Claims triage copilot at Meridian Mutual
A copilot that reads the claim, proposes the route and shows its evidence: built with the claims team, handed to them in twelve weeks.

−41%
claims-processing time
How it was measured
Measured as median hours from claim receipt to first decision, comparing the quarter after launch with the same quarter a year earlier, on the same four channels.
Stated this way because a figure without its method is a claim. Ask us for the workings: we will send them.
The copilot reads the claim, proposes the route and shows its evidence. These are the numbers the team held it to.
The challenge
Claims arrived by four channels into one queue, and were worked in the order they landed. A straightforward motor claim waited behind a complex liability one; the adjusters knew it and could do nothing about it at volume.
The approach
Eight weeks on the data first, one definition of a claim, one owner per field. Then a triage copilot built with the claims team in the room: it reads the claim, proposes a route, and shows the three sentences it based that on. Adjusters overrule it freely, and every overrule is logged as training signal rather than as an error.
The result
Processing time fell 41% across 11,400 claims in the first quarter after launch. The number the board watches is not the model's accuracy but that one.
The engagement, on paper.
Delivery
Where it sat in the method.
Digital & AI is stage 03 work. Every engagement opens with the audit whichever stage it lands in, and the roadmap from that audit is what priced this one.

“Tessel left us running the system ourselves, which was the point.”
More published cases.

Market entry and portfolio calls at Fennimore Foods

