Claims Triage GenAI Copilot
Claims adjusters were spending the majority of their time reading documents before they could even begin to evaluate a claim. The backlog was growing, SLAs were slipping, and the manual review process had become the bottleneck for the entire operation.
A mid-sized insurance carrier processing thousands of claims monthly across personal lines and commercial policies.
Manual first-pass review of claims documentation was creating a throughput ceiling — adjusters spent more time reading than deciding.
Quick Summary
We built a GenAI copilot for an insurance carrier that reads claims documents using OCR and Document AI, then recommends claim type, severity and routing with a full reasoning trace. Integrated with Pega, it cut manual first-pass review by 60%, tripled claims throughput and restored SLA compliance without adding headcount.
Challenges Identified
Every claim arrived as an unstructured bundle — PDFs, scanned forms, photos, and emails with no consistent format or routing logic.
First-pass triage required a trained adjuster — junior staff could not reliably route complex claims, creating a dependency on senior reviewers.
Audit requirements demanded full documentation — any automation had to produce a traceable decision log, not a black-box output.
Legacy claims system had no API — integration required a middleware layer to bridge the copilot with Pega without replacing core infrastructure.
Sakha Solution
Document ingestion pipeline with OCR and Document AI — extracts structured data from PDFs, forms, and images regardless of format.
GPT-4 powered triage reasoning layer — reads extracted content, identifies claim type, severity, and routing recommendation with a reasoning trace.
Automated routing rules engine — maps copilot output to adjuster queues based on claim type, value band, and complexity score.
Full audit trail via structured logging — every copilot decision stored with its evidence chain, meeting regulatory documentation requirements.
Technical Highlights
Impact
Measurable outcomes delivered to our client.
60% reduction in manual first-pass review — the copilot pre-qualifies the majority of straightforward claims without human involvement.
3× throughput increase in the claims queue — adjusters process more claims per day by starting from a structured summary, not raw documents.
SLA compliance restored — average time-to-first-decision dropped to within target for the first time in three years.
Zero additional headcount required — throughput gains came entirely from automation, not hiring.
"The copilot reads faster than any associate we have trained. And it never misses the routing criteria."