Image recognition is used for automation in factories, laboratories and medical centers. But can it be used in insurance industry to help underwriters estimate the risk of insuring a person? Or to estimate damage done to a car? Could we collect data without expensive equipment and use it to further minimize probability of a claim? Let’s see how the biggest insurance companies in the world use image recognition in their work.

Image recognition algorithms can help assess risk

Risk assessment is still done with legacy methods like the generalized linear model. But statistics is not as accurate as checking each case individually. In property insurance image recognition can significantly increase our understanding of details of the property that are a factor in risk assessment.

Satellite or drone footage allows underwriters to see images of the property from each angle and it allows them to know what material is the roof made from or to estimate the area of the building. 

Image recognition and machine learning can be used to estimate post-event response. Insurers can now understand and properly estimate the damage. And using algorithms will definitely make it faster.

Statistics are in favor of algorithms. An average property insurance carrier visits only 10%-20% of their portfolio. While visiting – error rate is 30%-35%. With machine learning algorithms using pattern and image recognition they could get to 100% coverage and 90% accuracy.

Automating motor insurance

While we almost have self-driving cars, more and more is done to automate not only vehicles but also motor insurance.

While now to assess the damage someone has to physically see the car, take photos of it and write a report, we are developing image recognition algorithms that will be able to understand if the car is repairable or not. This saves 1-4 weeks that an assessor takes to review a claim. With image recognition we will be able to decide the claim immediately after receiving a photo.

And although there are services like that they need to be adopted into the insurers claims process.

Image recognition algorithms can make that decision for an underwriter and automate the process without human interaction. And not in weeks, but in a day.

Using drone or satellite imaging for agricultural insurance

Hyperspectral imaging using drones or satellites can scan large agricultural areas and monitor it 24/7. Normally it would be done with a large human task force. It would take days to collect the data. Satellite and drone images let insurer to precisely evaluate risk.

It lets the insurer to evaluate damages from drought automatically without visiting the site of occurrence.

It also lets to precisely scan vegetation for more accurate pricing of the insurance.

Which, at the end of the day, will reduce risk and increase profit for the insurer.

There are still many obstacles in developing accurate image recognition algorithms for insurance companies. Small mistake can lead to a large financial loss.

But time and labour savings are definitely worth it.

There are many companies that are working on universal algorithms for insurance, but I believe they will have to “personalize” the experience for each insurance company so they will be able to seamlessly add them to their workflow.

Author: Marcin Słomiński