Risks and Rewards of Usage-Based Insurance

The insurance business has always been a challenging one. Insurers estimate the risk they will have to pay out in claims and balance that amount against the premiums they charge customers for coverage. Risk modeling, like any type of forecasting, is only as good as the data signals coming into the model. For insurers, the nature of those signals has achieved an unprecedented level of fidelity in recent years, or at least it has appeared to. 

Traditionally, risk modeling for auto insurance involved correlating historical data about claim frequency and severity to driver demographics such as age, gender, location, vehicle type, and other criteria. Initial premiums were determined by where a driver fell into those categories and were then adjusted over time based on their history of crashes and traffic citations. 

Enter Connectivity

The advent of cellular data connectivity in the past decade has transformed that paradigm by giving insurers much more granular access to driver behavior data. Insurers began offering drivers connected dongles that plugged into the onboard diagnostic port to track acceleration, braking, and speed patterns that were transmitted back to insurers’ servers in exchange for premium discounts for drivers. 

More recently, with nearly all new vehicles having embedded connectivity, automakers have begun partnering with insurers and risk management companies such as Verisk to provide the information directly when a driver opts-in. For the most part, insurers have used the carrot of premium discounts to entice drivers to opt into these data collection programs. However, it was inevitable that eventually, as connectivity became ubiquitous, the stick of higher premiums tied to certain behaviors would also be applied. 

Garbage In and Garbage Out

The fundamental problem with any sort of modeling is that it can only ever be as good as the input signals. If those signals don’t accurately represent what is happening, they will provide a false picture of what might happen in the future. Such is especially true if the signals go beyond direct measurement of what the driver is actually doing to recording what the vehicle is doing independently of the driver’s actions. 

For example, tracking how aggressively the driver presses the brake or accelerator pedal may be a good indicator of their probability of getting into certain types of crashes. However, tracking activations of traction or stability control can be more ambiguous and indicative of things beyond the driver’s control. An even more extreme example is tracking the frequency of forward collision warning (FCW) or automatic emergency braking (AEB). If those active safety driver assistance systems work well, they may well be a very good risk predictor. If a good FCW/AEB system is being activated frequently, it may be a sign that the driver is often following too closely.

The problem arises when those systems are unreliable but are nonetheless used to increase a driver’s insurance premium. Tesla vehicles have a history of false positive FCW and phantom braking AEB that is being investigated by the National Highway Traffic Safety Administration. But some Tesla owners that get their insurance through the automaker are finding their premiums being negatively affected by bad data coming from systems they don’t control, and one owner is suing Tesla over it. 

Insurers like the idea of getting more granular and real-time data because it helps them manage revenue and costs. Automakers like the idea of sharing that data as a potential new revenue stream, and drivers like the idea of lower premiums for safe driving behavior. But all of that goes away if the parties involved don’t ensure that data is actually valid.