For too long carriers have been looking to driver behaviour to provide the answer to the question of how to accurately assess auto risk. One potentially controversial view is that driver behaviour may not be the best indicator. Instead, carriers should be looking to vital contextual data that can paint a far more objective and accurate picture that can benefit both driver and insurer.
As we approach this year’s Underwriting Innovation Europe Virtual Event (June 28–30), Oleg Korol, chief technology officer of kasko2go, demonstrates why the traditional telematics parameters such as braking and driving speed paint only half the picture. He explains how analytics tools and artificial intelligence (AI) can delve into contextual, rather than telematic, data not just to improve your risk exposure but also to achieve a better rating from controlling authorities, shareholders and reinsurers.
Why should we be talking about improving traditional telematics data?
We have typically been dealing with two types of telematics models: ‘Pay as you drive’ and ‘Pay how you drive’. The first is a proven method: the number of kilometres driven on the road has a direct impact on risk. ‘Pay how you drive’, based on driver behaviour, is a different proposition. A lot of carriers have latched on to it because the technology makes this kind of analysis possible. But the problem is that no-one has checked how much behavioural data you need to come to any kind of solid conclusion.
Sometimes pilots go well, sometimes they don’t. The data is inconsistent. The problem is mathematical: there is a limit to what you can learn from a small dataset. The amount of information about driver behaviour at even the biggest providers is not enough to make any kind of consistent assessment. Instead, we need to look again at ‘Pay as you drive’ because it works, and we can upgrade it using current technology to create a product that is more accurate.
How is ‘Pay as you drive’ more accurate in the long run?
It comes down to assessing the situations people are actually driving in. It’s one thing to drive on a sunny day when you’re relaxed. It’s quite another to drive home in the rain after a long, stressful day at work.
That situation is problematic in terms of risk, and there are datasets out there that let us assess just how problematic it is. I can tell if one customer drives in the rain more often than another, or if they’re more cautious. I am able to tell if one customer habitually drives at night or during rush hour. Using accident and traffic data, insurers are able to make more objective assessments.
Telematics has typically relied on some kind of in-car installation—do you propose an alternative?
Trying to install anything on the car is a pain. The insurance industry has to work out how to keep moving forward without having to install tracking devices on everything that moves.
But with contextual data we can say for certain that the two most dangerous times to drive are during the morning and evening rush hours. We can use this insight to create an initial assessment of a driver’s risk without the need to install a tracking device. We use this understanding to create initial assessments for drivers that don’t require connectivity in-car.
How does AI enhance these insights?
AI is a multidimensional approximation system. It can take all the driving parameters—speed, road condition, time of day, weather, likely driver attitude and more—and select the weight of the parameters that we need. Driving at speed is not necessarily the most dangerous factor, and it becomes even more dangerous when there are other factors at play. Take the German Autobahn for example, where some stretches are derestricted—that is, there is no official limit. However, where road conditions are poor, busy or built up, lower enforceable limits apply. This policy recognises that speeding while surrounded by heavy traffic is several degrees more dangerous than driving at speed on an open road. Today, the insurance industry analyses each risk parameter in isolation. There needs to be a connection between all the contextual datasets. The lack of connection between different parameters is the reason, in our opinion, that insurance loss ratios haven’t gone down meaningfully despite the industry having a sophisticated, tech-driven ability to analyse the data for some time. We need to change something major within the insurance industry to get to a point where those loss ratios go down.
What are the wider implications for a contextual approach to assessing driver risk?
Quite simply, it makes roads safer. Using this information, drivers can make predictions about upcoming trips. The system can tell them alternative routes with different risk labels.
Presented with the options, drivers may choose to leave earlier to improve their risk profile while also potentially saving money in terms of fuel costs or helping the environment (less idling on congested roads) or reducing wear and tear (fewer stop-starts or incidents of harsh braking).
A contextual system can also feed information back to infrastructure, showing how different areas of roads are becoming more or less dangerous over time. Authorities can use it to assess how a new intersection impacts traffic and whether it needs to be redesigned.
What will be the most important takeaway from this session?
We need to stop thinking about ‘Pay how you drive’. The whole concept is flawed to its core. That’s not to say it’s not possible to get there but it’s hard to say what will happen sooner: that there are enough datasets to make ‘Pay how you drive’ a reality—or we’ll all be using cars that drive themselves, making it redundant. I’m betting on the latter.