OpenProposal is a draft technical specification from IAB Tech Lab that defines how a seller describes an advertising offer, and how a buyer's software agent reads, compares and adjusts that offer, before any money is committed. It covers the stretch of media buying that has long run on emailed requests for proposal (RFPs), spreadsheets and PDF media kits. It exists because the industry's transaction standards start once a deal is agreed, leaving automated agents no common format for the offer itself.

The specification is the core of version 3.0 of the Agentic Advertising Management Protocols (AAMP), the umbrella under which IAB Tech Lab develops its standards for AI agents in advertising. As of October 2026 it is a draft open for public comment, tagged v3.0-draft-1, and its own repository states that "nothing here is a ratified IAB Tech Lab standard."

What the specification describes

OpenProposal defines two objects. The proposal is the parent: it carries identity and lifecycle data such as proposal_id, version, seller_id, status, an offer window set by valid_from and valid_until, and a negotiation_history[]. It also carries what the draft calls agent comprehension fields, a plain-language description plus best_for[] and not_suitable_for[] lists. These lists are advisory: the draft says they are "never the basis for a match."

The line item is the buyable unit. Each one covers a single channel, its audiences, formats, inventory, creative requirements, pricing, guarantees and the route by which it will be executed. The draft states that "each line item MUST be fully self-contained": nothing is inherited from the proposal or from another line item. The rule stops an agent misreading a price set several levels away from the unit being bought.

In total the draft runs to 71 fields, which PPC Land counted under 11 headings grouped into merchandising (identity and lifecycle, agent comprehension, inventory composition, audience reach, creative requirements), optimisation (the targeting envelope), commitment (commercial terms, guarantees, transaction eligibility) and reconciliation (execution and settlement reference).

Pricing sits in a pricing[] array with cost_method, gross_rate, agreed_rate, seller_rate, currency, price_valid_until and an optional floor. Guarantees sit in committed_metrics[], where each entry names a metric, a value and a basis, alongside a measurement_source, a shortfall_remedy and reporting terms. The draft's examples use viewable rate for display, completion rate for video, audio and connected TV (CTV), and an audience-modelled metric for digital out-of-home (DOOH).

Constraint satisfaction instead of haggling

The design choice that most separates OpenProposal from a conventional RFP is field mutability. Every field carries one of seven markers that say who may change it:

  • immutable - fixed at creation;
  • seller-set - revisable by the seller while the proposal is in draft or published;
  • selectable - the buyer picks from seller-declared options, which may carry a price change;
  • settable - the buyer enters a value inside seller-declared bounds, such as budget or flight dates;
  • supplied - the buyer provides data, such as its own audience segments or suppression lists;
  • requestable - the buyer asks for something outside the declared space, which needs seller assent;
  • derived - calculated from other fields.

Only three fields are requestable: materials_due, hold_status and committed_metrics[]. That means 3 of 71 fields, about 4%, require a round trip to the other party. The rest can be resolved by the buyer's agent alone. According to a September 30 analysis by Bedrock, an ad tech vendor, the seller publishes a menu and the buyer knows the final price before contacting the seller. A brief is optional; one can still be sent, and a custom proposal comes back in the same format.

Sellers can also point to reusable definitions through catalog_ref, a version-pinned reference in the form <seller_namespace>/<collection>/<entry_id>@<version>. When a proposal's status becomes agreed, those references must be replaced with the resolved values.

The lifecycle and the hand-off to execution

According to IAB Tech Lab's announcement, OpenProposal supports seven stages. Sellers publish public packages and hold back products for custom proposals. Buyer agents discover packages or submit briefs, compare and qualify them against the brief, and negotiate custom terms. Line items are then bound and executed on existing standards, and performance is reported through existing ones.

Proposal status moves through draft, published, under_review, agreed, withdrawn and expired. A line item has its own status, which adds sold_out.

Execution is where OpenProposal stops and older standards take over. A field called transaction_mechanism takes one of five values: DIRECT_IO, PROGRAMMATIC_GUARANTEED, PREFERRED_DEAL, PRIVATE_AUCTION or OPEN_AUCTION. Each line item is then routed to OpenDirect 2.1, the Deals API or OpenRTB 2.6. AdCOM, IAB Tech Lab's shared advertising object model, supplies the vocabulary across every stage. A seller agent publishes and answers proposals, managed alongside the publisher's order management system and ad server; a buyer agent reads them. Messages can travel over the Model Context Protocol (MCP), Agent2Agent (A2A) or gRPC.

Origin and evolution

The gap OpenProposal fills is visible in the standards that preceded it. OpenDirect, the IAB specification for automated guaranteed buying, added price negotiation and buyer discovery of deals in version 2.0, released for comment in January 2018. The Deals API, released for comment on December 5, 2025, standardised how deal terms pass from a supply-side platform (SSP) to a demand-side platform (DSP), but its specification said it "does not support proposals, revisions, or negotiations in version 1.0."

Agentic buying exposed that missing layer. The Ad Context Protocol (AdCP), launched on October 15, 2025 by six companies including PubMatic, Scope3 and Yahoo, included a get_products task that takes a natural-language campaign brief and returns matching inventory. IAB Tech Lab named its own programme AAMP on February 26, 2026, building on OpenRTB, OpenDirect and the Deals API rather than replacing them. AAMP 2.3, released on July 30, 2026, added a pricing provenance field after an agent invented a cost per thousand impressions (CPM) where real pricing data was missing.

