Claims triage & extraction estimate

Context

An estimate for an insurance claims-triage and document-extraction build. The modeling is routine; the risk lives in data-readiness on heterogeneous claim documents and in the eval harness that proves extraction is accurate enough to act on.

The decision

Price the document long tail as a line item: data-readiness on non-standard claim forms and the eval harness are where insurance AI estimates blow up, so surface them as their own WBS rows and present the PERT range, not the point.

What most miss

Estimators anchor on the extraction model and treat document variety as a detail. In claims, the document long tail is the schedule — it's a data-readiness problem wearing a modeling costume.

Stakes

Bury data-readiness in 'build' and the estimate is weeks short before the first messy claim form arrives.