Medical claims intake with OCR and document extraction
Scanned hospital bills, receipts and referral letters are read, checked against the policy and queued for assessors, in English and Bahasa Melayu.
- SECTOR
- Finance & Insurance
- LOCATION
- Kuala Lumpur, MY
- YEAR
- 2025
- DELIVERED AS
The problem
Claims arrived by email and WhatsApp as phone photos and multi-page PDFs. Processors keyed every bill line into the claims system by hand, so a claim waited days before an assessor saw it, and typing errors surfaced only when a payment was disputed.
What we built
An intake pipeline that splits each submission into documents, classifies them, reads the fields and line items, and checks the totals and member details against the policy record. Every field carries a confidence score and a link to where it was found on the page. Anything below the agreed threshold, or failing a check, goes to a review screen with the scan side by side. Approved claims are posted to the existing claims system through its API.
Key features
Document splitting and classification
Field and line-item extraction with confidence scores
Checks against policy and member records
Side-by-side review queue for assessors
How the 4-week sprint went
Listen & prototype
Labelled 300 past claims with the processing team, agreed which fields matter and set the review rules.
Core build
Built extraction, validation checks and the review screen, measured field accuracy weekly against the labelled set.
Live & handover
Ran in parallel with manual entry for two weeks, then switched one claim type over and handed over the pipeline and evaluation scripts.