A Fortune 1000 insurance carrier deployed document intelligence and GPT-4, cutting claim cycle time by 5 days and saving $4.8M annually.
A top-50 US insurance carrier with 8,000 employees processed millions of claims annually. Claim adjusters spent 60-70% of their time extracting data from unstructured documents: police reports, medical records, repair estimates, photos.
Average claim took 14 days to process from submission to decision. Two-thirds of that time was adjusters manually reading documents, highlighting relevant sections, typing data into the Guidewire claims system, and writing summaries. For a claim with a police report, repair estimate, medical exam, and photos, an adjuster could spend 3-4 hours just on document handling.
The carrier had 3,200 adjusters. Each hour saved per claim, multiplied across annual volume, represented millions in productivity value.
Modern document intelligence (Azure Document Intelligence) combined with GPT-4 could extract structured fields from claims documents with high accuracy, classify claim type, and summarize key findings. That structured output could be delivered directly to Guidewire and the adjuster dashboard, reducing manual extraction from 3-4 hours to minutes of review and approval.
A human-in-loop escalation workflow would flag complex, disputed, or high-risk claims for full adjuster review.
We built a processing pipeline using Otonmi, Ingress's AI division, to integrate modern AI into an existing enterprise claims system.
Adjusters immediately trust 99% accuracy. Below 95%, skepticism grows. We spent 4 weeks tuning the extraction model to build that confidence baseline. The speed multiplier only mattered if accuracy was bulletproof.
We never tried to auto-approve claims. Fraud flags, disputes, high payouts, and novel claim types all escalate to human adjusters. That design decision eliminated concerns about AI making decisions without oversight. Speed and safety coexist.
Insurance is regulated. We built detailed logging: what model processed the claim, what data was extracted, why it was escalated, what an adjuster approved or changed. Every decision is traceable for audits and fair lending reviews.
The AI itself took 6 months. Training adjusters and rolling out the workflow took 3 more months. Organizational readiness, not technology, was the longest pole. We over-invested in communication and incremental rollout, and it paid dividends.
Bring the problem. We'll come back with a written brief: what to build, what to defer, and where AI actually moves the number. No deck pitches.