InsuranceLive
Laya Healthcare’s claims system recovered over €800,000 in four months.
We built Laya Healthcare’s receipt-reading claims system and call summaries, and we support four production capabilities under contract, up from one.
Laya Healthcare
- €800,000+recovered in the first four months, against a €300,000 business case
- 6,000receipts an hour absorbed at peak member upload
- 3 secondsfrom the end of a call to its summary, category and sentiment
- 1 to 4production capabilities, across 18 engagements
The situation
Laya Healthcare serves more than 600,000 members. Outpatient claims arrive as photographed receipts, handwritten dockets and photographs of provider letterheads, at peaks of 6,000 an hour: more than people can review by hand.
Laya set a business case of €300,000 for automated detection of overspend, fraud, waste and abuse in outpatient claims.
What we built
We began with a two-week proof of concept against 1,000 real receipts, then a 15-week build.
Vision-only extraction proved unreliable under test, so the pipeline runs in two stages: optical character recognition reads the receipt first, and its output goes to the vision model as context. Each receipt is processed inside a 20-second budget, with fallback to manual entry if it runs over.
A follow-on change moved work out of the model into pre- and post-processing, taking member response time from 23 seconds to under three with no loss of accuracy.
A companion service on Laya’s contact-centre platform returns a call summary, category and sentiment within three seconds of the call ending, grounded against a supplied list of providers and locations so that transcription reaches the accuracy the workflow needs.
The decision that mattered
Before going live, the system ran alongside production, processing real receipts beside the manual workflow, so accuracy was proved at full volume before any payment decision moved. Both design changes, two-stage extraction and the faster response, were proved alongside production, against live volumes.
Where it stands
The system recovered more than €800,000 in its first four months, against the €300,000 business case, and absorbs peak uploads of 6,000 receipts an hour.
Call summaries, with a category and sentiment, arrive within three seconds of the call ending, and the service has released capacity for 3,813 more service calls a year.
The relationship has grown across 18 engagements, from one production capability to four, and we keep them right under a continuous operations contract.
Laya Healthcare won Best Industry Application of AI at the itag Excellence Awards 2025, and Best Application of AI in Consumer/Customer Service at AI Ireland’s AI Awards 2025, for this work.