AI claims estimation is software that estimates the cost of damage to a car or a home from photographs, using an AI model, instead of waiting for an assessor to visit. An assessor, called a surveyor in Indian motor insurance, is a person who inspects the damage and estimates the repair cost. With AI claims estimation, a person still checks the estimate before it is paid.
Here is how it works. The customer photographs the damage in the insurer's app, from several angles. A computer vision model finds the damaged parts in each photograph. On a car, it might find a dented door, a cracked bumper and a broken headlamp. It decides whether each part should be repaired or replaced. The software then looks up the price of each part and the labour time for that car model, and adds them up. Older software did only this last step. An assessor entered the damage by hand, and the software priced it. Many American home claims are still estimated this way.
Consider a hatchback in Kolkata whose rear bumper and tail lamp are damaged in a parking area. The owner uploads six photographs. Within a minute, the model marks the bumper for repair and the tail lamp for replacement. The estimate is ₹18,400. A claims employee looks at the photographs and the estimate, agrees with both, and approves the claim. The repair is booked at a garage, and the garage's dealer management system receives the approved amount.
Sometimes the repair would cost almost as much as the car is worth. In Indian motor insurance, a car is treated as a total loss when repairing and recovering it would cost more than 75 per cent of its insured declared value (IDV). The IDV is the car's value written in the policy. The model then flags a total loss. The insurer pays the IDV, minus the deductible, and the damaged car may be sold through a salvage car auction.
The estimate is not the claim itself. The claim is recorded and paid in the insurer's claims management system. Data from usage-based insurance, such as the force of the impact, can help to check whether the damage matches the crash. Because a model drafts and a person approves, this is a form of human in the loop.