A make-good in an agent-negotiated deal is the same remedy it has always been: what a seller owes when it misses a delivery commitment. What changes when AI agents negotiate is that the commitment, the trigger and the remedy were agreed by software, often in seconds, and may exist only in each agent's own logs. A make-good is enforceable only if four things were fixed at agreement and recorded where both parties can see them: what triggers it, what form it takes, the deadline for delivering it, and who has authority to accept it.

TL;DR. A make-good is triggered when a named metric, measured by a named party, falls below a stated threshold over a stated period; anything vaguer becomes a negotiation after the fact. Agents can propose and accept make-goods within limits their mandates set, but credits, refunds and substitute inventory should go to a person. Each make-good should be recorded as a new, timestamped version of the original deal, held by both parties. Neither AAMP nor AdCP sets a binding make-good process, so the make-good lives in the deal terms.

What is a make-good in advertising?

A make-good is the remedy a seller provides when it fails to deliver what a deal promised. The practice comes from broadcast television, where a channel that under-delivered a guaranteed audience added airtime to cover the gap. In digital deals, the remedy is usually additional impressions, a credit against future spend, or a refund. Make-goods attach most naturally to programmatic guaranteed deals, where the seller has committed to volume or quality in advance. An open auction carries no commitment to fall short of.

What triggers a make-good in an agent-negotiated deal?

A make-good is triggered by a measured shortfall against a commitment written into the deal terms. A trigger that works has four parts: the metric (delivered impressions, viewable impressions, completion rate or invalid traffic), the measuring party (the seller's ad server, the buyer's ad server or a named verification vendor), the threshold, and the measurement window (daily, weekly or the whole flight).

Common triggers are under-delivery, viewability below an agreed floor measured against the MRC viewable impression guidelines of 2014, invalid traffic above an agreed level, and delivery outside the agreed placements. "Viewability below expectations" triggers nothing. "Viewable rate below 65%, measured weekly by the named verification vendor" triggers a make-good the moment the weekly report lands.

The weak point in agent negotiation is the threshold itself. In Concourse Bench v1, a benchmark of AI buying agents negotiating with software sellers, 95 of the 110 contracts signed promised a viewability floor below 70%. A low floor means a make-good rarely fires, however well the clause is drafted, so buyers should set minimum floors in the mandate that the buying agent cannot trade away for price.

Who agrees a make-good when agents negotiated the deal?

Two different agreements are involved. The first is the make-good clause, which buyer and seller agents negotiate within their mandates and which is approved with the rest of the deal.

The second is the make-good itself, agreed when a trigger fires. Here the mandate should draw a clear line. A buying agent can reasonably be allowed to accept additional impressions on the same placements, inside the flight, up to a stated share of the original volume. Anything else should go to a person: a credit or refund, which moves money; substitute inventory, which changes what was bought; or delivery after the flight, which changes when it runs. The same applies on the sell side, where a credit commits the publisher's revenue. An agent is not a legal entity, so a make-good it accepts or offers binds the organisation that deployed it.

How should a make-good be recorded?

A make-good should be recorded as a new, timestamped version of the original deal record, linked to the version it amends and held by both parties. Overwriting the original terms leaves finance unable to show which terms applied to which impressions.

A complete make-good entry holds the original deal identifier, the trigger that fired and the measurement report behind it, the shortfall as calculated under the deal terms, the form of the remedy, the placements and deadline, and the approver on each side with the time of approval. A closing entry records the measured outcome. If those fields sit in one record both sides can retrieve, the make-good is an auditable fact. If each agent logs its own version, it is the start of a dispute.

Do AAMP or AdCP standardise make-goods?

Neither framework sets out a make-good process both parties are bound by. AAMP, IAB Tech Lab's Agentic Advertising Management Protocols, standardises agent communication, how publishers describe products through OpenProposal, and how agents work across deal types through the Deals API. It does not by itself give buyer and seller one shared record of the commercial terms a make-good is calculated against.

AdCP, the Ad Context Protocol from AgenticAdvertising.org, is a separate protocol. Its documentation states that payments, legal agreements, settlement, disputes, credits and make-goods remain account or platform integrations unless the relevant AdCP contract is explicitly present. Under either framework, the make-good clause belongs in the deal terms, and its record belongs with the marketplace or the contract between the parties.

What does a make-good look like in practice?

The following is a hypothetical, illustrative example. A buyer agent agrees a £24,000 deal with a seller agent: 2,000,000 impressions at a £12 CPM over four weeks, with a 65% viewable rate measured weekly by a named verification vendor. The make-good clause provides additional impressions on the same placements, delivered within the flight or up to 14 days after it. The buyer's mandate lets its agent accept make-goods of up to 10% of deal volume on the same placements; anything else goes to the campaign manager.

In week two, 500,000 impressions run at a 58% viewable rate. The deal promised 325,000 viewable impressions for that volume and delivered 290,000, a shortfall of 35,000. The seller agent proposes about 54,000 additional impressions on the same placements, enough to cover the shortfall at the agreed viewable rate. That is under 3% of deal volume, so the buyer agent accepts, and the deal record gains a new version showing the trigger, calculation and remedy.

Now suppose the placements are sold out and the seller agent offers a credit of about £646 instead. A credit is outside the buyer agent's authority, so the offer goes to the campaign manager, who approves it with finance, and the record shows who decided and when.

How does a shared deal record help?

A make-good is a calculation against agreed terms, so it is only as reliable as the record of those terms. Alkimi, a neutral agentic advertising marketplace working within IAB Tech Lab's AAMP framework, records each agreement on a DealSheet that is written at agreement, timestamped and held by both parties, so either side can retrieve the clause a make-good is measured against, and people keep approval rights at the deal level.

Frequently asked questions

Can an AI agent agree a make-good on its own?

Yes, within its mandate. A buying agent can accept like-for-like impressions inside the flight up to a set share of volume, but credits, refunds and substitute inventory should be approved by a person.

What should a make-good clause include?

The trigger (metric, measuring party, threshold and window), the form of the remedy, the deadline for delivering it, and who on each side may accept it.

Do open auction buys get make-goods?

Generally not. Make-goods attach to commitments, so they belong in deals where the seller has promised volume or quality, most often programmatic guaranteed.

Does AdCP handle make-goods?

Not by default. The AdCP documentation leaves make-goods, credits, settlement and disputes to account or platform integrations unless a specific AdCP contract covers them.

How Do Make-Goods Work When an AI Agent Negotiated the Deal?