We have not had a serious accident in years. Why does the premium keep rising?

Because the premium was never mostly about the serious accident. A fleet manager who has avoided a catastrophic loss for years and watches the renewal climb anyway is reading the market correctly and reading the wrong number. Insurers price fleets first on how often something happens, and only second on how bad it was.

Frequency and severity are two different bets

Severity is what a claim costs when it happens. It is driven by things that are close to random from an underwriter’s point of view: the speed of the other vehicle, whether anyone was standing at the kerb, which hospital, which lawyer, which court. A fleet of any size will have a severe loss eventually, and nothing in its management predicts when.

Frequency is how many incidents the fleet produces per vehicle, or per distance travelled, in a year. It is stable. It repeats. It describes the organisation rather than the accident — the routes, the schedule, the hiring, the training, the vehicle turnover, the pressure on the last delivery of the day. An underwriter who knows a fleet’s frequency knows something about next year; an underwriter who knows only that a fleet had one very expensive claim four years ago knows almost nothing.

So the record that prices your renewal is mostly a count. A year of small collisions with no injuries can cost a fleet more at renewal than a single large loss, and fleet managers find that backwards until they see the arithmetic from the other side: the large loss is one draw from a distribution nobody controls, and the small ones are a description of how the operation runs.

There is a second reason, less discussed. Small claims are expensive relative to what they pay. Every one carries handling cost, and each opens a reserve that sits against the fleet’s record until the file closes — including reserves for injury claims that never materialise. A fleet with a long tail of small open files looks worse on paper than its actual paid losses, which is one of the few things a fleet can correct by asking for it.

What the record is being compared against

A fleet’s loss record only means something next to something else. Underwriters read it against the same fleet’s earlier years, against fleets of a similar class and use, and against the premium — the ratio of what has been paid out and reserved to what has been paid in. Where that ratio sits, and what a particular insurer considers acceptable for a particular class, is commercial and not public, and no article can give you the figure.

What is worth knowing is which of the inputs a fleet can move. The count is one. The reserves are another, and they are moved by pressing for review rather than by argument about the incident. The exposure base — the declared vehicles, the declared use, the declared distance — is a third, and a fleet whose declared distance no longer matches what it drives is being rated against a number from a different business.

What telematics actually changed

For most of the history of commercial motor insurance, an underwriter looking at a fleet with a thin record had to rate the class: this kind of vehicle, this kind of work, this region, this many drivers. It was rating by proxy, and it meant a careful fleet subsidised a careless one of the same shape.

Telematics loosened that, and it did so from an unexpected direction: not by detecting accidents but by counting the driving between them. A telematics device records speed, acceleration and braking, cornering, distance and the time of day the driving happens, and those events accumulate in weeks where claims accumulate over years. Driver behaviour derived from that record — a driving score, acceleration, the telematics data itself — now sits among the rating factors European motor insurers reported using to a supervisory review, in the group that review rated high for its relation to risk. Whether event rates predict claim frequency as closely as the market says they do is not something this piece has read evidence for, and no figure for it appears here. What the same review does record is the consequence a fleet cares about: firms gave the example of young drivers with no claims history being easier to price where richer data is available, and a fleet too small for its own record to be credible is in that position at every renewal.

Cameras changed a different thing. A forward-facing clip of a collision settles liability arguments that used to be settled by whichever account sounded more confident, and the fleet benefit shows up in the claims that are closed as not-at-fault rather than shared. That is a frequency benefit as well as a cost one, because a shared-liability outcome is an entry in the record and a not-at-fault outcome argued successfully is a weaker one.

Driver scoring is where the two streams meet, and where the questions get harder. A score is built from proxies. Harsh braking is a reasonable proxy for following too closely and it is also what happens on a route with badly phased traffic lights; a model comparing drivers on the same route can separate those, and a model comparing drivers across different routes cannot. A score used to coach is useful and a score used to discipline is a decision about a person’s livelihood made partly by a model, which is regulated as such in a growing number of places.

And the data is employees’ data. Continuous location, driving behaviour and — with inward-facing cameras — the driver’s face at the wheel are personal data under most regimes that regulate it. What that requires before installation, whether it includes consulting a works council or employee representatives, whether an impact assessment is mandatory, and what retention period is defensible are questions with local answers; the regulator for where you operate appears in the rules for your jurisdiction below.

A fleet that installed cameras for claims defence and then used the footage for performance management has changed the purpose of the processing, which is the point at which several of these regimes start to care. Deciding what the system is for, in writing, before it is switched on, is cheaper than deciding it during a dispute.

The trap that looks like good management

Frequency can be lowered, and it can also be hidden, and the two are indistinguishable in the first year.

A fleet that discourages reporting gets a cleaner record and loses the thing the record was for. The collisions still happen; they are now repaired out of maintenance budgets, undocumented, unanalysed, and invisible at the point where a pattern would have shown that one route, one depot or one shift produces most of them. Frequency management depends entirely on drivers telling the fleet about things the fleet would rather not hear, which is why a debrief that reads as a disciplinary hearing costs more information than it collects.

The insurance consequence is worse than the operational one. Late notification of an incident that later produces an injury claim can affect cover, and an insurer that discovers a pattern of unreported damage during a claim investigation is no longer only assessing that claim. Under-reporting is not a quieter version of good risk management; it is the same record with the evidence removed.

What we cannot tell you is which behaviours in your fleet drive its frequency. That answer is in your own incident file and nowhere else — route, time, driver tenure, vehicle, the assignment being run — and it exists only if somebody recorded those fields at the time. The fleets that reduce frequency are not the ones that bought the best system; they are the ones that could still read their own incidents two years later.