AI perspectives from Tractable.

Back to blog list: See all posts

14 Apr 2021

What is Computer Vision and why should insurers care about it?

By: Jimmy Spears, Head of Automotive at Tractable

Close up of eye covered in binary code

Can a computer assess images of a car and accurately recognize the damage present? 

Before we get to the answer, it’s worth a quick recap on the technology that is responsible for making these decisions: computer vision.

Computer vision: in full view

Computer vision is a type of AI that replicates the visual intelligence of the human brain (i.e. understanding what’s in a 2D or 3D image). It dates as far back as the early 1960s, with the last two decades being a turning point for the field. 

In the mid-2010s, researchers were able to create algorithms for image classification and object detection that for the first time surpassed human performance. This was a huge milestone and necessary to allow the real-world and commercial applications of computer vision to be explored. 

And this is why Tractable was founded in 2014. Computer vision had reached an accuracy level that meant we could build AI models to make a genuine difference in people’s lives. 

Putting insurers in the driving seat

After the success of projects like ImageNet, it is now possible to train a computer to interpret and understand what’s in an image, like a human would. 

And that means we can train an AI model to complete a task that’s usually carried out through visual assessment – for example, evaluating the damage to a vehicle after an accident. By developing deep learning algorithms from hundreds of millions of images, AI can match the performance of a human assessor. 

Policyholders simply need to upload photos of the damage. Then the AI will analyze the photo pixel by pixel and classify the amount of damage based on its training. It will then create an estimate of the extent of the damage.

All estimates are provided with a certainty score (ie – how sure it is of the assessment it has made), depending on part visibility, photo conditions, damage severity, etc. If the AI gives a low confidence in the predicted repair cost, the claim can then be directed to a human to review. However, if the certainty score passes the required threshold, the estimate can continue through the claims process. 

One key fact to remember is that computer vision is not designed to replace humans in the claims process. Instead, it enables them to work smarter and allocate their expertise more efficiently. The AI augments human experience and knowledge, automating the claims that can be and directing those that need extra attention to the experts. 

This means that more claims can be processed and to a higher overall standard, benefiting insurers, collision repairers, and policyholders simultaneously.

Looking to the future

The foundations have been set. Computer vision can successfully and accurately assess vehicular damage. 

The next steps come down to two key areas: (1) continue to grow the dataset that the AI learns, increasing accuracy over time, and (2) expand the use cases of this technology beyond automobiles. 

These two steps need not be sequential but should be simultaneous. The bigger the dataset, the smarter the AI. The smarter the AI, the more it can be applied to different situations. 

The end goal for Tractable is to deploy its AI in disaster recovery, being used by insurers, governments, and relief organizations to help those who need it most. 

To learn more about how the technology works, book a demo with the team today

Image by The Digital Artist from Pixabay

More content we think you’ll like:

5 things you need to know when integrating AI into your claims environment

Previous post: Three key takeaways on claims automation

Back to blog list: See all posts

Next post: Tractable selected to join Future Fifty

Get updates