Machine Learning & Insurance rating

Motor insurance is out of line with almost every other industry when it comes to using artificial intelligence (AI). 

For an insurer, the ability to quantify risk and uncertainty is a key driver for success in a market where the prediction of risk – the output of which is a price offered – plays such a pivotal role in a sale to a customer.

Currently, the job of managing risk for an insurance company falls under the remit of actuaries. An actuary will collate relevant data, organise the data, run the data through pre-built models and software, tweak the rules and results using their experience and intuition, and diligently test, qualify and finally version pricing models to value risk. It’s a tried and tested method.

The iterative nature of the actuarial process makes it perfectly suited to machine learning (ML), where a computer algorithm trains itself through experience and by the use of data to achieve a required outcome. Machine learning is a subset of artificial intelligence (AI), which is commonplace in a huge range of applications, from predicting your next purchase on Amazon to driving you to your desired destination in your Tesla. It’s now a well-established tool for greatly accelerating and improving decision-led processes in almost every industry.

Every industry, it seems, except motor insurance. Whilst AI is being talked about by many insurers, the application of it is thin on the ground, and procrastination is routine - knowing where to start can be intimidating. It could be suggested that the current ‘big-picture’ vision of an efficient, automated end-to-end insurance stack through the use of AI (AI chatbots, AI pricing, AI claims, the list goes on) is stifling innovation, and insurers need to think ‘small steps’ first.

The introduction of AI to insurance is often seen as a wholesale replacement of the classical actuarial sciences, but instead of taking it out of the picture entirely, machine learning could be used to enhance traditional actuarial processes. Large data sets, too cumbersome and overwhelming for a human to find meaning in, could be fed into a ML system to create distinct factors on claims predictions or credit risk, which could then be incorporated into a classical rating model. ‘Pay-how-you-drive’ motor insurers have been applying a similar idea of taking a nebulous score (a ‘driving score’) to create rating factors for some time now.

Regardless of the first steps the industry takes, the rewards to both insurer and consumer are aplenty. For the insurer, starting now means future-proofing for the inevitability of data from drivers and their vehicles becoming ubiquitous. It will allow them to price better, support innovation inside and outside of their domain, and deliver a more customer-centric experience. Insurance will be dominated in the future by insurers who use the power of artificial intelligence to innovate, and starting sooner rather than later is crucial.