Clarity Underwriting reviews commercial insurance policies for a mid-sized carrier. Manual review was thorough but slow, and the backlog it created was starting to cost the business renewals.
The Challenge
Every policy required an underwriter to read through submission documents and flag risk factors by hand. The process was accurate, which mattered enormously in a regulated industry, but it didn’t scale with submission volume, and the resulting backlog meant some renewals were being decided later than clients expected.
The Approach
We built an evaluation-first AI copilot rather than an automation tool: it reads submission documents and flags risk factors for a human underwriter to confirm or override, never issuing a decision on its own. Before it touched a single real policy, we built an evaluation harness that measured its flagging accuracy against a large set of Clarity’s own historical decisions, and we did not consider it ready until that accuracy held steady across edge cases, not just the common ones.
Human-in-the-loop review stayed mandatory for every policy, and the system logs its reasoning for every flag, which mattered as much for audit requirements as for underwriter trust.
The Results
Review time per policy dropped by 58%, and critically, underwriting accuracy showed zero measurable regression against the pre-AI baseline, the metric Clarity cared about most going into the project. The backlog that had been costing renewals is gone, and underwriters report the tool catches details they’d occasionally miss late in a long review session.
