The Telematics Holy Grail: Predicting a Driver’s Risk of an At Fault Claim

A decade ago, Usage Based Insurance (UBI), often known as “Black Box Insurance”, using telematics was envisaged as a panacea for the insurance industry as a way of linking insurance premiums to driver behaviour. This new wealth of data was meant to enable insurers to identify trends with drivers who have ‘At Fault Claims’ and then incorporate these insights into intelligent premium pricing.

Ten years later, the penetration of UBI in motor policy books of business remains a long way below the 50% figures originally expected. One of the challenges is that telematics generates a huge amount of data, with the device in each vehicle generating a wide variety of data every second that the vehicle is driving. Notwithstanding the massive size of the overall dataset, traditional methods also required a lengthy and tedious process of data cleansing to correct any occasional errors, meaning more effort has to be taken to analyse individual chunks at a time.

Therefore, although the process is possible to be done manually, it is a lengthy process which needs to be done by someone highly knowledgeable in data science, a rare and valuable skillset within the sector.

At Massive Analytic, our data science artificial intelligence platform ‘Oscar’ provides a solution to all of these problems simultaneously. It simplifies the process, allowing it to be done in greater detail, in a shorter period of time, by those who do not have the same level of data science experience.

Massive Analytic’s unique precognition machines are effective at reducing decision outcome uncertainty for businesses across a variety of different sectors. Through our recent research, we have identified how Oscar can be effectively applied to improve the accuracy of predicting insurance claims in a telematics environment, which could mark a fundamental change to the future of the industry.

The standard use of a regression model using various data, such as monthly scores across components such as acceleration, braking, cornering & speeding together with mileage, generates a value called the 'Regression Claim Result'. When this is plotted against the Overall Monthly Mean score for each policy holder it generates a scatter plot, as can be seen from the graph, which fails to identify an obvious pattern that differentiates those policy holders who have had an 'At Fault Claim' (shown as blue dots) from those who haven’t (shown as orange dots). They are randomly distributed amongst the plot and ultimately does not deliver any meaningful increase in understanding of the data.

The solution to this is the use of Artificial Precognition on the data.

This involves the initial step of ‘Coarse Tuning’ which is a method to amplify the information, without amplifying the noise. This can be done for example using a possibilistic decision tree, which breaks down the data and builds a model at the same time. The model is then deployed against the data set to get a new variable that tells us more. This bins unnecessary noise and groups the data together, which can be used for prediction.

This cognised variable can then be used for a Regression Analysis. As can be seen from the graph, this creates much more useful results – as there are clearer groupings of data and gives insurers a much clearer idea of which drivers were likely to be at risk and thus make claims in the future.

Overall, the differences between the two graphs highlights that the use of Massive Analytic’s Artificial Precognition not only includes a far wider selection of variables, but simplifies the outputs into visibly clear results, which dos not require a specialist data scientist to analyse them.

The ability to predict which drivers are the most likely to make an “at fault claim” and therefore price their premiums accordingly is something of a “holy grail” for insurance companies. The potential cost savings are certainly enormous. With Oscar and AP we have not only moved closer to being able to do this but have automated the process as well, for more accurate insights in a fraction of the time. Bringing us closer than ever before to delivering on the promises UBI made a decade ago.