AAMP: AI agents buy ad space for you

A human team can evaluate 20 publishers per campaign at most. An AI agent can analyze hundreds in a few minutes, negotiate the price and book the space, without anyone opening an email.

Is it really worth letting an AI agent buy ad space for me?

It depends on how many campaigns you run and how much time your team spends on repetitive tasks like requesting quotes, comparing rates and booking inventory. The higher your volume, the bigger the gain.

AAMP, short for Agentic Advertising Management Protocols (protocols for automated advertising management through AI agents), is the name IAB Tech Lab, the international body that defines technical standards for digital advertising, has given to the system that lets intelligent software buy and sell ad space without direct human intervention.

This isn't science fiction: version 3.0 has been in public draft since September 2026, and some platforms are already testing it. The right question isn't "does it exist?" but "when will it affect me, and what should I do now?"

What really happened: how did AAMP start and where do we stand today?

IAB Tech Lab created AAMP because companies and platforms were already connecting AI agents to advertising systems without shared rules. The standard brings order to something that was already happening.

It all started in November 2025, when LiveRamp donated its user data management protocol to IAB Tech Lab. Over the following months came more building blocks: the framework for real-time auctions and the APIs for programmatic deals. On February 26, 2026, the program was given the name AAMP, to end the confusion in the market.

The underlying logic is simple: instead of inventing a new language, IAB Tech Lab built an AI interface on top of the standards the industry already uses: OpenRTB, OpenDirect, AdCOM. As IAB president Alison Levin put it: "building a new car doesn't mean you have to build a new racetrack too."

In April 2026 came version 2.0, with agentic private marketplaces and guaranteed programmatic deals. In July 2026, version 2.3 fixed a critical problem: agents inventing prices when data was missing (the so-called "CPM hallucination", meaning a hallucination about the cost per thousand impressions). In September 2026 came version 3.0, still in draft.

How it works in practice: who does what in an AAMP campaign?

The process is split into three levels: a central orchestrator manages the budget, channel-specialized agents look for the best inventory, and operational agents handle research, negotiation and booking.

Think of AAMP as an automated buying team. There's a director (the Portfolio Manager) who decides how to allocate the budget. Below them are specialists for each channel: display, video, connected TV, mobile. Further down are the executors: those who research, check availability, negotiate the price, book and then report the results.

On the seller side (publishers, sales houses) it works in a mirror image: one agent manages offers and deals and responds to requests from buying agents.

One important detail: the agents don't replace the DSP you already use. They negotiate and book first, then the deal is activated in your usual tool. For anyone running programmatic campaigns, the integration is less painful than it looks.

To learn more about governing these agents without losing control, read our guide to AI agents in media buying and risk governance.

What changes with OpenProposal, the big news in AAMP 3.0?

OpenProposal turns commercial proposals between publishers and advertisers into machine-readable documents, eliminating the emails and Excel sheets that slow down every campaign today.

Today, when a company wants to buy inventory from a publisher, the process looks like this: emails, PDFs, spreadsheets, phone calls, waiting. It can take days. With OpenProposal, the centerpiece of AAMP 3.0, the publisher posts a structured offer that an AI agent can read, compare and accept (or reject) in seconds.

An offer can be "standard" (published by the publisher without anyone requesting it) or "custom" (built in response to a specific brief). Every field has precise rules about who can change it: some values are fixed by the seller, others are picked from a menu, and others are negotiable within stated limits.

According to the IAB Tech Lab specification, the result is that the buying agent does "constraint satisfaction" instead of negotiation: it compares the available options and picks the one that best fits the configured parameters. Faster, more predictable, less prone to human error.

Three real-world scenarios: what changes for an e-commerce store, an agency and a manufacturer?

The impact varies a lot depending on campaign volume and the complexity of the media mix. Here are three typical Italian situations with an estimate of the benefit.

Illustrative examples: numbers are meant to show the reasoning, they are not real customer data.

Is AAMP the only standard out there, or are there alternatives?

There is a direct competitor called Ad Context Protocol (AdCP), launched by Scope3. The two standards overlap on 13 functions, and the debate over which to adopt is still open.

A common idea in the industry is that AdCP handles "planning" and AAMP handles "execution." IAB Tech Lab has denied this split: the two protocols overlap on at least 13 functions, from natural-language briefs to agent registries.

If you have to decide today: neither one runs at production scale yet. AAMP has the advantage of being built on standards the industry already uses (OpenRTB, OpenDirect), which lowers the risk of obsolescence. AdCP has the advantage of being newer, with a different approach to semantic planning.

