Document-heavy claims intake automated end to end, with every straight-through decision reconstructable by an ombudsman.
Where it started
Claims arrived as scanned forms, photographs and email threads. Adjusters spent the first day of every claim simply extracting facts, and the backlog meant simple claims waited behind complex ones. Any automation had to remain fully reviewable — an ombudsman can ask why a claim was handled as it was, years later.
- 01
Built an intake pipeline that extracts structured facts from scans, photographs and email, with confidence scores attached to each field.
- 02
Routed claims by predicted complexity so simple cases clear straight through and complex ones reach a specialist immediately.
- 03
Grounded every generated summary in cited source documents, with low-confidence extractions flagged rather than guessed.
- 04
Recorded the model version, prompt version and evidence for each decision in an immutable case record.
- Claude
- Azure Document Intelligence
- Python
- .NET
- Azure
Where it landed
Handling time fell by more than half, and simple claims that previously waited days now settle the same day. The immutable case record has since been used successfully in two ombudsman reviews.
A short conversation with an engineer, not a sales qualification call. If we're the wrong people for it, we'll say so and point you somewhere better.