Driver scoring is easy to implement and easy to get wrong. Done well it reduces incidents, fuel use and wear. Done badly it produces gaming, resentment and a metric that measures route difficulty rather than driving quality.
What gets measured
Most systems combine some subset of:
| Event | Typically measured as | Caveat |
|---|---|---|
| Harsh braking | Deceleration above a threshold | Threshold sensitivity varies wildly by vehicle |
| Harsh acceleration | Acceleration above a threshold | Loaded vs empty changes the profile |
| Harsh cornering | Lateral force | Vehicle type dependent |
| Speeding | Over posted limit, or over a margin | Requires accurate speed limit data |
| Idling | Stationary engine time | Legitimate reasons exist |
| Seat belt | Unbelted while moving | Requires vehicle data |
| Distraction / phone use | Camera-based detection | Sensitive, needs care |
| Following distance | Camera-based | Genuinely predictive of collisions |
| Fuel efficiency | Consumption vs baseline | Strongly route-dependent |
Following distance and speeding correlate most strongly with collision risk in most fleet studies. Harsh braking is the most commonly used and the most contaminated by route conditions.
Normalisation: the fairness problem
A raw event count punishes drivers on difficult routes. A driver doing 90 urban drops will trigger more harsh events than one doing four motorway trunk runs, regardless of skill.
Minimum normalisation:
- Per 100 miles or per 100 km, never absolute counts.
- By vehicle class, since a loaded 18-tonne rigid and a small van have different physics and thresholds.
- By route type — urban, suburban, rural, motorway — where your data supports the split.
Better programmes compare drivers to peers on similar work rather than to a fleet-wide average. If your system cannot segment, be extremely cautious about publishing comparative scores.
Designing the score
Keep it simple and explicable. A driver who cannot understand their score will not act on it.
- Three to five components maximum. More becomes opaque.
- Weight by risk, not by event frequency. Speeding and following distance should dominate.
- Publish the formula. Secret scoring is corrosive.
- Show the events, not just the number. "Six speeding events, three on the A34 on Tuesday" is coachable; "72/100" is not.
- Use a rolling window — 30 or 90 days — so improvement is visible and old events age out.
The coaching process, which is the actual programme
The score is a targeting mechanism. The intervention is coaching.
- Weekly exception list — drivers whose score has deteriorated or who sit significantly above peer norms.
- Conversation within days, by a supervisor who knows the route, not by email.
- Specific evidence — the events, the locations, the context.
- Listen first. Many events have explanations: a vehicle with a braking fault, a badly designed junction, an unrealistic schedule. Some of the most valuable output of a scoring programme is discovering that the schedule causes the behaviour.
- Agree one or two changes, not a lecture.
- Re-measure and acknowledge improvement. Programmes that only ever deliver criticism stop working within months.
What destroys a programme
- Using it punitively from day one. Announce a measurement period with no consequences, then introduce coaching, then — much later, if ever — link to performance management.
- League tables published fleet-wide. They humiliate the bottom and teach the middle to game.
- Ignoring vehicle causes. A vehicle with worn brakes generates harsh braking events. Blaming the driver destroys credibility instantly.
- No supervisor capacity. If coaching does not happen, the score is a number that annoys people.
- Linking to pay too early. Incentives on a metric that is not yet trusted or properly normalised produces gaming, not safety.
- Never adjusting thresholds after installation.
Measuring whether it works
Track outcomes, not just scores:
- Collision rate per million miles
- At-fault collision rate
- Insurance claim frequency and cost
- Fuel consumption per mile
- Brake and tyre wear intervals
- Driver turnover — a programme that improves safety but drives away experienced drivers has not succeeded
Expect a visible improvement in event rates within the first few months, largely from awareness. The harder question is whether collision rates follow, and that requires enough exposure to be statistically meaningful — usually a year or more in a mid-sized fleet.
Frequently asked questions
Should driver scores be linked to bonuses?
Not until the programme is trusted, normalised properly and stable for at least a year. Incentivising an immature metric produces gaming — drivers avoiding difficult routes, coasting to avoid braking events — that can reduce safety rather than improve it.
How do we handle drivers who consistently score badly?
Investigate before acting. Check the vehicle, the route, the schedule and the thresholds. If the behaviour is genuine after those checks, structured coaching with clear expectations and follow-up is the appropriate path, escalating only if there is no improvement.
Is harsh braking a good safety metric?
It is weakly correlated with collision risk and heavily contaminated by route conditions. Following distance and speeding are better predictors. Use harsh braking as a coaching prompt rather than as a headline score component.
Do drivers game the system?
If the metric is used punitively, yes — predictably and creatively. Transparency about how scoring works, and using it for coaching rather than punishment, substantially reduces the incentive to game.
Should we tell drivers about every event?
Give them access to their own data continuously, and discuss patterns rather than individual events. Alerting on every event creates noise and defensiveness; a weekly summary the driver can review themselves is more effective.