AAMP 3.0 followed on September 22, 2026, the first AAMP release to define a specification rather than only reference code, according to Bedrock. "OpenProposal fills a major gap between a buyer's brief and a buy commitment," Anthony Katsur, chief executive of IAB Tech Lab, said in the announcement. Mike Treon, head of CTV strategy at PMG, said: "What's new in AAMP 3.0 is that the reasoning travels with the proposal." IAB Australia said it was running a seller-side pilot in a sandbox.

Why it matters for marketers

The RFP stage is labour that rarely shows up in a media plan. A planner briefs a handful of publishers, receives proposals in different formats and normalises them by hand. IAB Tech Lab's Barnaby Edwards has used 10 to 20 publishers to illustrate how many a human team can review, a bandwidth illustration rather than measured data. A common format would let an agent rank far more offers against the same brief.

For publishers, the stakes are visibility: packages that are not machine-readable may never be considered. A shared format also lowers integration cost. Adform's chief technology and product officer, Jochen Schlosser, estimated that a custom deals integration with one SSP takes 250 to 350 days of work, against about 50 days on a shared standard.

Limitations and disputes

The draft is incomplete by its own account. Machine-readable schemas are not published and are tied to the first revision after comments close. The repository warns against building production systems on it. Seven open questions remain, including whether per-field mutability and catalog_ref belong in 3.0 at all, how to compare guarantees across channels, and the absence of audio, CTV and DOOH entries in the format registry.

Bedrock's analysis found the buyer side "still thin": the draft does not define what a brief contains or how a buyer returns its selections. It also noted that the draft's own display example adds a clean room add-on to the final rate without any field recording that the buyer chose it.

Competition is the larger question. IAB Tech Lab's chief operating officer, Shailley Singh, listed 13 business functions that AAMP and AdCP both address, including natural-language briefs, product discovery and negotiation. Schlosser told ExchangeWire that very little money runs through either protocol, and that one may die, the two may merge or both may survive. A proposal format only delivers comparability if most sellers adopt the same one.

OpenProposal versus adjacent terms

  • Ad Manager proposals. In Google Ad Manager, according to Google's help centre, a programmatic proposal is a working draft that seller and buyer revise until the buyer accepts, when orders and line items are created. It is a feature of one platform, not an open schema.
  • OpenDirect. OpenDirect handles ordering of guaranteed buys once terms are known; OpenProposal sits upstream.
  • Deals API. The Deals API pushes agreed deal terms from SSP to DSP and excludes proposals.
  • AdCP get_products. AdCP's discovery task answers a brief with inventory, overlapping OpenProposal's discover stage under a rival protocol governed by AgenticAdvertising.org.

Recent developments

The comment period was due to close on October 22, 2026. On October 23, IAB Tech Lab extended public comment on AAMP 3.0 and OpenProposal to November 10, 2026, citing demand. "Wow! Public comment on AAMP 3.0 and OpenProposal has been immense!" Katsur wrote. The 30-day window now runs 49 days. Neither IAB Tech Lab post disclosed how many comments arrived. As of today the repository shows 99 open issues, including its own tracking issues, and 9 open pull requests. The governance file still leaves working group members, chairs, the decision rule and publication dates blank, and IAB Tech Lab has not said whether the schema timeline moves with the deadline.

Timeline

  • January 2018: IAB Tech Lab releases OpenDirect 2.0, adding price negotiation and deal discovery, for comment through March 12, 2018
  • October 15, 2025: Ad Context Protocol launches with a brief-driven get_products task
  • December 5, 2025: Deals API 1.0 released for comment, excluding proposals and negotiation
  • February 26, 2026: IAB Tech Lab names its agentic programme AAMP
  • July 30, 2026: AAMP 2.3 adds a pricing provenance field
  • August 20, 2026: IAB Tech Lab's Shailley Singh lists 13 functions shared by AAMP and AdCP
  • September 22, 2026: AAMP 3.0 and OpenProposal v3.0-draft-1 published for public comment
  • October 22, 2026: Original comment deadline
  • October 23, 2026: IAB Tech Lab extends comment to November 10, 2026
  • November 10, 2026: Extended comment period closes

Summary

Who: IAB Tech Lab develops OpenProposal through a working group under its AAMP programme, led publicly by chief executive Anthony Katsur. Supporters at launch included WPP Media, PMG and IAB Australia. Publishers, SSPs, DSPs and agencies building buyer and seller agents are the intended implementers.

What: A draft specification of 71 fields defining a proposal object and self-contained line items, with seven mutability markers that let buyer agents resolve most terms without negotiation. Only 3 fields require the other party's assent.

When: Published as v3.0-draft-1 on September 22, 2026. Public comment, first due to close October 22, was extended on October 23 to November 10, 2026. Schemas follow the first revision.

Where: The specification lives in a public IAB Tech Lab GitHub repository, open to comment from anyone with a GitHub account. Agreed line items execute through OpenDirect 2.1, the Deals API or OpenRTB 2.6, across display, CTV, mobile apps, audio, DOOH, linear TV and walled gardens.

Why: Existing standards began at the deal. The brief-to-commitment stage ran on email and PDFs that AI agents cannot reliably compare, and OpenProposal aims to make offers machine-readable in one format that any agent can read.