Now that Agentic Advertising Is Getting Its Standards, what about Measurement?

IAB Tech Lab released AAMP 3.0 in late September, including a new draft specification called OpenProposal, which is open for public comment through October 22. It is another sign that agentic media buying is starting to move from a concept into something the industry is actually building around.

OpenProposal focuses on the part of media buying between the brief and the buy. An advertiser creates an RFP, sellers figure out what they have that fits, proposals go back and forth, people compare them, terms get negotiated and eventually a buy gets made. A good amount of that still happens through email, spreadsheets, PowerPoint and conversations between people.

The specification gives buyer and seller agents a common structure for handling much of that work. Sellers can describe their products in a standardized, agent-readable format, while buyer agents can discover those products, communicate what the advertiser is trying to accomplish, receive proposals and evaluate them against the brief. From there, agents can negotiate terms and connect an accepted proposal to the systems used to execute the buy.

AAMP is broader than OpenProposal. IAB Tech Lab is building it around many of the standards already used across digital advertising, including AdCOM, OpenDirect, the Deals API and, more broadly, OpenRTB. Instead of replacing the existing infrastructure, the idea is to make more of it usable by agents.

There is another effort happening at the same time. AdCP, from AgenticAdvertising.org, is also building an open protocol around agentic advertising. It covers many of the same functions, including media discovery, planning, buying, creative, signals and reporting.

The overlap between the two is significant. IAB Tech Lab has even published its own comparison showing that AAMP and AdCP both touch planning, discovery, negotiation, ordering, governance, audiences, execution and reporting. The approaches are different though. AAMP starts with many of the standards already running digital advertising, while AdCP defines its own cross-channel agent contract and connects to existing standards where it makes sense.

I don't know that it is particularly useful yet to try to predict which approach the market settles around. Will it be VHS or Beta? I don't know, but what is more interesting to me is how quickly both efforts are putting structure around the information an agent needs to understand what is available, determine whether it fits a brief and transact against it.

AAMP 3.0 makes that pretty tangible. OpenProposal allows more of the context around a proposal to travel with it. A seller can describe the audience, formats, terms and why a particular package fits what the buyer asked for. That gives an agent more than inventory and a price. It gives it information it can use to evaluate the proposal, which raises the harder question of what the agent should make of the results once they come back.

Reporting gets us part of the way

Both AAMP and AdCP support reporting. AAMP allows performance to flow back through existing reporting standards so results can be evaluated against the objectives and commitments in the original plan. OpenProposal actually goes further than basic reporting, allowing proposals to specify committed metrics, a measurement source and attribution window, and the type and cadence of reporting. AdCP includes delivery reporting in its media-buy protocol and can carry information used by existing measurement and attribution systems.

That is important infrastructure, but getting the result and its reporting context back to an agent doesn't necessarily tell the agent how to interpret the evidence behind it.

Take a CTV campaign where the objective is incremental sales. When the campaign ends, the agent could receive impression data, reach, conversions, sales and an attribution result showing a 3.2x ROAS. All of that information can move cleanly between systems while still leaving a fairly basic question unanswered: how many of those sales happened because of the advertising?

How the campaign was measured matters too. An attribution provider may give the campaign one level of credit, an incrementality study another and an MMM something different again. There may be nothing wrong with any of those results. They can differ because the methodologies are answering different questions and using different assumptions.

This is normal for people who work in measurement. We look at the methodology, understand what question a study was designed to answer and decide how much weight to give the result. We know that an attributed conversion and an incremental conversion aren't the same measure simply because both are reported as campaign outcomes. Agents will eventually need to make some of those distinctions themselves if campaign performance is going to influence what they buy next.

Neither framework gives an agent a universal methodology for determining what caused an advertising outcome. AdCP is explicit about this in its documentation, saying it does not specify attribution or viewability and instead works with existing measurement providers and standards. AAMP takes a similar practical approach by routing performance back through existing reporting and measurement infrastructure.

That makes sense at this stage. There are already established systems for reporting exposures, events and outcomes, and there isn't one methodology that can answer every advertising question. The more difficult issue is what information needs to accompany a measurement result if another machine is going to use it as evidence.

Making the evidence understandable

A media outcome can mean very different things depending on how it was produced. An agent may need to know whether the result is attributed or incremental, whether it came from an experiment or a model, what population it represents, what the comparison was and how much uncertainty surrounds the estimate.

Most measurement teams already document this information somewhere. It lives in methodology documents, research decks, PDFs and the knowledge of the people doing the work. That has been fine because another person usually sits on the other side of the result and can interpret it, but that assumption changes when agents begin evaluating the evidence directly.

There is a useful parallel with what OpenProposal is trying to solve. Media proposals have always contained the information needed to make a buying decision, but much of that information has been trapped in emails, presentations, post-campaign reports, spreadsheets and conversations. OpenProposal gives it more structure so an agent can understand and use it.

Measurement has a version of the same problem. The result is usually there, and the methodology is usually there somewhere too. They aren't consistently packaged together in a way that makes the meaning of the result easy for another system to interpret.

I don't think the answer is one industry-wide measurement methodology. Experiments, attribution, MMM, brand lift and other approaches all have legitimate uses. The more realistic opportunity may be to create greater consistency around how the evidence behind those results is described.

That would let an agent understand what kind of result it is looking at, how it was produced and whether it is comparable to another result before using it in a future buying decision.

Why this matters for publishers

AAMP 3.0 is especially relevant for media sellers because OpenProposal gives publishers another way to make their products accessible to buying agents. If agents can evaluate more opportunities than a human planning team could realistically review, the number of sellers that can participate in a plan could expand.

That also means publishers have to think about how their products are understood by the agent, not only how they are presented to a planner. Measurement will likely follow the same path.

Publishers have spent years building evidence that their media works through brand-lift studies, attribution analyses, incrementality tests, audience research and custom studies. Most of that evidence was created for people. It lives in sales decks, white papers and campaign reports, with researchers and sellers providing the context needed to understand what the results mean.

If agents start using past effectiveness as an input into future buying decisions, that context becomes part of the product too. A strong incrementality study has limited value to an automated system if the lift number is separated from the methodology, comparison and assumptions needed to understand it.

AAMP and AdCP are both pushing the industry toward a world where more of media buying can be understood and acted on by machines. A lot of progress is being made on inventory, audiences, briefs, proposals, pricing, transactions and reporting.

Measurement is a different kind of standardization problem because there isn't a single field that can reliably answer whether a campaign worked. The answer depends on the objective, the methodology and the evidence behind the result. As agents begin using those results to decide what happens next, making that context understandable to them will become part of the infrastructure as well.

Sources

• IAB Tech Lab, IAB Tech Lab Introduces AAMP 3.0 with OpenProposal, September 22, 2026.
• IAB Tech Lab, Clarifying the AAMP Scope: Why So Many AAMP vs. AdCP Comparisons Miss the Mark.
• IAB Tech Lab, OpenProposal: AAMP 3.0 Object Model Specification, draft opened for public comment September 22, 2026.
• AgenticAdvertising.org, AdCP: Advertising Context Protocol.
• AgenticAdvertising.org, AdCP Frequently Asked Questions, including current guidance on attribution, viewability and its relationship to AAMP.