Motor damage assessment from photos for an insurer

The client is a UK motor insurer where claims experience is a major driver of customer retention, and every day a claim sits open costs money and goodwill. Speed and fairness in claims are core to how they compete.

They also faced persistent staged-damage fraud, which is hard to catch reliably by eye at the volume and pace their claims team operates.

Every motor claim waited on a physical assessor visit, adding days to settlement and cost to each file in a market where claims experience drives retention. The delay frustrated genuine customers at exactly the moment they most needed the insurer to perform.

Staged-damage fraud slipped through when photos were reviewed by eye, especially at volume when reviewers were under time pressure and inconsistencies were easy to miss. Fraud that gets paid is a direct loss and pushes up premiums for everyone.

The insurer needed to fast-track the routine cases without opening the door to leakage. Speeding up settlement and tightening fraud control usually pull against each other, and they needed both at once.

We built a vision model that grades damage severity and identifies affected parts directly from claimant photos, pricing straightforward cases straight through. The routine, honest claims (most of the volume) settle in hours instead of days.

The same model flags inconsistencies (mismatched damage, reused or manipulated images) at first notification, so suspicious files surface before money moves rather than after. Catching fraud early is far cheaper than clawing back a paid claim.

Low-confidence or high-value cases still route to a human assessor, keeping judgement where it belongs. We trained on the insurer's historical claims and shadow-tested against assessor decisions before going live, so the model earned trust against real outcomes first.