Insurance design-partner focus

When Your AI Denies a Claim, What Evidence Do You Have?

A denial decision is not just another model output. It can trigger complaints, audit questions, legal review, or direct regulatory scrutiny. When that happens, the team needs one defensible record that shows what checks ran, what the AI recommended, who approved the outcome, and whether the record changed afterward.

The problem

In most teams, the evidence for an AI claim denial is scattered. Some of it lives in vendor traces, some in tickets, some in workflow logs, and some only in a reviewer’s memory. That is painful during internal review and worse when an external party asks, “What exactly happened here?”

Compliance and model-risk teams do not need more dashboards in that moment. They need one portable case file that answers four questions clearly: what decision was made, what checks supported it, what human approved it, and whether anyone changed the record later.

What EPI does

EPI captures the claim workflow into a sealed .epi case file. That file can include the claim checks, the AI recommendation, the human approval step, the rule evaluation, and a printable Decision Record. Anyone can later verify whether the record still matches what was created at the time of the decision.

For insurance teams, that means a claim denial is no longer “something the model did.” It becomes a defensible record you can hand to compliance, legal, internal audit, or a regulator.

What the demo shows

The current insurance starter workflow loads a claim, runs a fraud check, checks coverage, requests human approval because the amount exceeds the threshold, records the denial reason, and seals the final denial into a case file. Then EPI exports a Decision Record that can be printed or turned into a PDF directly from the browser.

That is the proof point: not a slide, not an architecture diagram, but the exact record a reviewer would read.

Try it

Run the insurance starter kit locally:

cd examples/starter_kits/insurance_claim
python agent.py
epi view insurance_claim_case.epi
epi export-summary summary insurance_claim_case.epi
Open the insurance starter kit Open the Colab walkthrough

Why this matters now

AI teams often start with speed and only later realize they also need evidence. Claim denials are where that realization arrives quickly, because the cost of “we can reconstruct it later” is too high. EPI exists to make that reconstruction unnecessary.