An agency running AI buying agents should send each client a weekly report in seven sections: spend and pacing against the mandate, deals agreed and their terms, approvals and overrides, exceptions and escalations, deal-record reconciliation, quality measures, and what the agent will do next week. A standard campaign report shows what the media delivered. An agent report also has to show what the agent committed the client to, under which rules, and which person signed it off.

TL;DR. The weekly report for agent-bought media is an accountability document first and a performance document second. Every line should point to a source the client could check: the mandate, the deal record, the approval log or the delivery data. Figures that are still provisional belong in pacing, not in reconciliation, and a short monthly add-on reviews the mandate, the agent's autonomy and the finance close.

How are agencies using AI agents for programmatic buying?

Agencies are moving AI agents from planning support into live buying, and most pair that move with human sign-off. In October 2026, Digiday reported a run of agency launches around Advertising Week, with executives working to reassure clients that agents would not act unchecked. "We require explicit controls, approval thresholds, accountable human oversight," said Lauren Wetzel, global president of data and tech solutions at WPP Media, speaking about decisions with a financial or activation impact.

Client reporting sits inside that oversight rather than outside it. In the same piece, Rebekah Shalit, head of platforms at Dentsu, named billing and campaign reporting among the tasks that still require human oversight.

Logging is becoming the practical starting point. Digiday reported in September 2026 that Rise, part of the Quad agency group, had been testing AI media buying agents with a supermarket client since June, using an audit log feature developed by PubMatic to monitor when agents drift outside pre-set parameters. "I can make sure everything is recorded and that we can reconstruct what happened and why," said Klaudia Smykowska, group director at Rise.

The same article reported that Dept is developing monitoring tools that produce decision logs and quality control reports in a digestible format. "Every deliverable has an accountable human," said Jonathan Whiteside, global EVP of technology at Dept.

Why does an AI agent report need sections a campaign report does not?

An AI agent report needs extra sections because the agent makes commitments, not only optimisations. A human trader's deals usually arrive as insertion orders the client can read, while an agent can agree many smaller deals, each with its own price basis, viewability floor and makegood rule, and amend them mid-flight. The client therefore needs to see the terms, the authority behind them and the record they live in.

A decision log held by the agency explains why the agent acted, but it does not prove what the seller agreed. That second question needs a record both parties hold.

Alkimi offers one concrete example of where such a record can come from. On its Marketplace, each deal produces a DealSheet that is written at agreement, timestamped, held by both parties and retrievable by either side independently, and people keep approval rights at the deal level. In its Agree. Transact. Verify. simulation, 95.3% of deals ended with mismatched records when each side kept its own, against 0.19% when both sides used a shared record.

What should a weekly AI agent client report include?

The weekly report has seven sections, and each should name its source of record so the client can check it without asking the agency. The table summarises the structure; the field lists below it form the template.

Section	The client's question	Source of record
1. Spend and pacing against the mandate	Is the agent spending what it was allowed to, at the right speed?	Mandate and delivery data
2. Deals agreed and their terms	What did the agent commit to this week?	Deal record held by both parties
3. Approvals and overrides	Who signed off, and where did people step in?	Approval log
4. Exceptions and escalations	What went wrong or needed a person?	Agent decision log
5. Deal-record reconciliation	Do the agency, the seller and the delivery data agree?	Deal record and delivery reports
6. Quality measures	Did the media meet the contracted standard?	Independent verification data
7. Next week	What will the agent do, and what must the client decide?	Agent plan and mandate

1. Spend and pacing against the mandate

Total budget in the mandate, spend to date, spend committed but not yet delivered, and budget remaining.
Pacing against plan for each deal or package, with any variance explained in one line.
Spend outside mandate limits (expected to be zero), and any limit changed this week, with its approver.
The mandate version in force, so every figure ties to the rules that applied.

2. Deals agreed and their terms

This section lists every deal the agent agreed or amended during the week. Since AAMP 2.0, IAB Tech Lab's Agentic Advertising Management Protocols have supported programmatic guaranteed, preferred deals, private marketplaces and the open marketplace through the Deals API, so the deal type field can use those familiar names.

Deal ID, deal type, seller, and the seller agent's verified identity.
Volume, flight dates, formats, targeting and price basis.
Measurement terms: viewability floor, brand safety rules, which party's count governs billing, and the makegood rule.
Amendments, each with the old value, new value, time and approver.
A link to the deal record that the client can open directly.

