When an AI buyer reports completion on a campaign buy, a planner sees a deal state. They do not, automatically, see a brief state. At the pace that agentic buying now moves, those two things can separate — and the gap may not surface until the campaign window is already closing. This is not a theoretical failure mode. It was recorded.

TL;DR: In 55 completed buys recorded for Concourse Bench v1 (concourse.agency), one model-driven buyer signed two valid contracts and returned a completion state. Net forecast reach was 124,511. The brief target was 137,156. The shortfall was 9.2%. A second finding: 95 of 110 contracts negotiated across all 55 completed buys had viewability floors below 70%. Both problems sit behind the word 'completed.'

The buy that closed but did not finish

The buyer was operating in a firmer-seller configuration: sellers held their pricing positions more firmly than in looser variants tested. The model bought two packages. Both deals were valid. Both contracts were signed. The harness registered no technical failure, no rejected offer, no disputed term. Net forecast reach across the two committed packages was 124,511. The gap against the brief target of 137,156 was 12,645, approximately 9.2%. A completion metric would not surface this. Each deal completed. The brief did not.

What completion looks like when a buy satisfies the brief

A reference buy from the same study illustrates the contrast. A GPT-6 Astra model, operating against the same brief and budget of £19,528.69, committed £18,706.89 and assembled a portfolio with a gross forecast reach of 149,565. After accounting for 8,709 audience overlap, net forecast reach was 140,856, clearing the 137,156 target. Both buyers completed their transactions at the deal level. Only one satisfied the brief.

The second problem that hides behind 'completed'

Reach shortfall is one failure mode that completion state obscures. There is a second. Across all 55 completed buys in Concourse Bench v1, 95 of 110 contracts had viewability floors below 70%. These are negotiated contractual promises, not catalogue values. A buy that hits its reach target may still carry this problem. If the evaluation layer stops at completion, neither gap surfaces in the system record. Two failure modes. One word covering both.

What a planner needs to check after an AI buyer reports done

When an agentic buying system returns a completion state, a campaign planner is working with a deal-level signal. The brief-level question requires a separate check covering at least three things. First: did the aggregate of committed packages, adjusted for audience overlap, clear the net reach requirement? No individual deal result answers this. Second: what percentage of contracts carry viewability floors at or above the threshold the brief implies? A portfolio built from low-floor contracts is weakly protected regardless of whether reach was achieved. Third: does whatever the system surfaces as its output distinguish between deal state and brief state? If the only signal is a completion indicator, it is measuring the wrong variable.

Why the timing matters

In conventional planning, the gap between deal and brief is caught quickly. A planner checks the summary sheet; the reach figure is short; the brief goes back before budget is committed. Agentic buying closes deals quickly. A model-driven buyer that returns a completion state early creates a system record that reads as successful. If the evaluation layer is not tracking brief-level outcomes separately, the shortfall does not appear until post-buy analysis, after the budget is locked and the campaign window is contracting. That is a structural feature of how completion is currently reported. The evaluation question is not how many deals completed. It is whether the system distinguishes between 'deals closed' and 'brief satisfied' — and whether that distinction is visible to a planner before the campaign window closes. Full data: concourse.agency.