Demand · 24 Sep 2026
What Is AAMP 3.0? How OpenProposal Standardizes the RFP-to-Buy Process
IAB Tech Lab wants AI agents to compare ad proposals the way a spreadsheet compares rows. OpenProposal is the shared form. Here is what that means, without the protocol jargon first.
On 22 September 2026, IAB Tech Lab released AAMP 3.0, the next version of its Agentic Advertising Management Protocols. The new piece is OpenProposal: a standard way for buyers and sellers to go from a request for proposal to a buy commitment when AI agents do the reading.
The specification is in public comment until 22 October 2026. It is a draft the industry can still change. It is not a switch that turns every campaign over to an agent tomorrow.
AAMP 3.0 standardizes the messy middle of advertising: the stretch after someone writes a brief and before someone agrees to spend.
Start with a normal buying problem
Say a brand wants to launch a shoe in India next month. The plan needs mobile ads, a few podcasts, and some screens in malls. The budget, the dates, and the audience are written down. That document is an RFP — a request for proposal. It asks sellers, “What can you offer, at what price, and can you actually deliver it?”
Today that request usually becomes email. A planner sends it to 10 or 20 publishers and platforms they already know. Each reply comes back as a different PDF, a slide, or a thread. Someone copies prices into a spreadsheet. Hundreds of other sellers never see the brief, because no team has time to ask them. Sellers have the same limit: they answer the RFPs they can staff, and skip the rest.
IAB Tech Lab’s point is that this is a time problem, not a talent problem. People are good at judging a proposal. They cannot personally read every proposal in the market.
An AI agent can read far more replies than a person. It still fails if every seller describes the product in a private dialect. One PDF says “premium mobile.” Another says “in-app high impact.” A third hides the guarantee in a footnote. The agent cannot tell which rows match.
OpenProposal is the agreement to use the same boxes.
A plain example of the same form
Picture three caterers bidding on a dinner for 200 people. You asked for a date, a city, a vegetarian option, and a ceiling on price.
- Caterer A sends a glossy brochure.
- Caterer B sends a one-line email: “We can do it.”
- Caterer C sends a spreadsheet with a different column order.
A human can sort that out for three vendors. An assistant asked to compare 200 vendors cannot, unless every reply has the same fields: menu, headcount, price, what is guaranteed if the count changes, and how you pay.
Advertising is that dinner, with more sellers and more ways to buy. OpenProposal asks each proposal — a standard package on a shelf, or a custom answer to this brief — to state, in a structure an agent can read:
- what the product is
- who it is meant to reach
- what it costs and what is guaranteed
- which buying method is allowed
- how it will be executed and settled
The person still approves the spend. The form does not choose the campaign. It stops the comparison from depending on whoever wrote the prettiest PDF.
What AAMP 3.0 actually adds
AAMP is IAB Tech Lab’s framework for letting software agents take part in advertising workflows with shared rules: who the agent is, what it is allowed to do, and how an instruction stays consistent if it is sent twice. Version 3.0 extends that framework into planning, which older ad-tech standards mostly left to humans.
Standards already exist for later steps. OpenRTB runs auctions. AdCOM is a shared vocabulary for ad objects. OpenDirect covers automated guaranteed buys. The Deals API sets up deals. None of those is a common language for “here is my brief” and “here is my proposal.”
OpenProposal fills that gap. A seller can publish products in an agent-readable catalog, and can keep some products back for custom proposals only. A buying agent can browse the public packages, or send the brief and ask for a custom reply. It can then rank what comes back against the brief: audience, flight dates, format, price, and other stated requirements.
IAB Tech Lab CEO Anthony Katsur described the goal as a common framework between the buyer’s brief and the buy commitment, connected to the execution standards the industry already runs. The spec does not ask the market to throw out DSPs, SSPs, or existing programmatic pipes.
The seven steps, in order
IAB Tech Lab describes a lifecycle. You can read it as one campaign moving from a catalog to a report.
- Publish. Sellers list public packages. Other products stay private until a buyer asks.
- Discover. A buyer agent reads those packages, or submits a full brief for a custom proposal.
