
Alan O’Loughlin discusses how actuaries can build a deeper understanding of motor insurance risk using probability-to-cancel models and historical quote data
In 2014, a contributory database was created for the motor insurance market so that it could share policy history data. The purpose at that time was to validate no-claims discount entitlements (NCD) at point of quote in order to create cost efficiencies and reduce deliberate or accidental misstates of the years earned by an individual.
Fast forward six years and the policy history gathered from more than 85% of the motor market is now enabling data scientists to find correlations between an individual’s past policy behaviour and the risk that they will cancel mid-term or make a claim, as well as the cost of any claim. Much like policy history data, quote behaviour data (gathered for more than half a decade) is bringing a further layer of understanding to those risks, including the risk of fraud.
From this data we can extract new attributes based on previous cancellations, quote manipulation, mid-term adjustments, vehicle cover, length of time associated with a vehicle, and gaps in cover. These attributes are now used alongside traditional rating factors to help refine pricing, retention and underwriting strategies – and to help prevent fraud.
The power of this data in underwriting risk is evident in retrospective analysis* undertaken by LexisNexis Risk Solutions, which found that:
• Past cancellations can equate to 70% higher loss cost relativities
• An individual with two prior mid-term cancellations (in recent history) is more than twice as likely to cancel in the subsequent policy year than average
• An individual with multiple NCD entitlements at any one time has a 33% higher loss cost relativity
• A person is 60% more likely to cancel if they have, in the past, claimed a higher NCD than they were entitled to and subsequently been downgraded
• The more often people switch vehicles, the more likely they are to cancel a policy mid-term
• Those with gaps in cover show a 50% higher loss cost relativity and have a 55% increased likelihood of cancelling a policy mid-term.
Cancellations
Cancellations are a pain point for the motor market, especially for brokers. By taking a very large sample of 2m policies we identified that, on average, 15% of new business policies don't make it to the end of the term. The majority of those cancellations (86%) happen after the cooling off period, when the cost is heavier for the insurer, broker or managing general agent. It is therefore important to know at point of quote if there is an increased likelihood of cancellation. Industry-level policy history data can give a unique indication of how likely someone is to cancel, based on having history across the market.
To demonstrate this point, we built a prototype boosted model using a limited set of variables on the 2m sample to see what increased measure of cancellation relativity could be achieved. We split that research sample into train and test, and ranked the test set into deciles ranked from the highest to the lowest likelihood of cancellation.

This shows that, based on behaviour, there are huge differences in the likelihood of people cancelling a policy. The cancellation rate is only 6% in the lowest decile (0.4 x 15%), compared to 40% in the highest decile (2.69 x 15%).
In cost terms, based on a 15% cancellation rate on 100,000 new policies per year, 14% of cancellations occur during the cooling off period, at an average cost of £25. The remaining 86% of cancellations occur after the cooling off period and up to renewal; the cost for these cancellations is £85 on average, in large part due to aggregator fees. This all adds up to £1.2m per annum – and this doesn’t account for the lost income or bad debt.
Using a cancellation model means insurance providers can start to identify individuals with the highest cancellation risk at the quote stage and apply a strategy for handling those quotes. This isn’t necessarily decline or accept. Different payment terms could be set to overcome the bad debt risk, mitigating it to a large degree – perhaps through a larger deposit, or even payment entirely upfront. In contrast, those applicants identified as having a low cancellation risk could be offered a lower premium to attract and retain that business.
Switching early and switching late – the difference in risk
Now that we understand cancellation risk, we have started looking at switching behaviour, and specifically what increases the risk of higher claims losses and higher cancellations in the future from how often people switch insurer.
Motor insurance is the most ‘switched’ of all personal finance products, so we wanted to understand if it’s possible to identify not only the risk of switching, but also the differences in risk between those who switch early and those who switch late.
Quote behaviour analysis was initially only used for fraud prevention, but we are seeing an uptake in its use for pricing and underwriting, as well as cancellation modelling. Initial analysis shows that people who quote on the same day as they want their cover to start have an increased risk of claim, and we’ve seen loss cost relativities rise 32% higher than average and early cancellation risk increase by 91%.

It is only recently that the breadth and depth of policy history data shared by the market has reached the critical mass to enable a much wider understanding of the risk related to an individual’s car insurance history. Now the market has an opportunity to enrich data further and gain valuable insights by combining policy history data and quoting behaviour data to create an enhanced holistic view of the risk. This way, motor insurance providers can be in a stronger position to deliver fairer pricing to customers. The costs incurred through cancellations and claims losses can be reduced, while loyalty can be rewarded and promoted.
Alan O’Loughlin is a director of data science at LexisNexis Risk Solutions.
*This article contains results of analysis carried out by LexisNexis Risk Solutions UK Limited on available data within the LexisNexis® Motor Policy History database. The analysis was completed within a fixed period and does not purport to represent the results of any identifiable customers. The statistical analysis reported is provided ‘as is’, nothing arising from the data should be taken to constitute the advice or recommendation of LexisNexis Risk Solutions.
Picture Credit | iStock