- H2O Document AI reads police reports, estimates, statements, photos and correspondence into the claim facts that point to a responsible third party.
- An agent checks each claim against your subrogation guidelines and flags recovery potential, citing the document behind every flag.
- Subrogation evidence judgment lists what supports a recovery, what is missing and which deadlines your team has to watch.
- Your recovery specialists and counsel decide whether to pursue a claim; the agent drafts the file and the demand.
Recovery money sits in files nobody reread
A claim closes when the policyholder is paid. Whether a third party should pay you back depends on details buried in the file: a line in the police report, a remark in a recorded statement, a part number on a repair estimate.
Adjusters are measured on closing claims, not on spotting recoveries, and a busy adjuster can miss the sign that another driver, a contractor, a landlord or a faulty product caused the loss. By the time someone looks again, evidence has gone and deadlines are close, and a recovery team ends up working the obvious cases and never sees the rest.
How subrogation identification works
H2O can be used to read every claim file for recovery potential, judge the evidence, and hand your recovery team a file ready to act on, sitting on top of the claims system you already run.
Step 1: read the claim file
Claim documents enter an Enterprise h2oGPTe collection by upload, a connector to a shared repository or the API, which reads each page, including scans, tables, photos and handwriting.
Step 2: extract the liability facts
H2O Document AI extractors turn the file into structured output that matches a schema your team defines: parties, other carriers, vehicles or products involved, cause of loss, statements of fault and amounts paid.
Step 3: flag recovery potential
Agentic RAG reads the claim facts against your subrogation guidelines and flags claims with a possible responsible third party, each flag citing the document and the guideline behind it.
Step 4: judge the evidence
The agent sorts the evidence into what supports a recovery and what is missing, such as a repair invoice or a witness statement, and notes the deadlines your team tracks for each claim.
Step 5: draft the recovery package
AI agents assemble the recovery file: a summary, an evidence index with citations, the amounts paid, and a draft demand letter. Your recovery specialist edits and approves it.
Correspondence becomes structured facts too
Recovery work creates its own paper: demand letters, other carriers' replies, counteroffers, arbitration filings and litigation correspondence. Document AI reads each reply into structured facts such as the amount offered, the position taken and the next date that matters, so your team sees where every open recovery stands without reading the whole thread.
Your recovery specialist decides whether to pursue, settle or close each case, and counsel owns anything that turns on the law.