Case study
Remote Damage Assessment Model Speeds Insurance Claims
How SquareShift built a computer-vision model to remotely assess vehicle damage and speed up insurance claims for a Southeast Asian insurer.
Sending an inspector to every fender-bender doesn't scale, but skipping inspection entirely is worse. A model that flags severity and checks it against the claim is the safer middle path.
AI & Generative AI Practice Lead, SquareShift
A top-tier insurance provider in Southeast Asia, offering automobile and personal insurance across multiple countries.
It carries a large base of policyholders and claims volume, and previously relied on manual, in-person inspection to assess vehicle damage.
- Remote assessment — Fewer minor claims need an in-person inspection.
- Faster claims — Damage details and payouts reach inspectors automatically.
- Consistent payouts — Estimated and final claim amounts checked against each other.
How SquareShift delivered it.
The challenge
A top-tier insurance provider in Southeast Asia, offering automobile and personal insurance across multiple countries, was settling claims the slow way: an inspector had to physically evaluate each damaged vehicle before a claim could move forward.
Against thousands of vehicles and a large claims volume, that model couldn’t scale, and there was no automated way to check whether an estimated claim actually matched the damage once it was assessed.
What we delivered
SquareShift built a real-time video and image processing model to detect a vehicle’s make and the severity of its damage directly from images, trained on a purpose-built dataset of car images. The model classifies the vehicle, the damaged parts, and the severity of the damage automatically.
Claim amounts are then matched against what the model detects, and inspectors receive an automatic notification with the damage details and the proposed payout — so they review a flagged case rather than starting every assessment from zero.
The payoff
Fewer minor claims need a full in-person inspection, since the model can assess vehicle damage remotely from images alone. Damage details and payout information now reach inspectors automatically instead of through manual write-ups.
Estimated and final claim amounts get checked against each other as a matter of course, giving the insurer a consistency check that wasn’t there when every assessment was manual.
Where this work sits