By Nika Lee, Chief Underwriting Officer, Aioi Nissay Dowa Insurance Europe
To avoid the very real risk of underinsurance, here at Aioi Nissay Dowa Insurance Europe (AND-E) we have been working in partnership with AI experts, Mind Foundry, to develop an innovative machine learning approach to efficiently identify drivers who may be using their vehicle for commercial purposes. The process has provided an incredibly effective and efficient means for telematics insurance providers such as ourselves, to identify potential underinsurance or misrepresentation at scale.
At a time of financial vulnerability it could be argued that proper insurance protection is more important than ever before for UK motorists. However, with average quoted motor insurance premiums surging by 34% in the past year and with lower-priced deals hard to come by, the temptation of non-disclosure is great, meaning the need to identify fraud, opportunistic or otherwise, grows with it.
While actuarial professionals have been using effective analytical tools and statistical modelling with huge success, it makes perfect sense to exploit machine learning to deep dive into data and augment insurance professionals’ decision-making.
Indeed, machine learning can be used to research incredibly specific circumstances that can help lead to uncovering underinsurance, therefore supporting better risk management. Our new white paper ‘A Bayesian Approach for Prioritising Driving
Behaviour Investigations in Telematic AutoInsurance Policies’ explains how common data captured by black box telematics systems including location, speed, acceleration or braking, fuel consumption, idling time and vehicle faults, can provide in-depth insights that can be analysed for events and patterns to identify vehicles which may be used for deliveries.
Of course, the volume of data generated by telematics insurance is far too great for manual review alone. In order to extract useful knowledge from blackbox GPS data that’s suitable to be passed to a machine-learning algorithm, domain knowledge is used to construct a feature set from the raw data.
For example, it is likely that delivery drivers perform multiple deliveries in one journey, which would suggest long trips with an alternation between residential and high street-like destinations. Other trip-level characteristics such as time of day, and day of week, also inform the model.
In addition, trip stationary points are analysed to determine if a vehicle is stopped at a traffic light, for example or has made a genuine stop. Then by extracting these datasets including average trip duration, total number of stops, average wait at a stop, number of commercial and domestic destinations and time of day, a model is made for analysis.
It would be too time-consuming to employ an insurance professional to extract this detailed information without the help of AI. Yet, it is important to note that machine learning is by no means a replacement for human-decision making, rather it enhances the process. In fact, machine learning blends in analysis of GPS data to provide a prioritised ordering of policies for human investigation.
The automation ensures a consistent application of investigation, so enables fairer risk assessment and treatment of customers.
Provision of the resulting predictions in a ranked score then enables the insurer to focus on the highest risk and manage risks appropriately with available resources.
The ultimate goal is to produce driver-level referrals which are given a priority score meaning individual drivers can be reviewed top to bottom starting with those likeliest to be delivery drivers, thus making the most efficient use of time.
Impressively, 99.4% of the top 0.9% of policies identified as being most likely to be being used for commercial driving have been confirmed as correctly identified.
The integration of AI and machine learning into data analysis to identity potential underinsurance and misrepresentation puts insurance professionals in a more informed position to help customers avoid underinsurance before it becomes an unpaid claim.
After all, some may be unaware that they are driving illegally without the appropriate cover and that by investing in ‘top up’ Hire and Reward insurance to supplement their existing policy, they can avoid the negative outcomes of underinsurance.
As insurance providers fight for advantage in a competitive marketplace, the results of this innovative human-AI collaboration clearly shows how this approach has achieved a significant improvement in efficiency of human resource allocation compared to manual searching alone.
There is clear potential to expand the approach to other behaviours of interest to telematics insurers, including business use, private hire or other driving behaviour. More general applications could also be explored, including classification problems where the approach could be adopted as part of a model to detect fraud, theft or other high risk events. It could also be used to measure confidence around prediction in pricing or risk cost models. With such potential, AND-E will continue to work in partnership with Mind Foundry on AI initiatives such as this to tackle everyday insurance problems.
|
