Claims triage and underwriting a regulator and a board signed — together.
How a European insurer put agentic AI into claims and underwriting under a human-in-the-loop policy — every decision logged with lineage — and cut claims cycle time by 42% while passing model-risk review.
Three moves between a model nobody would sign and agents in production.
No demo theatre. A short discovery, a governance-first design, and agents that execute claims and underwriting under a policy the regulator and the board both signed.
AI in claims the regulator wouldn't let ship.
Claims and underwriting drowned in manual triage, and earlier models had stalled in model-risk review — no lineage, no human-in-the-loop policy, nothing an auditor or the board could sign.
- Manual triage queues, days of cycle time
- Model risk and the EU AI Act blocking go-live
- No decision lineage the regulator would accept
Governance-first agents, three weeks of design, eight weeks to live.
We designed the human-in-the-loop policy and the model-risk framework first, then built agentic triage and underwriting support on top — agents read, classify and progress each case; adjusters and underwriters approve and own the edge cases. Every action is logged with full lineage.
- Risk register + eval gates before the first model shipped
- Human-in-the-loop policy mapped to EU AI Act tiers
- Decision lineage and incident pathway built in from week one
Live in eleven weeks. Signed by the regulator and the board.
Agentic triage and underwriting went live across the claims book — escalating only true exceptions to a human — and passed model-risk review with the regulator and the board signing the same policy. The insurer's own teams now operate it.
- −42% claims cycle time across covered lines
- +19% underwriting precision against gold sets
- A+ model-risk grade · ISO 27001 + EU AI Act audit trail
"They didn't show us another model. They wrote the human-in-the-loop policy with our risk team, agreed it with the regulator, and shipped agents we operate ourselves. That's the difference between a pilot and production."
Three things this story is not.
A demo we built for a deck.
A system the client operates in production today.
A model the client can't audit.
An ISO 27001-aligned audit trail their DPO signed off.
A platform we own forever.
A reference stack their own engineers now run end-to-end.
The firm behind this outcome — EU-based, senior-led, marquee pedigree.
If this is the kind of partner your board can defend, the next page tells you who we are, where we come from, and how we work.
From their pilotto your platform.
Same approach. Same governance. Your use case next — bring the constraints, leave with the path.