A Gulf public health ministry
Healthcare / Government · Gulf
The shape of this problem
A human reads it and types it in again
The problem
Teams needed to convert paper and scanned records into structured data. Health records on paper cannot be queried, audited, or linked to anything - the information exists but is unavailable to every downstream system that needs it. The scans are also not uniform: form layouts vary, quality varies, and handwriting turns up exactly where it is least convenient, so a fixed-template extractor fails on precisely the documents that matter most.
What we built
We used vision models to segment pages, extract text, preserve each value's location on the page, itemize fields, and prepare the outputs for downstream processing.
What moved
| Measure | Before | After |
|---|---|---|
| Handling time per record | 18 minutes | ~2.5 minutes |
| Extraction accuracy | not measured | 96% clean digital forms / 84% poor scans / 61% handwritten entries |
| Fields routed to human review with their source region highlighted | not measured | 14% |
Handling time per record
Fields routed to human review with their source region highlighted
14%
Fields routed to human review with their source region highlighted
The ring is the whole of that measure. The grey arc is the other 86%.
Figures are drawn from the practice's own delivery records for the engagement named, measured against the process that preceded it. They have not been through third-party audit, and none is presented as an average across clients.
What it turned on
Preserving where a value was found, not just the value, is what makes a low-confidence field reviewable - the reviewer is shown the source region instead of having to re-read the document. Relevant to any organisation whose paper archive is the reason a digital system still runs on manual entry.
- Service line
- AI Document Intelligence
The proof is telecom and financial-services documents, not a construction subcontractor invoice. The machinery is identical; say which it was.
Start here
Which of the four is yours?
Tell us the documents and the monthly volume and we will send the two closest records, with the proof behind each and the honest note where the match is partial.
You get a reply within one working day, from the engineer who would do the work - not a sales sequence.