AI · 2026
Claims document processing
Health insurer · UAE
An extraction pipeline that reads submitted claim documents and routes them to the right adjuster with a confidence score.
9.2k
Documents processed monthly
-71%
Manual handling time
The challenge
Claims arrived as scans, phone photos and PDFs in two languages. A team of nine read every one by hand, and the backlog grew every quarter.
Our approach
- 01
Ran a six-week pilot on 4,000 historic claims with a labelled test set so accuracy could be argued with numbers, not impressions.
- 02
Built extraction with a mandatory human review step for anything below the confidence threshold.
- 03
Shipped an adjuster interface that shows the source region of every extracted field, so corrections take seconds.
The result
Manual handling time is down 71% and the backlog cleared in the first quarter. The review team moved to complex claims, and accuracy is reported monthly against a live test set.
- Python
- TypeScript
- pgvector
- AWS
- React
Tell us what you're trying to build.
Send a short description of the problem. We'll come back within one business day with questions, a rough shape and an honest view on whether we're the right team.