- Compare. The agent ranks options against the brief. This is the step that can cover more publishers than a human shortlist.
- Qualify. The agent checks whether the option actually fits this audience, these dates, and this plan — not just whether it looks similar.
- Negotiate. If the terms need to change, agents can negotiate a custom version while keeping a link back to the original offer.
- Bind and execute. Once someone commits, the deal uses existing transaction standards. Programmatic guaranteed might become a deal. An open-auction plan might become bidding rules. The proposal is not a second auction protocol.
- Report. Delivery and results come back through reporting the industry already has, so the outcome can be checked against the original promise.
AdCOM is meant to keep the words stable across those steps. “Audience” at discovery should mean the same thing at reporting. That sounds small. It is the difference between a plan you can audit and a plan that was rewritten in five tools.
The eleven headers, without the spec voice
IAB Tech Lab groups OpenProposal into eleven headers and four jobs. You do not need the field names to see the jobs.
| Job | What the agent is supposed to learn |
|---|---|
| Merchandising | Identity and lifecycle of the offer, a summary an agent can understand, what inventory is in the package, who it reaches, and which creatives it accepts. |
| Optimization | The targeting envelope: the bounds of who can be included, not a vague “premium users.” |
| Commitment | Commercial terms, guarantees, and whether this buyer is allowed to transact this product this way. |
| Reconciliation | How the buy will be executed, and a settlement reference so finance can tie the proposal to the bill. |
A proposal is not an impression and a CPM. The buyer also needs to know whether measurement is supported, whether a banned adjacency is excluded, and whether the seller can deliver what the price assumes. Those checks are why a shared proposal is more than a rate card.
Mike Treon, head of CTV strategy at PMG, made the practical version of that point: the useful new part is that the seller’s reason for the fit — audience, formats, and the logic behind the price — travels with the proposal in a form a planning model can read.
It is not only for open-auction display
OpenProposal is written to be agnostic about how the ad is bought and where it runs. IAB Tech Lab’s examples include:
- direct insertion orders
- linear television priced on GRPs and CPPs
- programmatic guaranteed
- preferred deals
- the open auction
The same shape is meant to cover open-web display, connected TV, mobile apps, audio, out-of-home, linear TV, and walled gardens. A shoe launch that mixes in-app ads and mall screens should not need two unrelated proposal languages if both sellers adopt the spec.
That does not mean every walled garden will publish its catalog this year. It means the standard was not designed as an open-web-only form.
What this does not replace
Three boundaries matter if you buy or sell programmatically.
It does not replace the auction. When the committed buy is programmatic, bidding still happens over OpenRTB, at auction speed. An agent that compares proposals is doing a slower job: planning. Our earlier note on what happens when AI agents start buying ads separates that planning layer from the milliseconds inside a bid. AAMP 3.0 is a standard for the planning layer, not an instruction to put a language model in the bid path.
It does not force DSPs and SSPs to rebuild current deals. IAB Tech Lab says existing programmatic implementations of AdCOM, OpenDirect, and the Deals API can continue. OpenProposal is supposed to connect into those pipes when a proposal becomes an order. If you already run preferred deals or programmatic guaranteed, the change is upstream: how the proposal was written and compared, not a new bidder.
It does not remove human approval. The interesting claims from agencies are about a wider set of options and less time reconciling documents. Judgment — brand suitability, whether the guarantee is real, whether the seller is authorized — still has to live in policy. Agents that can send an order twice must not create two orders. IAB Tech Lab calls that property idempotency: a retry confirms the original instruction instead of duplicating the buy. That is an engineering guardrail, and it belongs next to spend limits.
Who gains if sellers actually publish this way
Publishers and SSPs can be considered for briefs they never had staff to answer one by one, if their products are in the catalog in this format. Differentiated audiences and packages are easier to find when they are fields, not a paragraph in a PDF. Being absent from the catalog is the new version of never receiving the email.
Agencies can apply a strategy across more sellers and spend human time on the shortlist, not on copying cells. WPP Media’s Michael Palmer framed AAMP 3.0 as a step from pilot projects toward agents that can transact, because authentication, authorization, and the existing IAB standards are in one framework.