What real risks do I need to manage if I adopt AI agents for media buying?

There are three main risks: agents that make up prices, a lack of transparency on who bought what, and underestimated technical integration.

The "CPM hallucination" problem, where an agent produces a convincing price when real data is missing, had already surfaced in the early versions of AAMP. Version 2.3 addressed it by adding a price provenance field and an optional approval gate linked to the IAB verification platform.

A second risk is transparency: who bought what, at what price, with which agent? The Agent Registry, a public register of active agents, is one of the three pillars of AAMP precisely to answer this question. In March 2026 it already had 10 entries, including Amazon Ads.

The third risk is technical: the AAMP 3.0 repository explicitly warns against building production systems on the current version. If you want to experiment, you can do it in a test environment; if you want to go live, you'll have to wait for the stable version.

For a broader view of how AI agents change the customer journey and what you can still control, read this deep dive on AI agents in the customer journey. And if you're figuring out how to scale campaigns without growing your team, this article on AI agents for campaigns offers a practical overview.

Why now: is the digital advertising market really ready?

The European digital advertising market topped €131 billion in 2025, and programmatic is growing fast. The pressure to automate is real and measurable.

The global agentic AI market was worth about $7.3 billion in 2025 and is projected to exceed $139 billion by 2034, according to Forbes Tech Council. This is no niche market: it's the direction the entire advertising supply chain is heading.

To see how this trend fits into the bigger picture of automated media buying, read our analysis of agentic buying in 2026 — part of our hub on Marketing and ADV performance.

How to get started: what are the steps to approach AAMP without risk?

You don't need to integrate everything at once. The smartest path starts by mapping where your team loses the most time today, then testing in a controlled environment before going into production.

What mistakes are most common when introducing automation into media buying?

The most common mistake is delegating everything without setting limits: an agent without clear parameters can spend budget inefficiently or buy placements that don't fit the brand.

Here are the mistakes we see repeated when companies approach media buying automation:

• Not defining negotiation limits. An agent with no maximum CPM configured can accept rates that blow your budget. Set the parameters first, not after.
• Confusing automation with no control. AAMP automates the operational phase, not the strategy. Goals, targets and messaging remain human decisions.
• Going into production too early. Version 3.0 is a draft: building critical systems on a non-final specification exposes you to costly rework.
• Ignoring the problem of invented prices. Before AAMP 2.3, agents could produce plausible but fake CPMs. Always check that the system you use has the price provenance field enabled.
• Not training the team. Whoever manages the campaigns needs to understand what the agent does and when to step in. Automation works best when the people supervising it know how it works.

How do I know if adopting AAMP is delivering real results?

The indicators to track are both operational (time saved, publishers reached) and financial (average CPM, cost per acquisition). You need both for the full picture.

So, is it worth moving on AAMP for your company?

It's worth it if you run multichannel campaigns with a significant monthly budget, you have a team losing hours on repetitive negotiation and quoting tasks, and you already work with DSPs like The Trade Desk or DV360. In that case, starting to explore AAMP today puts you ahead when native integrations reach production.

Somewhere in between if you have medium volumes and are weighing whether to automate. The right move is to map your current bottlenecks and follow the standard's evolution over the next 6-12 months, without investing in custom development now.

Not yet if you run few campaigns a year with a limited budget and few publishers. The integration cost outweighs the short-term benefit. Better to wait for DSPs to integrate AAMP natively, with no extra effort on your part.

How we see it at SHM

For us, AAMP is a precise example of how AI should work in a company: it handles the repetitive part — searching, comparing, negotiating, booking — and gives people their time back for the decisions that really matter. It's not automation to shrink the team; it's automation to help the people you already have work better.

We believe technology is the means and people are the end. An agent that buys ad space doesn't replace the judgment of someone who knows the brand, the market and the customer: it frees them from mechanical work to focus on strategy.

In our experience with companies running multichannel campaigns, the bottleneck is almost never strategy: it's the time the team loses on operational tasks that could be automated. AAMP tackles exactly that problem.

If you're evaluating how to integrate AI into your company's marketing processes, the starting point is understanding where you lose the most time today. We can help you do that with a structured approach: from strategic AI integration to business intelligence on campaign data, all the way to team training on process automation.

Frequently Asked Questions

AAMP isn't a future promise: it's the signal that the manual phase of media buying already has its days numbered — and those who understand how it works today will be ready when it becomes the standard.