ClaimSetu: turning a motor insurance claim from a two-week phone saga into a guided, photo-first journey

ClaimSetu's head of claims sent us something unusual before our first meeting: an export of six months of customer complaints. More than half were about motor claims, and almost none were about the settlement amount. They were about not knowing what was happening — "I sent the photos twice", "the surveyor never called", "the garage says they are waiting for you".

Behind every complaint was the same pattern. A claim started with a phone call, then moved through email, WhatsApp, a surveyor's personal phone and a garage's paper job card. Each hand-off lost information, and every missing piece meant another call to the policyholder.

In discovery we followed live claims with the claims team's permission: a minor bumper dent, a windscreen replacement and a larger collision. We sat with a surveyor on inspections and spent a morning in a partner garage. Four parties touched every claim — the policyholder, the claims handler, the surveyor and the garage — and none of them could see what the others had done.

The design principle that came out of this was simple: one claim, one timeline, visible to everyone with a role in it. Each party would see the parts relevant to them, and every action would appear on the shared timeline automatically.

The most important screen in ClaimSetu is the camera. Most delays started with poor photos, so we built a guided capture flow: an outline of the car shows which angle to take next, the app checks each photo for blur, distance and lighting before accepting it, and it records the time and location automatically.

For minor damage, those photos go straight to an AI damage triage model that identifies the affected panels and suggests whether a physical survey is needed at all. Clear, low-value claims can be assessed on photos alone; anything larger or ambiguous goes to a surveyor with the photos already attached.

ClaimSetu needed mobile, computer vision and enterprise workflow skills in equal measure, so the team was built around three small pods that shared one backlog and one weekly demo with the claims leadership.

Insurance mixes customer experience, fraud risk and regulation. These were the decisions we took with the claims and compliance teams before building.

Twenty-six features across four products, all sharing one claim record and one timeline.

We trained and evaluated the damage triage model on historical claim photos provided by ClaimSetu, labelled with the panels and repair decisions surveyors had actually made. The model never approves a claim on its own; it recommends a route — photo assessment or physical survey — and handlers can always override it.

The fraud signals were deliberately simple and explainable: photos taken far from the reported location, the same vehicle appearing in multiple recent claims, image metadata that did not match the claim date. Each signal shows its reason so a handler can judge it.

ClaimSetu launched first for minor motor claims in two cities, then expanded to all motor claims nationally. The claims team's favourite change was not the AI — it was that policyholders stopped calling to ask for status, because they could see it.

Garages adopted the portal faster than expected once they realised approved estimates came back the same day instead of waiting for an email chain.

Better inputs beat smarter models. The guided camera did more to speed up claims than any model improvement, because it removed the most common reason for going back to the customer.

Shared visibility changes behaviour. When the garage, surveyor and handler all see the same timeline, delays become visible — and visible delays get fixed.