A Triage Score Near a Gold Thing

In practice

An insurer's claims team wants a triage model that scores each new claim for complexity, so that simple claims move faster. The analyst, Dev, gives the approved AI tool the Claim Decision's Gold Pack and a three-sentence description of the feature, and asks the book's four questions, using only the Pack and the description. The tool answers that the score could influence which handler receives the claim and how quickly it is decided. It names a risk the team had not discussed: if handlers see the score, a low complexity score could be read as a low-value claim. It suggests using the Pack's definition of the Claim Decision as a guardrail: the decision's outcome and reasons must come from the policy clauses, never from the score. Its questions before deploying include who sees the score, whether a handler can override the routing, and how the team will check scores across different kinds of claimant. Dev reviews the draft. He removes one suggested risk about fraud, which the description never mentioned, and keeps the rest. The team decides that the score will fill a Claim Assessment field that routes work, and will not appear on the decision screen.