2 Nov 2026 · 3 min read
What Happens When an AdCP Deal Goes Wrong
Every explainer about agentic media buying describes the happy path: agent finds inventory, agent negotiates terms, deal closes, campaign runs as planned. Worth asking the less comfortable question too, because this is advertising, and advertising campaigns under-deliver, get miscounted, or get disputed often enough that "what happens when it goes wrong" matters more than the sales deck version of the story.
The scenario
A buying agent agrees a deal with a publisher's agent: £20,000 for CTV inventory at a minimum 75% viewability, two-week flight. The AdCP exchange that produced this agreement is clean, both systems have a structured record of what was proposed and accepted. The campaign runs. Three weeks later, the buyer's measurement partner reports delivered viewability at 61%, well under what was agreed.
How this used to play out
Two account teams, each holding their own version of the agreement, usually in an email thread or a signed insertion order neither side's ad server actually enforces automatically. The buyer's side argues for a make-good or a refund. The publisher's side argues the measurement methodology differs from what they track internally, or that external factors affected delivery. Neither side has a single, shared, timestamped record of exactly what was promised, so the dispute becomes an argument about whose version of events is more credible, which usually gets resolved by whoever has the stronger negotiating position rather than by what was actually agreed.
Why a clean AdCP exchange doesn't prevent this
The protocol did its job. The proposal was structured, the acceptance was structured, both systems logged the exchange correctly. None of that creates a single record both parties independently trust after the fact, because each side's log is still just their own internal copy. A structured message format solves ambiguity in the conversation. It doesn't solve disagreement about the conversation's outcome once both sides have walked away with their own version of it.
What a shared deal record actually changes
A DealSheet, written once when terms are agreed and held by both sides as the same cryptographically verifiable document, removes the "whose version is right" question entirely, because there's only one version. The 75% viewability floor isn't a term one side remembers and the other disputes, it's a line in a record both sides signed off on at the time. When delivery comes in at 61%, the comparison is mechanical: agreed term versus measured outcome, not a negotiation about what was actually promised three weeks ago.
The part that still needs a human
None of this makes the underperformance itself disappear. Inventory can genuinely under-deliver for reasons outside anyone's control. What a shared record changes is how fast that gets resolved and how much of the resolution depends on who has the stronger relationship rather than facts both sides already agreed to. The human checkpoint still matters here too, for a deal this size, someone should have seen and approved the terms before they went live, so if the flight does need to be renegotiated or made good, there's a person who understood the original agreement well enough to act on it quickly rather than starting from a cold email thread.
A clean negotiation and a clean dispute resolution are two different problems. AdCP, or whatever protocol got the deal agreed in the first place, only ever solved the first one.