Health Insurance / InsurTech · 2024
Claims triage AI that reads 40-page submissions
LLM-backed pre-adjudication agent that flags missing fields, summarizes claim history, routes by complexity. 6× faster intake review.
Client
Health Insurance
Role
Lead engineering team, with a contract data engineer
Year
2024
Stack
Claude 3.5, Pinecone…
6×
Faster intake review
40 pp
Avg. claim length read
11 wks
Brief to production
94%
Adjuster acceptance of summary
THE WORK
STACK
Claude 3.5PineconeAWS BedrockFastAPIPostgres
Challenge
Adjusters were spending the first 40 minutes of every claim reading prior history before they could start work. Backlog was growing faster than headcount could keep up.
Approach
A retrieval system over the member's full claim history, paired with a structured-output agent that produces a one-page pre-adjudication brief: flagged missing fields, similar past claims, recommended complexity tier.
Outcome
Adjuster intake time dropped from 40 minutes to under 7 minutes per claim. 94% of adjusters reported the AI summary "matched what I would have written myself."
“The summaries are good enough that we use them as the front page of the file now. The model writes better adjuster notes than half our adjusters.”
, Director of Claims Operations, Regional Health Plan
MORE WORK
Other projects from the studio.
Have a project like this in mind?
No project is too big or too small. Start with a 20-minute call.