A 30-day agentic advertising pilot should answer one question: can an AI agent buy media within a written mandate, escalate what falls outside it, and leave a record of each deal that both parties agree with? Thirty days is too short to prove a lasting lift in campaign outcomes at pilot budgets, but it is long enough to prove or disprove control. The pilot that works is narrow in scope, starts with a person approving every deal, loosens authority one step at a time, and has stop criteria written down before the first pound is spent.

TL;DR. Scope the pilot to one channel, a short list of named sellers and a fixed budget, with a control budget bought the usual way. Write the mandate first, including minimum acceptable terms as well as maximum prices. Measure governance (mandate breaches, escalations, record agreement) alongside price and delivery. Move from recommendations in week one to bounded automatic action in week three only if the evidence supports it. Stop the pilot at the first spend outside the mandate or the first deal with no agreed record.

What should a 30-day agentic advertising pilot prove?

A 30-day pilot should prove that the agent can be governed, not that it outperforms a trading team. The evidence a pilot can produce in a month is operational: whether the agent stayed inside its mandate, whether it escalated the decisions it should have, whether the terms it agreed were acceptable, and whether buyer and seller ended the month holding the same version of each deal.

Performance comparisons still belong in the pilot, but as early indicators. A small budget over four weeks produces price and delivery differences that are suggestive at best.

How should the pilot be scoped?

A sound pilot covers one side of the market (most advertisers start on the buy side), one channel, and one or two deal types, usually preferred deals or private marketplaces, where terms are negotiated but no volume guarantee has to be policed. It names a short list of sellers in advance rather than letting the agent discover new ones, and it runs a control budget through the existing buying method with the same sellers, so the comparison is like for like.

Scope also covers people: a budget owner who approves the mandate, a campaign manager who handles escalations within a stated turnaround, and someone in finance to reconcile the pilot deals.

What goes into the pilot mandate?

The mandate is the written scope of the agent's authority, and in a pilot it doubles as the test specification. It should state the sellers, formats and flight dates in scope; the maximum price per deal type; minimum acceptable terms, such as a viewability floor measured against the MRC viewable impression guidelines of 2014, a frequency cap and a make-good clause; excluded content categories; which audience data the agent may use and share; and the approval thresholds for each week.

Minimum terms matter as much as maximum prices. An agent told only to keep price down will trade away the protections the brief needed.

Which thresholds should a pilot use?

Approval thresholds in a pilot should start at zero and rise only on evidence. In week one, a person approves everything. By week three, the agent may commit to deals below a set value on its own, while larger deals, new sellers, new data uses and any term below the mandated minimum still go to a person. Every escalation is logged with its reason and turnaround time.

What should a pilot measure?

A pilot should measure four groups of result, and the governance group decides whether the pilot passes.

Governance: mandate breaches (the target is zero), escalations by reason, approval turnaround, and the number of deals where buyer and seller records of agreed terms differ.
Terms: share of deals that met every minimum term, viewability floors agreed, make-good clauses included, and deal completion rate.
Delivery and price: delivered against agreed volume, measured viewability against the agreed floor, and effective CPM against the control budget.
Effort: hours of human time per deal for the agent budget and for the control, including reconciliation.

The record-agreement measure is the easiest to skip and the most telling. If finance cannot reconcile the pilot deals against one agreed set of terms at month end, the agent has created work, not removed it.

What are the stop criteria?

Stop criteria should be written before day one, so nobody argues about them while money is moving. The pilot pauses immediately, pending review, if any of the following happens:

the agent commits spend outside its mandate, of any amount;
a deal runs without a retrievable record of agreed terms;
the agent uses or shares audience data the mandate did not permit;
a brand safety incident occurs on inventory the agent bought;
buyer and seller records disagree on any deal and the difference is not resolved within five working days;
escalations go unanswered beyond the agreed turnaround, which is a failure of the human side, not the agent.

A pilot that stops on a clear criterion has still produced the evidence it was designed to produce.

What does the week-by-week plan look like?

Week one (days 1 to 7): set up and recommend. The budget owner signs the mandate, the counterparty agents are confirmed as authorised by the sellers they represent, and the control budget starts. The agent proposes deals and terms; people decide and execute. The measure is how often the agent's recommendations would have stayed inside the mandate.

Week two (days 8 to 14): negotiate with approval. The agent negotiates with the named sellers, and a person approves every deal before it is committed. Each approved deal is checked against the mandate's minimum terms.

Week three (days 15 to 21): bounded automatic action. If week two produced no breaches and acceptable terms, the agent may commit to deals below the threshold on its own. Everything above it still escalates. A first reconciliation of week-two deals tests whether both parties hold the same record.

Week four (days 22 to 30): hold and review. No further loosening of authority. Finance reconciles every pilot deal, the team compares the agent budget with the control, and the budget owner decides whether to extend, scale or stop.

What budget does a pilot need?

The following budget is hypothetical and illustrative. An advertiser sets aside £30,000: £15,000 for the agent and £15,000 for the control, bought the usual way from the same sellers. The agent's budget is capped at £1,500 in week one and £4,500 in each of weeks two to four, with no single deal above £3,000. From week three, deals below £1,000 may be committed without approval. The shape matters more than the amounts: a fixed total, weekly and per-deal caps, and an approval threshold, all in the mandate.

Where does the deal record fit?

The record-agreement measure needs a record to measure. Pilots run through Alkimi, a neutral agentic advertising marketplace working within IAB Tech Lab's AAMP framework, produce one DealSheet per deal, written at agreement, timestamped and held by both buyer and seller, while people keep approval rights at the deal level.

Frequently asked questions

Is 30 days long enough for an agentic advertising pilot?

Long enough to test control, escalation and record-keeping. Too short to prove a lasting performance difference at a pilot budget, so treat price and delivery results as early indicators.

Should a pilot include a control group?

Yes. Running part of the budget through the existing buying method with the same sellers is the only way to compare price, terms and effort fairly.

What is the single most important stop criterion?

Spend outside the mandate. If the agent commits beyond its authority even once, the pilot pauses until the cause is understood.

Can a pilot start with full agent autonomy?

It should not. A person approves every deal at first, and authority widens only after approved deals show no breaches.

How Do You Run a 30-Day Agentic Advertising Pilot?