documentation laced with APIs, technical jargons, code and thousands of hyperlinks providing you the best experience of escape room (or a labyrinth) you can’t find a way out. So, I thought of making an attempt to simplify it and kick off a discussion after spending more than a month to read through everything. My disclaimer still exists that I can be wrong in my understanding due to a fact that the project is still evolving and it is too cumbersome to learn everything by just reading hundreds of pages.

Anyway, Lets Begin.

What is Privacy Sandbox ?

As Google wants to deprecate 3rd party cookies and block advertising IDs that can identify a user (no judgements on right or wrong, let the courts and industry decide), they needed to provide an alternate to the world. While Apple and Firefox did the same without worrying about what happens to advertising industry, Google took a lead in building a solution. Rightly because if they don’t do it, their whole existence will have a question mark.

Privacy Sandbox is an attempt by Google to provide a solution to facilitate personalized advertising or targeting advertising in somewhat similar way as it exists today while still taking away the 3rd party cookie from the system. In a simple language, remove the support for 3rd party cookie (data) and build cohorts ( groups similar users) to facilitate targeted advertising on these cohorts rather than individuals. By this change, all the advertisers (and related parties like SSPs, DSPs, DMPs and such platforms) will not be able to leave their trace (3rd party cookie) on the user device so that they can look back at it to identify user. Look at the oversimplified diagram below.

I hope the above diagram could successfully explain the key difference between cookie ( or in future, ID ) based solution and Privacy Sandbox by Google.

Lets deep dive a bit more and I’ll make sure not to use any references of code, APIs, and other technical stuff. Privacy Sandbox is a collection of various modules. But the three main key modules or components are :

1. Topics : Pre-defined cohorts published by Google and Chrome assigning users to cohorts as and when users are browsing the internet. As of today, there are approximately 500+ predefined cohorts ( oops!! Topics) and there are multiple ongoing discussions to increase the number to 1500 or so to match IAB spec of seller defined audiences.

2. Protected Audiences : again, These are cohorts just that they are now defined by publishers/retailers/ad platforms. The key difference is that these cohorts are still managed by Chrome. There is an ongoing discussion about it being server side ( truster server etc). but more on it later.

3. Attribution & Reporting : Anyone who is part of advertising wants to get a report and also, figure out the conversion or attribution to sale etc. Chrome manages all attribution and reporting and ad platforms ( or interested parties) can get this data.

There are a bunch of other modules but to understand Privacy Sandbox, the above 3 modules are the most important ones. But I’d like steer away from the confusion and complexity at this stage.

Topics ( umm… Cohorts)

Think of zodiac signs. A long long time ago (IDK when; too lazy to google), someone decided to divide the human beings into 12 zodiac signs. So, 7 billion people on earth are now classified into 12 categories with an assumption that they must be having similar traits. So, if there are 400 million scorpios (human beings, not the arachnids) in the world, they must be having similar traits. What Google decided was that 12 categories are too less to divide 7 billion people, let’s create 500+ categories (and may 1500+ in the future). and if these 7 billion people can be into 500 categories (ofcourse with permutation and combinations), then the privacy problem goes away and targeting advertising will work magically like it works today.

Jokes apart !! Google wanted to address the problem of privacy by deprecating 3 party cookies (and rightly so !). So, as explained in the above diagram, they introduced topics as a set of APIs where ad platforms can register their interest for any topics and Chrome can then allow the auction to happen for any ad slot if the publisher allows it. Each Topic has an identifier called topic ID. This list of topics is pre-defined and published so that all the interested parties can follow this standardized taxonomy and interact based on the topic ID. Typical steps would look something like this.

1. User browsing a website gets allocated to a pre-defined topic by Chrome.

2. Chrome refreshes the topic list of a user every week ( or every day, not sure).

3. When an ad spot occurs on a publisher website ( like yahoo finance), a bid request is sent with a topic ID.

4. Ad Platforms who are interested in that topic would recognise the topic ID and decide whether they want to bid or not.

5. Google Chrome will receive the bids from the interested ad platforms.

6. Google Chrome will run an auction on the user device itself. (there are mechanisms to connect auction algos etc. but that discussion is for another time).

7. Based on the auction algo, Chrome fetches the winning ad creative and gives it to publisher website.

8. publisher shows the Ad to the user.

The steps are pretty simple like the usual bidding process. The key difference is that ad platforms now do not know about the individual user as their is no 3rd party cookie so they need to bid based on their understanding of cohort. Also, the second key difference is that he whole process is managed on user device (a.k.a Google Chrome)

That’s it about Topics in a simple manner. And everything else is just the engineering documentation to understand the integration details.

Protected Audiences ( umm… platform defined cohorts)

The key drawback of topics is that these cohorts worked like zodiac signs as they are pre-determined by Google. What happens if any publisher or ad platform wants a different set of cohorts as they feel the current taxonomy is limiting and not doing justice to their business. So, the module protected audiences is created by Google. for the lack of any better word, I call them platform defined cohorts as they can defined by any player in the online advertising ecosystem. Lets look at the following diagram.

There a few key differences to note in Protected Audiences compared to Topics.

1. Topics is a set of pre-defined cohorts (standardized taxonomy) where as Protected Audiences are the cohorts ( or audience segments in legacy terms) owned by ad platforms and managed by Chrome on the device. As I mentioned earlier, there is a discussion on moving this to server-side, but let’s keep that out of scope for this article.

2. Any Ad platform can register interest in Topics and place bids but as protected audiences are owned by ad platforms, they can either use it only for the campaigns running on their platform or may permit others depending on the terms of usage of protected audiences.