Advertisers get a plan where each line is easier to check against the brief. That only helps if the fields are true. A structured lie is still a lie. Supply-path transparency, authorized sellers, and the curation questions in why advertisers are moving toward curated supply do not disappear because a proposal is machine-readable.
What is happening between now and 22 October 2026
The spec is open for public comment through 22 October 2026. IAB Tech Lab is building OpenProposal into its buyer and seller agent SDK reference implementations during that period. Comments can still change headers, required fields, and how a proposal points at AdCOM, OpenDirect, and the Deals API.
IAB Australia has asked publishers to join a seller-side pilot in a sandbox, using today’s reference implementations, before any live spend. That is a useful signal: the first serious tests are discovery, proposal, and approval workflows, not a demand that every SSP flip a switch in production.
There is also a webinar on 8 October 2026 (12:00–13:00 Eastern) with Matthew Jacobs, IAB Tech Lab’s head of agentic technologies, and AAMP architect Aidan Cardella, walking through the headers that keep a proposal transactable and billable.
Primary sources, if you want the spec rather than this summary: the 22 September announcement, the OpenProposal explainer, and the comment package at iabtechlab.com/AAMPv3.
What a buyer or seller can do this quarter
You do not need to wait for the comment window to close before you know whether your own process is agent-readable.
- Write down how a proposal leaves your building today: PDF, email, a deal seat, or an API. If a stranger’s software cannot parse it, an agent cannot rank it.
- List the fields you would hate to leave implicit: price, dates, geo, format, guarantees, measurement, blocked categories, and how the buyer is allowed to transact.
- Separate catalog products you would show every agent from products you only quote when asked.
- Decide who may approve a commitment. An agent that can negotiate should not also be the only signature.
- If you sell programmatic, keep the execution path boring. OpenRTB, deal IDs, and reporting should stay the system of record after the proposal is accepted.
The short version
AAMP 3.0 is IAB Tech Lab’s attempt to give AI agents a shared proposal form. OpenProposal covers the RFP, the comparison, and the path to a commitment. Execution stays on standards the market already uses. The draft is open for comment until 22 October 2026.
If it works, buyers see more of the market than a short email list, and sellers get considered without answering every brief by hand. The work that remains is the same as any other standard: publish true fields, keep a human on the approval, and do not confuse a tidy proposal with a delivered impression.
FAQ
What is AAMP 3.0?
AAMP 3.0 is the September 2026 release of IAB Tech Lab’s Agentic Advertising Management Protocols. It adds OpenProposal, a shared format so buyer and seller AI agents can move from an advertising RFP through proposal comparison and toward a buy commitment, then hand the deal to standards the industry already uses for execution.
What is OpenProposal?
OpenProposal is the specification at the center of AAMP 3.0. It describes an ad product or a proposal in a fixed, agent-readable structure: what is being sold, who it reaches, the price and guarantees, whether the buyer is allowed to transact that way, and how the deal will be executed and settled. It can be a public package or a custom reply to a brief.
Does AAMP 3.0 replace OpenRTB, DSPs, or SSPs?
No. IAB Tech Lab says existing programmatic uses of AdCOM, OpenDirect, and the Deals API can keep working. OpenProposal covers the planning gap between a brief and a commitment. OpenRTB still runs the auction when the buy is programmatic. A DSP or SSP does not have to rebuild its bidder because an agent compared proposals upstream.
When can the industry comment on AAMP 3.0?
IAB Tech Lab put OpenProposal in public comment until 22 October 2026. During that window it is being built into the Buyer and Seller Agent SDK reference implementations. The spec can still change from feedback.
What buying types and channels can OpenProposal describe?
IAB Tech Lab designed it to be agnostic about the transaction and the channel. The published examples include direct insertion orders, linear GRPs and CPPs, programmatic guaranteed, preferred deals, and the open auction, across open-web display, CTV, mobile apps, audio, out-of-home, linear TV, and walled gardens.
← All posts · AI agents and the DSP · What is a DSP? · What is OpenRTB? · Demand partners