AI could change car insurance

Claiming from your insurance for a dent or scratch on your car is rarely pleasant. You need to take the vehicle to an inspector to assess the damage and there may be uncertainty about whether your claim will be considered valid.

If, after all of this, the insurer declines the claim you’d have wasted time.

The machines are here

Automating this process would, however, have benefits for both parties. It already seems likely that things will move quickly in this direction.

At the Actuarial Society of SA Convention in Cape Town, artificial intelligence (AI) solutions provider NumberBoost showed how this would work.

When a car’s insured, the owner or inspector would use an app to take pictures of the car from different angles, which would be used to create a condition profile. This would then be stored as a basis for future claims. When the client lodges a claim, all they need to do is take a picture of the damage and submit it via the app.

The images wouldn’t be assessed by human inspectors, but by computer tools with the ability to identify different types of damage. This could be done immediately, thus claims could be approved within seconds.

Augmented reality

“Computers do the bulk of the work,” says NumberBoost’s Adriaan Rowan.

“But you have customer satisfaction, where results can be shown instantaneously, so you have faster claim settlements, more accurate appraisals and less fraud. You have the before and after pictures, fewer humans in the process who can commit fraud, and geolocation of images that can track where they were taken and if it makes sense in the context of the claim.”

The human element wouldn’t, however, be taken out of the picture entirely. As NumberBoost’s Alex Conway explains, the algorithms used will always come up with a probability of what it believes it has identified.

Most of the time, this probability will be high enough for the model to be sufficiently confident it’s right. However, there’ll be cases where a human is needed to confirm it, potentially due to the angle of the picture or shadows on the car.

This also creates an active learning loop for the algorithm, because the feedback from the human decision is passed back to guide future decisions.

“You want to have a system where man and machine work together.

“In many cases the machine is better, but when you combine the two you have more accurate output,” says Rowan.