3. The most important difference is that protected Audience can leverage first party data and can be a lot more accurate (or relevant) than topics.

If both Topics and Protected audiences are available, Chrome will make sure to make both of them available on the bid stream (ofcourse, with required permissions).

Typical steps to build and user protected audiences can look like this :

1. Ad Platform ( or publisher) informs Chrome that it wants to create a protected audience. Typically, these audiences will be created based on first party data.

2. Chrome will register the owner and the rules to add users to protected audiences defined by the ad platform.

3. if a user falls into the realm of the protected audience rule set, Chrome will add that user to that audience segment.

4. When an ad spot occurs on a publisher website ( like Yahoo finance), Chrome will decide whether the bid request needs to content protected audience signal or not based on ad platform partnership with the publisher.

5. ad platform gets a bid request containing the protected audience signal and decides to place the bid.

6. Rest of the auction process remains the same where Google Chrome receives the bids, runs the auction algo, fetches the creative and gives it to publisher to render it.

Protected Audiences is a key module where ad platforms and publishers can create their own audiences and still retain the rights to their own data and where they can sell their audiences to advertisers and generate data revenue as they do it today.

Attribution and Reporting

This is a module that may need significant upgrade. One of the biggest gaps in the online advertising industry is accurate measurement of attribution and providing as much granular reports as possible. With Privacy Sandbox, this becomes a lot more probabilistic that deterministic. Basically, Chrome is classifying the user actions in two sections :

1. Source : when a user views an ad or clicks on an ad, it’ll be classified as source.

2. Trigger : When a user takes any action post the view or click, that action or event is called as Trigger. For example: adding a product to cart, purchase a product, view a product etc.

With the help of this module, Chrome can provide two levels of reporting:

1. Event Level : The detailed event log will be sent by Chrome if its within specified attribution window. For example: If a user views an ad, then clicks on after a few seconds and then adds a product to cart and purchases it, a bunch of event logs will be sent to the registered ad platform.

2. Aggregate Level : Chrome will aggregate all data and create a report of that aggregated data. I don’t know why will any ad platform be interested in aggregate data unless they are licensing data to others and building controls over it.

There is nothing magical in attribution and reporting module. But there are a quite a few constraints as of today.

1. The attribution window is supposedly 3 day window. I am not sure if its configurable or dynamic based on publisher or ad platform needs. What it means is if a user views an ad or clicks on ad today to browse the product but makes a purchase on the 4th day, then the purchase event will not be attributed to the ad provider.

2. In the event level reports, only last 3 triggers per source will be maintained. So, if a product sales lifecycle had 5 events ( example: view product, add to cart, remove from cart, change sku by changing color or size, add to cart, purchase), only last 3 will be sent back to the ad platform and that too if all of them happened with in the 3 day attribution window.

3. although a lower ( sub 1%), but some level of noise will be added to attribution data. In a simple language, some fake data will be added to the actual data.

4. Reporting windows are set to 2 days, 7 days or 30 days.

5. Custom Parameters are not supported in event level reports. For Example: if you want to see the product info, price etc in the event log, you cannot get that data back.

My take on Privacy Sandbox ( Questions I am still looking answers for )

I read through a lot of Privacy Sandbox documentation published by Google and other industry players. Generally, I am always in awe of Google products. Whether its Gmail, Android OS, Drive, Docs, Play store, Google Pay and ofcourse, Search, Google has been great at building products and making it simple for users and developers alike. But with Privacy Sandbox, I felt a bit different. So, I am listing out a bunch of questions that are open for me and I am still looking for answers. If you know the answers, feel free to message me on twitter or LinkedIn or just comment in this article.

1. More of a request to Google : Can Privacy Sandbox team create a human understandable documentation or workflow for Privacy Sandbox ? I mean create a simple non-tech version first before you publish the links to github.

2. Chrome & Android seems to be the centre-piece for Privacy Sandbox. How does it solve the industry problem ? There are still a significant number of users who use iPhones, Safari and Firefox browsers. So, how can Privacy Sandbox address the Apple ecosystem challenges ?

3. Even within Google ecosystem, if Chrome retains the control, isn’t the whole system biased towards Google ? I mean how will Google make sure the Google services ( DV360, Performance Max, Ad Sense or Google properties) are not getting more benefits than open internet players ?

4. I could not find “user consent” much discussed in any of the workflows. Does Privacy Sandbox assumes consent not being necessary anymore as individual targeting is not happening anymore ?

5. While there are discussions about cohorts being server side ( trusted server stuff), why didn’t Google take that approach from the beginning and invested a huge deal in letting Chrome control it ?

6. Why can’t Google hide all the API complexity and just release a bunch of SDKs ( whether client side or server side) to make life for the ecosystem players easy in integration ?

I strongly feel that privacy sandbox is a step in the right direction but I think the odds are stacked in favour of Google. On top of it, there is another discussion of ID (email/phone number) based identity and privacy discussion going on that conflicts with privacy sandbox. and my opinion is that for most of the ecosystem players, it’ll be integration requirements to support for ID and ID less environments. any company that takes only one approach might lose in the long term unless one of the strategy itself fizzles out.

This was my first attempt to simplify Privacy Sandbox topic. Now, I am on to my next version of Privacy Sandbox to do a bit more deeper dive. or shall I make an attempt to build a service around it🙂 ..

by the way, if you like the article or have disagreements or would like to highlight any errors, please leave a comment. you can follow me on twitter or LinkedIn too for further discussions.