Before an AI agent spends any media budget, it needs a written mandate: a fill-in document that states the objective, the money, the price limits, the permitted inventory and data, the approval points and the conditions under which it must stop. The template below sets out every field, each with a one-line explanation and a hypothetical example value, ready to copy, fill in and hand to an agent.

TL;DR: A media buying mandate turns a campaign brief into limits an agent can be checked against. This template has ten sections, from objective to owner and review date, each written so a reviewer can compare it with the deal record. The same mandate should run in a sandbox against counterparty agents before it governs live spend.

The concept is explained in What Is a Media Buying Mandate? and How to Write a Media Buying Mandate for an AI Agent; this page is the working document.

How can AI agents help with media buying, and why do they need a mandate first?

AI agents can take a brief, find suitable inventory, negotiate with seller agents and assemble a plan, removing much of the repetitive back and forth of a media buy. What an agent cannot do is guess the limits its principal had in mind, and its actions can only be audited against limits that are written down.

The standards bodies make the same point. IAB Tech Lab lists "built in governance and trust" among the capabilities of AAMP 2.3, part of its Agentic Advertising Management Protocols (AAMP). AgenticAdvertising.org, which publishes the Ad Context Protocol (AdCP), governs through a principle it calls Embedded Human Judgment: "humans stay in the loop on decisions with real consequences".

Can AI agents be trusted to manage advertising budgets?

AI agents can be trusted with advertising budgets to the extent that their limits are explicit and their actions can be checked against those limits. Trust here is a property of the set-up, not of the model.

Buyers remain cautious. A Digiday+ Research survey of more than 100 marketing professionals, run in the first quarter of 2026, found that only 32% of marketers using AI in their social or retail media campaigns used it to buy that inventory.

Completing a buy is also not the same as protecting the buyer. In Alkimi's Concourse Bench v1, a controlled benchmark of AI buying agents, 55 of 96 runs completed a buy, yet 95 of 110 signed contracts promised a viewability floor below 70%. A viewability floor written into the mandate, rather than held in someone's head, is a term the agent has to defend and a reviewer can check.

How do agencies use AI for programmatic buying today?

Agencies are running agents inside defined limits, with people signing off wherever money or activation is involved. Lauren Wetzel, global president of data and technology solutions at WPP Media, told Digiday in October 2026: "particularly where there's a financial or activation impact... We require explicit controls, approval thresholds, accountable human oversight."

In the same report, David Dworin, chief product officer at FreeWheel, said: "Most agentic systems are either designed with a human that has to approve everything, or very specific controls that can keep it from getting into trouble." A mandate template writes those controls down.

Monitoring matters as much as the controls. Digiday reported in September 2026 that the performance media agency Rise uses an audit log from PubMatic to see when agents drift outside pre-set parameters. Jonathan Whiteside of Dept put it plainly in the same piece: "Every deliverable has an accountable human."

What goes in a media buying mandate template?

Every example value below is hypothetical, written for an imagined UK homeware retailer running a six-week video campaign. Replace each one with a value a reviewer could test against the deal record.

1. Objective

Campaign objective: the single outcome the agent is buying towards, in one sentence. Hypothetical example: build awareness of the autumn range among UK adults aged 25 to 44.
Primary measure: the one metric that decides whether a deal fits the objective. Hypothetical example: completed video views.

2. Budget and pacing

Total budget: the most the agent may commit across the whole campaign. Hypothetical example: £120,000.
Flight: when delivery starts and ends. Hypothetical example: weeks one to six of the autumn trading period.
Pacing rule: how spend should spread across the flight. Hypothetical example: even weekly delivery, with no week above 20% of total budget.
Maximum single commitment: the largest amount the agent may commit in one deal. Hypothetical example: £30,000.

3. Price limits

Maximum CPM: the highest cost per thousand impressions the agent may agree. Hypothetical example: £24 for video.
Minimum CPM: the price below which the agent treats an offer as suspect rather than a bargain. Hypothetical example: £6.

4. Inventory and environments

Channels and formats: where the agent may buy. Hypothetical example: connected TV and online video, 15 and 30 second spots.
Approved sources: the publishers, packages or lists the agent may buy from. Hypothetical example: inclusion list v3, held as Appendix A.
Seller checks: what a seller must pass before an offer is considered. Hypothetical example: authorised in the publisher's ads.txt file and identifiable in sellers.json.
Deal types: the structures the agent may agree without escalation. Hypothetical example: private marketplace and programmatic guaranteed only.
Viewability floor: the lowest viewability the agent may accept in the contract. Hypothetical example: 70%.