3. Approvals and overrides

Deals approved by a person: who approved, when, and under which threshold.
Agent proposals rejected or amended, with the reason given.
Overrides: changes people made to the agent's live decisions, and why.
The autonomy stage for each task on an earned autonomy scale: Observe, Recommend, Draft, Human-approved action, Bounded automatic action.

4. Exceptions and escalations

Each escalation: the triggering rule, who received it, how long it waited and the outcome.
Actions the agent attempted that the mandate blocked.
Negotiations abandoned, identity checks failed and counterparty errors.
Open items carried into next week, each with a named owner.

5. Deal-record reconciliation

This section checks that the agency's records, the seller's records and the delivery data describe the same deals. Delivery figures need a status flag, because the AdCP documentation for get_media_buy_delivery states that delivery rows not yet marked final are safe for pacing and reporting but not for "reconciliation, accruals, or finance close". The simulation cited above suggests mismatches are the norm when each side keeps its own log, so this section should report them rather than assume none.

Deals checked, and deals with any mismatch by dimension: price basis, volume, flight dates, targeting or measurement terms.
Provisional or final status for each delivery figure.
The source that governs billing for each deal, as written in its terms.
Unresolved discrepancies, the spend affected and the owner.

6. Quality measures

This section reports delivered quality against each deal's contracted terms, not against a general average. Viewability lines should name the standard used: the Media Rating Council's guidelines count a display impression as viewable when at least 50% of its pixels are in view for at least one continuous second, and a video impression when two continuous seconds play.

Viewability measured against each deal's contracted floor, with the measurement provider named.
Brand safety and suitability: blocked impressions, incidents and share of spend in low-risk environments.
Invalid traffic rate, and completion rate for video.
Any deal where delivered quality fell short of its terms, with the makegood status.

7. What the agent will do next week

Deals the agent plans to renew, renegotiate, pause or end.
Budget moves proposed and awaiting approval.
Mandate changes the agency is requesting, with the reason.
Proposed changes to any task's autonomy stage, up or down.
Decisions the client needs to make, each with a deadline.

What should the monthly add-on cover?

Once a month, the report should step back from individual deals to the system that produced them.

Mandate review: the versions in force during the month, the changes made and who approved them.
Autonomy review: which tasks moved stage, and the evidence for each move.
Finance close: a three-way match of agreed terms, delivery and invoice, as set out in How Should Finance Teams Reconcile Agent-Negotiated Media Deals?
Counterparty summary: the sellers and seller agents dealt with, and identity checks passed or failed.
Agent configuration changes: model version, tools and instructions, each with a date.
Incidents and lessons: what failed and what changed as a result.

How can a client audit what an AI agent agreed?

A client can audit an agent's deal only if the weekly report links each deal to a record the client can retrieve independently. The agency's decision log explains the agent's reasoning, while the agreed record shows the terms both sides accepted. The step-by-step procedure is covered in How do programmatic buyers audit what was agreed in a deal?

How can agencies test the report before real money moves?

The template is only as reliable as the data behind each field, and the easiest time to find a missing field is before launch. A sparring environment lets an agency check that every section populates against a realistic counterparty. The Alkimi Gym, available now, lets buyers and sellers build their own agents, set the rules they work to and spar against counterparty agents before any real money moves, and its Marketplace works within IAB Tech Lab's AAMP framework today and is being built to be AdCP compatible.

Frequently asked questions

How do agencies use AI for programmatic buying?

Agencies use AI agents to plan, negotiate and optimise programmatic buys, usually with approval thresholds and human sign-off on decisions that commit budget. Several are also adding audit logs so every agent action can be reconstructed for the client.

What KPIs should marketing teams track for AI-agent media buying?

Alongside media measures such as viewability, brand safety and completion rate, teams should track agent measures: spend inside the mandate, approval and override rates, escalations and record mismatches. The agent measures show whether the automation is behaving, which campaign metrics alone cannot show.

How do programmatic buyers audit what was agreed in a deal?

Buyers audit a deal by comparing the agreed terms with delivery data and the invoice, using a record of the terms that both parties hold. A decision log kept by one side explains its reasoning but cannot settle a disagreement with the other side.

How often should agencies report on AI agent activity?

Weekly reporting suits most live agent programmes, because agents can agree and amend deals faster than a monthly cycle can review them. A monthly add-on then covers mandate changes, autonomy stages and the finance close.

AI Agent Reporting for Agencies: What Clients Should See Every Week