How a computer can ‘see’ what’s in an image is known, naturally enough, as “computer vision.” And in 2014, this technology was at a tipping point.
Advances in computing and processing power meant that we were finally able to train an AI to understand what’s in an image, like a human would. That meant you could, in theory, accelerate any task that relied on visual assessments – like facial recognition, or detecting cancer in medical images.
At my company, Tractable, we applied this technology in an unusual way – to help people recover from accidents and disasters. We started with cars, mainly because cars are similar globally (making the training easier), they get damaged all the time, and because the first thing that happens when they get damaged is that a driver or a body shop takes lots of photos – which is what we needed to train the AI.
By 2019, we’d worked with some of the world’s leading insurers to prove that our technology could make them more efficient, applying our AI to enable them to check whether their repairs and claims had been carried out safely and correctly.
But we wanted to have a bigger impact. So, working with companies like global insurer Ageas, we sought to apply our technology at First Notice of Loss (FNOL) – when a driver first calls to report a claim.
This was difficult, to say the least. It’s one thing for an AI to analyze a perfectly taken photo of one dent to one panel, another for it to assess a collection of grainy pictures of a well-worn station wagon with multiple impacts, taken at night with low lighting. And for the technology to be useful, it had to provide a return almost immediately – perhaps, even, on that initial phone call.
It was a huge technological challenge – but, driven forward by some of the world’s leading AI researchers, one we were able to meet. By 2019 our AI was providing line-by-line estimates of damage at FNOL – and you can see us talking about this here with Ageas, at the DIA conference in Amsterdam.
Taking the human out of the loop
At this point, the AI was accelerating most of the process – but, quite sensibly, the insurance carrier had a human involved to ensure quality control and provide support.
But we wanted to know if we could take the obvious next step and make some of these claims touchless – i.e., driven entirely by AI. That way you free up your appraisers’ and claims handlers’ time, directing them to cases where they can make an impact and use their expertise effectively, instead of expending effort on claims that could be easily resolved with technology.
So, last October we convinced an insurer to take the plunge and let our technology finish a claim on its own – and since then we haven’t looked back. Our AI has now carried out tens of thousands of touchless claims worldwide. It is generating estimates for major insurers in the US, Japan, UK, Spain, Italy, to name a few – and you can see Admiral talk about the impact on their Spanish unit and operations at the video at the top of this piece (check out the killer quote at 12:54).
It’s no exaggeration to say AI has made an impact for millions of households worldwide and helped return vehicles to their drivers far quicker than previously possible.
Trust in the machine
So why isn’t the impact of our technology more widely recognized?
Perhaps the answer is that people just don’t realize. After all, if you’re a policyholder, you have no idea that an AI has just assessed your damage – you’re only aware that your claim has been completed in record time, and you’re now able to carry on with your life.
But in reality, since October 2020 Tractable’s touchless claims have been live across the world, making the insurer process more efficient, helping body shops carry out more repairs, and processing claims and returning vehicles to drivers far more quickly than was possible before.
The challenge for us? The introduction has been so seamless and smooth, they haven’t noticed.