5. Audience and data permissions

Permitted data: the audience data the agent may use or request. Hypothetical example: the retailer's own first-party segments and contextual signals only.
Lawful basis reference: the documented basis for using each data source. Hypothetical example: privacy notice v7, signed off by the data protection lead.
Prohibited data: what the agent must never request or combine. Hypothetical example: special category data and precise location.

6. Brand safety

Excluded content: the content the brand will not appear against, named in a shared vocabulary. Hypothetical example: exclusions mapped to the IAB Tech Lab Content Taxonomy.
Verification source: whose measurement settles a disagreement about where ads ran. Hypothetical example: the advertiser's contracted verification vendor.

7. Approval thresholds

Autonomy stage: where the agent sits on the earned autonomy scale of Observe, Recommend, Draft, Human-approved action and Bounded automatic action. Hypothetical example: Human-approved action.
Approval triggers: the conditions that pause the agent until a named person approves. Hypothetical example: any deal above £15,000, any seller not on the inclusion list, any CPM above £20.
Approver and wait time: who signs off and how long the agent holds. Hypothetical example: senior trader, within four working hours.

8. Reporting

Deal record: what must be recorded for every deal the agent agrees. Hypothetical example: final terms, both parties' identifiers, a timestamp and the mandate version.
Decision log: the reason the agent gives for each offer, counter and refusal. Hypothetical example: one line per action, exportable.

9. Stop conditions

Hard stops: events that halt the agent entirely rather than trigger an approval request, and who may restart it. Hypothetical example: committed spend reaches the total budget; only the mandate owner may restart.
Record mismatch: a stop that fires when the buyer's and seller's records of a deal disagree. Hypothetical example: any difference in agreed price, volume or dates.

10. Owner and review date

Mandate owner: the person accountable for the mandate and its outcomes. Hypothetical example: head of programmatic.
Version and effective date: the identifier recorded against every deal. Hypothetical example: v1.2, effective from the first day of the flight.
Review date: when the mandate is next checked, whatever the results. Hypothetical example: the end of week two.

How should a mandate be tested before it controls live spend?

A mandate should run in a sandbox before it governs real money: give the agent the mandate and a realistic brief, let it negotiate against counterparty agents that push on price, volume and terms, and check every simulated deal against every field.

Three results matter: whether the agent stayed inside the price limits and the viewability floor, whether it paused at each approval trigger, and whether it stopped when a stop condition fired. Any field the agent ignored in the sandbox needs rewording before launch. How to test an AI media buyer before you deploy one sets out the wider testing method.

The sandbox tests the mandate as much as the agent: an agent will read a vague field such as "premium inventory" in whatever way a counterparty encourages.

What happens to the mandate once the agent goes live?

Once live, the mandate becomes the reference for every deal review, which works only if both sides keep the same record of what was agreed. Alkimi offers one route for both steps: its Gym, available now, lets buyers and sellers build their own agents, set the rules those agents work to and spar against counterparty agents before any real money moves. Its Marketplace works within IAB Tech Lab's AAMP framework today, is being built to be AdCP compatible, keeps human approval rights at the deal level and writes each agreement to a DealSheet that both parties hold and can retrieve independently.

Frequently asked questions

What is a media buying mandate template?

A media buying mandate template is a fill-in document listing every limit an AI buying agent must work within, from budget and price to data, brand safety and stop conditions. Filling it in turns a campaign brief into terms that can be checked against each deal.

Can AI agents be trusted to manage advertising budgets?

They can be trusted to the extent that budget, pacing and approval limits are written down and every commitment is recorded. Without those limits there is nothing to hold the agent to and no way to check its decisions afterwards.

Who should own a media buying mandate?

One named person, usually a senior trader or the head of programmatic, should own the mandate and sign off every change. Finance, data protection and brand safety leads should review the sections that affect them before the agent goes live.

Does the template work with AAMP and AdCP?

The template describes the buyer's limits rather than a protocol message, so it can sit behind an agent built to AAMP or to AdCP. Teams then map each field to the relevant protocol settings, such as deal types and approval steps, when they configure the agent.

Media Buying Mandate Template: What to Give an AI Agent Before It Spends