Finding the money left in closed claims: subrogation detection

The problem

When another party is liable for a claim, the insurer can recover what it paid — subrogation. But at intake volume those opportunities get missed, and once a claim is closed, nobody goes back to look. The result is recovery dollars leaking out the back door, invisible because closed claims are, by definition, done.

The hypothesis

The evidence of missed recovery is already sitting in the closed-claim file — the notes, the loss description, the police-report fields. Our hypothesis: a model could re-score closed claims for recovery potential, and a document agent could assemble the demand package, so a recovery specialist reviews ranked opportunities with the paperwork already drafted instead of hunting.

The build

• Recovery-signal model — scores closed claims on liability indicators drawn from notes, FNOL fields, and structured loss data.
• Document AI extraction — pulls the liability-relevant facts from the file and maps them to what a demand requires.
• Demand drafting — the agent assembles a demand package for a recovery specialist, who decides whether to pursue. The system finds and drafts; it never files.

Rollout

We back-tested the model against known historical recoveries to prove it caught what specialists would have, then ran it over a backlog of closed claims with specialist review of every flag. Coverage expanded line by line as accuracy held.

Results

Leakage that used to disappear at claim closure became recovered dollars, and specialist time shifted from digging through files to actually pursuing recovery — the part of the job that returns money.

What we'd tell you

• Closed claims are an asset, not a graveyard. The recovery signal is already in them.
• Keep pursuit human. The system's job is to find and draft, not to file.
• Measure recovered dollars, not flags raised — that's the number the business feels.