Every feature of an AI buying system attracts attention at the evaluation stage: the model's reasoning capability, the harness's brief enforcement, the negotiation speed, the price performance. The feature that receives the least attention at evaluation and the most attention after something goes wrong is the audit trail. When a dispute arises about what was agreed, when it was approved, who authorised it, and whether delivery matched the committed terms, the audit trail is the only source of truth. If it is incomplete, the dispute cannot be resolved cleanly.

What is an audit trail in the context of agentic media buying?

An audit trail is a structured, time-stamped record of every consequential action taken during the lifecycle of a deal. In the context of agentic media buying, that includes every negotiation round, every proposed and counter-proposed term, every human approval or rejection, every amendment, and the final agreed position. It also includes the state of the campaign brief at the point each action was taken, so that any action can be evaluated against the brief that governed it.

An audit trail is not a reporting dashboard. A reporting dashboard shows current performance aggregated across many deals. An audit trail shows the history of a specific deal, action by action, from the first negotiation round to the final delivery confirmation. The two serve different purposes and neither substitutes for the other.

Why does the audit trail matter more when an agent is acting?

When a human trader negotiates a deal, the chain of accountability is short. The trader knows what they agreed, the counterpart knows what they were told, and any dispute is between two people who were present in the conversation. When an agent negotiates the deal, the chain of accountability is longer and less transparent. The buyer's organisation needs to know what the agent proposed, what it accepted, and whether the accepted terms were within the scope the agent was authorised to agree. The seller's organisation needs the same certainty about what was committed on their side.

Without a complete audit trail, neither side can answer these questions from the record. They rely on system logs that were not designed for dispute resolution, or on the memory of whoever reviewed the approval at the time, or on performance data that shows outcomes but not the agreed terms that should have produced them. Each of these is an inadequate substitute.

What does a complete audit trail need to contain?

A complete audit trail for an agent-negotiated deal needs seven things. First, the original campaign brief, preserved as it existed at the start of the negotiation. Second, every offer and counter-offer, with the timestamp of each exchange. Third, every brief-compliance check run by the harness and its outcome. Fourth, every approval gate event: who was presented with the decision, what information they saw, what they decided, and when. Fifth, any amendments made by a human reviewer between agent recommendation and execution. Sixth, the final agreed terms at commitment, signed or confirmed by both parties. Seventh, delivery confirmation against those committed terms.

Any gap in this sequence creates a window of ambiguity. If the audit trail shows negotiation rounds but not the brief-compliance checks, there is no way to verify that the agent operated within its permitted scope. If it shows the approval gate event but not what information was presented, there is no way to verify that the approval was informed. If it shows commitment but not delivery confirmation, there is no way to verify that what was agreed was what was delivered.

How does the DealSheet function as an audit trail?

Alkimi's DealSheet is designed from the ground up as an audit instrument, not a summary view. It records deal events as they occur, not as a batch export after the campaign ends. Each event is time-stamped and associated with the actor responsible, whether that actor is the buying agent, the selling agent, or a human approver. The record is bilateral: the buyer and seller access the same DealSheet, not separate versions of it. This prevents the most common form of post-deal dispute, which is each party reading a different record of the same negotiation.

The DealSheet captures the negotiation history, not just the final agreed position. A deal record that shows only the committed terms cannot tell you whether those terms resulted from a two-round negotiation that went smoothly or from a six-round negotiation in which the agent exceeded its authorised scope at round three and then revised back. The negotiation history is where the governance evidence lives.

Why is the audit trail relevant to compliance?

Media buying sits within a regulatory environment that is tightening around automated decision-making. Demonstrating compliance with agency obligations, advertiser instructions, and any applicable trading standards requires evidence that the buying system operated within its defined mandate. The audit trail is that evidence. An organisation that cannot produce a complete audit trail for an agent-negotiated deal cannot demonstrate compliance; it can only assert it.

This matters particularly for agency relationships, where the agent's buying activity is conducted on behalf of a third party. The advertiser is entitled to see the record of what was negotiated in their name. An audit trail that is complete and accessible to the advertiser is a transparency obligation, not just an operational convenience.

Why is the audit trail not optional?

Agentic buying systems are being deployed in environments where the volumes, speeds, and complexity of decisions exceed what a human team can manually track. That is precisely the point of deploying them. But that efficiency gain creates a governance gap if the system does not record what it does in a form that humans can review and rely upon. The audit trail closes that gap.

The organisations that will trust AI-negotiated deals at scale are the ones that can see, plainly and completely, what the AI negotiated and on what authority. The audit trail is not a feature added on top of an AI buying system. It is the foundation on which the trust that makes scale possible is built.