Internal AI Workflow Tool
- Client
- Regional insurance group
- Sector
- Insurance
- Service
- AI
- Timeline
- 10 weeks
- Year
- 2024
- Stack
- Next.js / TypeScript / PostgreSQL / Claude API
Forty staff hours a week were disappearing into reading, sorting, and retyping claim documents.
The problem
Claims intake arrived as unstructured email and scanned PDFs. Staff manually read each document, extracted key fields, and retyped them into the core system, slow, error-prone, and impossible to scale during seasonal spikes.
Our approach
We shadowed the claims team for a week before writing code, then built an extraction pipeline with a human review queue. High-confidence documents flowed straight through; anything ambiguous was routed to a reviewer with the AI's reading pre-filled for one-click correction.
What we built
A document pipeline that classifies, extracts, and validates claim data, paired with a review dashboard the team actually likes using. Every decision is logged, every extraction is traceable to its source text, and the system gets measurably better from reviewer corrections.