Insurance claims document processing

Claims intake is the front of the queue

A claim cannot move until someone has read what came in: the loss notice, the estimate, the police report, the photos of a repair invoice. In claims, a wrong value costs more than a slow one. A misread policy number or estimate total sends the claim down the wrong path, and someone has to unwind it later. docRefinery is built for that trade-off. It extracts what it can stand behind and holds back what it cannot.

What a claim file has to yield

Repair invoices are in the 20-type catalogue. First notice of loss forms, ACORD forms and declarations pages are set up as a workflow or a custom type, trained at no charge.

Loss notices, estimates and the systems that print them

Claims look varied, but the paperwork behind them is not. A carrier receives its own forms back, plus industry-standard loss notices such as ACORD forms from agents and brokers. Repair estimates come from a handful of estimating systems that most body shops and contractors use, and police reports follow each force's own form. The same layouts come back constantly. Each is learned once and remembered, so a carrier's thousandth estimate from a familiar system is a known layout, not a fresh problem.

Confidence thresholds that hold claims back

You set a confidence threshold per field and per document type. Set the policy number and estimate total strict and the loss description loose. A field that clears its threshold is ready to use; one that does not waits for a person. A packet uploaded as one PDF is separated into its documents first, so a strict threshold on an estimate does not hold up the loss notice beside it. See confidence scores in document extraction for how to choose thresholds.

Workflow rules add checks on the values within a document, such as estimate lines summing to the total or the date of loss falling before the report date. A failed rule flags the run. A field the form leaves blank comes back null, not a plausible guess.

Coverage calls stay with adjusters

Coverage and settlement decisions stay with adjusters. In the review queue they see the source page next to each held value, with the cited region highlighted, and correct values or add and remove estimate lines. Every change is kept on the run, and exports always carry the corrected values. docRefinery reads documents. It does not assess damage from photos, decide coverage or detect fraud.

What else claims teams can lean on

Claim files are long and mixed. These are the parts of docRefinery that help most.

• Long packets at scale. A file of loss notice, estimate, report and invoices is segmented into its documents, each recognised for what it is.
• One format across the claim. A loss notice, an estimate and a police report all return the same field and table shape.
• Claim system names. A Return as alias sets each output key, such as policy_no or loss_date, so values land in the right place.
• Estimate lines as rows, with labour, parts and materials kept apart rather than as one block of text.
• Forms outside the catalogue. ACORD forms and declarations pages become custom types, trained, hosted and re-trained at no charge.
• Safe changes. A workflow starts from one sample document, and versions, replay and compare let you test a change before it goes live.
• Handlers' fixes reused. A correction on one estimate can be offered as a fix for that whole layout, once it has been tested on that layout's own documents.
• A view of the queue. Insights tracks flagged fields and auto-approval trends, so you can see which forms hold claims back.

Writing intake data to the claim

Your code takes the JSON from the API (extract a document) and writes it to the claim, or a handler exports CSV, Excel or JSON from the dashboard. Every document type arrives in the same format, so that mapping is written once. No claims platform connector or webhook exists today; integrations are in design on the roadmap. Claim files that cannot leave your network can run on Enterprise in a private VPC or fully air-gapped on premises.

What the claims cycle costs today

J.D. Power's 2026 U.S. Property Claims Satisfaction Study, as reported by Decerto, puts the average time from first notice of loss to final payment at 40.7 days. That is a third-party benchmark for the whole cycle, not a measure of docRefinery. Intake is only the front of it.

Claims work it does not cover

• You want damage assessed from photographs. docRefinery reads documents, not damage.
• Your claims arrive mostly as free-form letters and handwritten narratives with no repeating layout.
• You need fraud scoring or automated coverage decisions.
• Your intake already arrives as structured data from a portal, so there is little paper to read.