The Insurer That Knows How You Actually Drive
Motor insurance was priced from proxies for risk including age, postcode, and vehicle. Telematics measures the driving directly, which improves pricing accuracy and raises questions the proxies never did.
Pricing From Proxies
Traditional motor insurance pricing uses characteristics correlated with claims: age, driving history, vehicle type, annual mileage as declared, postcode, and in some markets credit based scores.
None of those measures driving. They are proxies, and they are statistically valid at a population level and frequently wrong about individuals.
A careful twenty year old pays substantially more than a careless fifty year old, because the model prices the group. A driver in a high claim postcode pays for their neighbours.
Telematics replaces proxies with observation. A device, or increasingly the driver own phone or the vehicle built in connectivity, records how the car is actually driven.
What Gets Measured
| Variable | Relationship to Risk |
|---|---|
| Mileage driven | Strongest single predictor |
| Time of day | Night driving carries higher risk |
| Harsh braking and acceleration | Correlated with claims |
| Cornering forces | Correlated with claims |
| Phone handling while driving | Strongly correlated |
Mileage is the most important and the most obvious. Exposure to risk is roughly proportional to time spent driving, and self reported annual mileage has always been unreliable.
Behavioural variables add further predictive power, and the evidence on which ones matter is stronger for some than others. Harsh braking frequency has reasonable predictive validity. Some marketed metrics have considerably less and persist because they are easy to measure and easy to explain.
The largest and least contested gain from telematics is simply knowing how far somebody actually drives. Everything else is refinement on top of a variable the industry had been guessing at for a century.
The Selection Effect
An underappreciated feature is that offering a telematics product changes who buys it, before any data is collected.
Drivers who believe they drive carefully, and those who drive very little, opt in expecting a discount. Drivers who know they drive badly do not.
That produces favourable selection into the programme independent of the measurement, which is why early telematics books performed better than the pricing model alone would predict.
It also means the discount offered to attract participants is partly funded by the selection rather than by the behavioural improvement, which matters when a market moves toward telematics as standard and the selection effect disappears.
Does It Change Behaviour
Insurers claim that feedback improves driving, and there is reasonable evidence that it does, at least temporarily.
Drivers receiving scores and feedback show measurable reductions in harsh braking and speeding during the monitoring period. Whether the improvement persists after monitoring ends is less clear, and some studies suggest a decay.
The mechanism appears to be attention rather than skill, which is consistent with the finding that the effect is strongest early and diminishes as the novelty of being scored wears off.
The Privacy Question
The genuine concern is not that the insurer knows how hard you brake. It is that continuous location data reveals a great deal more than driving quality.
Where somebody sleeps, works, worships, receives medical care, and spends their weekends is all derivable from location traces. That information has value beyond insurance and is a target for subpoena, breach, and secondary use.
Regulatory frameworks differ substantially. Data protection regimes requiring purpose limitation constrain use to the insurance purpose disclosed. Where such frameworks are weaker, the constraints are contractual and depend on what the policyholder agreed to in terms they did not read.
A related issue emerged where vehicle manufacturers collected driving data through built in connectivity and shared it with data brokers who sold it to insurers, in some cases without drivers understanding that enabling a vehicle feature had that consequence. That practice attracted regulatory attention and litigation, and it is a meaningfully different situation from a driver choosing a telematics policy.
The Fairness Argument in Both Directions
Telematics is defended as fairer because it prices individual behaviour rather than group membership, which reduces reliance on characteristics like postcode that correlate with protected attributes.
The counterargument is that some measured variables also correlate with circumstance rather than choice. Night driving is riskier and is also when shift workers travel. Urban driving produces more braking events regardless of skill. Pricing those accurately transfers cost onto people whose driving pattern is determined by their job rather than their care.
Both points are correct. Replacing a proxy with a measurement improves accuracy and does not automatically improve fairness, because the thing being measured accurately may itself be a function of circumstance.
The Bottom Line
Telematics prices motor insurance from observed driving rather than from proxies, and the largest gain is simply measuring mileage properly. Early performance benefited from favourable selection as much as from measurement, the behavioural improvement is real and appears to fade, and the substantive concern is that continuous location data reveals far more than driving quality. Accuracy and fairness are not the same thing, and pricing risk correctly can still allocate cost to people whose driving pattern was never a choice.