Ninety per cent of what? What the driver-error statistic really shows

It appears in vendor decks, conference slides and insurer marketing in much the same form: around 90% of commercial motor claims are caused by driver error, and most of that is distraction. Traced back to its sources, the claim does not hold as stated. It merges studies of different populations, using different definitions and measuring different outcomes, and none of them measures commercial insurance claims.

The figure matters because it does commercial work. It is used to argue that a camera, a coaching programme or a monitoring platform addresses the bulk of an insurer's motor loss. A pricing, partnership or investment decision built on it rests on a share of losses that nobody has measured.

Where the number comes from

Five studies usually sit somewhere behind it. Each is credible in its own terms. None says what the headline says.

Four ways the claim goes wrong

1. The measure is a critical reason. Crash causation studies record the last failure in the chain of events before impact. A driver who fails to react to a vehicle pulling out is assigned the critical reason, even if the other vehicle created the hazard. NHTSA is explicit that this is not a finding of cause or fault.
2. The population is mostly general traffic. The largest numbers come from studies of the general driving population. Professional drivers face different tasks, schedules, vehicles and supervision, and the fleet-specific studies use far smaller samples.
3. The category bundles several failures. The composite figures bundle misjudgement, fatigue, impairment and distraction together. Several factors are often recorded for the same crash, so the categories cannot be summed or swapped for one another.
4. The outcome is a crash or an event. A commercial motor book includes third-party-at-fault collisions, theft, weather and parked damage, and its cost depends on severity, coverage and settlement as much as on frequency. How often a factor appears before an impact is not the share of losses that removing it would prevent.

What can be said with confidence

Rejecting the headline does not mean distraction is unimportant. The evidence supports a narrower and more useful statement:

The more precise findings are also the more actionable ones. In naturalistic truck data, using or looking at a dispatch device carried roughly ten times the odds of a safety-critical event [4]. That points to a specific, controllable source of risk in the operation itself, which is far more useful to a fleet, a risk consultant or an underwriter than a single percentage attached to the whole of driver behaviour. A companion piece sets out how that risk is designed into the working day and what to change first.

Questions to ask when the figure appears

• What is the denominator: claim count, paid losses, severe collisions or reviewed camera events?
• Which country, period and vehicle classes does it cover?
• How were third-party fault, theft, weather and parked damage treated?
• Was distraction observed, inferred or coded after the event, and by whom?
• Is the figure the share of crashes where a factor was present, or the share of loss that a specific intervention has been shown to reduce?

Why it matters for technology partners and insurers

For technology companies, an inflated statistic is a liability in front of the audience that matters most. Underwriters and actuaries who know the source studies will discount the rest of the case once they see it. Measured against the proof ladder, the 90% claim sits at the level of correlation drawn from other populations. A filed rating factor needs credible causal evidence in the insurer's own claims experience. The route there is a well-designed pilot.

For insurers and brokers, the discipline is to separate three questions about any driver-behaviour factor: is it associated with risk, can it be measured reliably and fairly in this portfolio, and does changing it reduce losses? Evidence for the first answers neither of the others.