A driver safety score is only useful if drivers believe it is fair. Fleet managers who have built scoring systems that generated resentment rather than behaviour change consistently identify the same root cause: the score measured inputs that were outside the driver’s control — road conditions, vehicle age, route difficulty — and then presented the outcome as a reflection of driver performance. Drivers noticed, disengaged, and the scoring system became a source of friction rather than a coaching tool.
Building a behaviour-based safety score that works requires making deliberate choices about what to measure, how to weight it, and how to present the result in a way that a driver can act on. The score is not a ranking. It is a feedback mechanism.
A driver safety score is a summary metric that combines multiple individual driving behaviour indicators into a single number or rating, typically for a defined period (weekly, monthly, quarterly). When built correctly, it gives a fleet manager a quick indication of which drivers need coaching attention, and gives drivers a single number to improve rather than a list of separate metrics to track.
A driver safety score is not a productivity metric. It should not include delivery completion rates, hours worked, or mileage — those are operational outputs that belong in a separate performance framework. Mixing safety and productivity in a single score produces a number that does not tell the manager or the driver anything specific enough to act on.
It is also not a ranking mechanism in the sense of a competitive leaderboard. Framing the score as “who is your worst driver” is the framing that destroys driver buy-in. Framing it as “here is where you are this month and here is what would move your score” is the framing that generates improvement.
The behaviours most directly linked to incident risk, and most consistently captured by in-vehicle camera and telematics systems, are:
Harsh braking frequency. The number of harsh braking events per 100 miles driven, weighted by severity (G-force). Harsh braking is the most consistent predictor of following-distance failures and late hazard responses. It is also measurable, objective, and directly within the driver’s control through forward planning and speed management.
Harsh acceleration and harsh cornering. Aggressive acceleration and sharp cornering both correlate with risk — the first with tailgating and reaction time, the second with load stability and vehicle control. Both are measured by the same telematics sensors as harsh braking.
Speed compliance. The percentage of time spent above the posted speed limit or the fleet’s internal speed policy. Not all speeding events are equal — 5mph over a 30mph limit on a quiet B-road is different from 5mph over on a 20mph residential street — but the headline compliance percentage provides a useful comparative metric when normalised by route type.
Proximity sensor alerts. For vehicles equipped with blind-spot or nearside sensors, alert frequency per shift provides a measure of spatial awareness and positioning on approach to junctions, pedestrian crossings, and reversing manoeuvres. This metric is particularly valuable for large vehicles where side and rear visibility is limited.
Driver-facing camera events. For vehicles with driver-facing cameras, distraction events (confirmed by human review, not raw AI output) — phone use, attention lapses, fatigue indicators — provide the score’s behavioural dimension beyond vehicle dynamics.
The metrics that consistently produce unfair scores, and that experienced drivers are quickest to identify as such:
Raw event counts without mileage normalisation. A driver who covers 40,000 miles per year will generate more absolute events than a driver who covers 15,000 miles, even at a lower event rate. Always normalise by distance or shifts worked.
Events from routes that generate structural alerts. Some routes — speed bumps, steep gradients, particular junctions — generate telematics events regardless of driver behaviour. If the same section of road generates braking events across multiple drivers across multiple shifts, it is a route characteristic, not a driver characteristic. Remove or discount those route-specific events before scoring individual drivers.
AI-only distraction events without human review. AI driver-facing camera systems have documented false positive rates — mirror checks, gear changes, face-touches all generate alerts on poorly calibrated systems. Including unreviewed AI events in a safety score produces a number that drivers correctly identify as unreliable, and that damages the credibility of the entire framework.
Not all behaviours carry equal risk. A safety score that weights harsh braking the same as a minor speeding event misrepresents risk. A reasonable starting weighting framework:
These weightings should be documented and shared with drivers before the score is introduced. A driver who understands that harsh braking carries the most weight, and why, can make a direct connection between their forward planning behaviour and their score. A driver who receives a score without understanding how it was calculated cannot.
The most common complaint from drivers about safety scoring is that the system is unfair to those on harder routes — urban multi-drop, high-pedestrian-density routes, busy loading areas — compared to motorway or rural drivers. That complaint is often correct.
Normalisation approaches that address this include: comparing drivers against others on the same route type rather than against the whole fleet; using a rolling baseline for each route so that route difficulty is accounted for in the expected event rate; and flagging events on known high-risk route sections for review rather than automatic inclusion in the score.
The goal is a score that measures driver behaviour, not route difficulty. If two equally skilled drivers on different routes produce very different scores because of route characteristics, the score is measuring the wrong thing.
A safety score presented as a number without context generates either relief or defensiveness, depending on whether the number is good or bad. Neither response produces behaviour change.
Effective score presentation includes: the score itself; the component breakdown (so the driver can see which metric is driving the number); the trend over the last three periods (so they can see whether they are improving, stable, or declining); and the specific event or events that most affected the score this period.
The score presentation should happen in a conversation, not in an email. A driver who receives their safety score in an inbox has no opportunity to discuss context, ask questions, or commit to a specific change. A driver who receives it in a ten-minute conversation with their line manager has all three. The conversation does not need to be a formal coaching session — for a driver with a good score, it can be a brief positive acknowledgement. For a driver whose score has declined, it is the opening of a coaching conversation.
The fastest way to destroy driver engagement with a safety score is to use it as a disciplinary trigger. “Your score fell below threshold so you are receiving a written warning” produces drivers who focus on gaming the score rather than improving their driving.
The approach that produces behaviour change is: the score tells the manager where to focus coaching attention, and the coaching session uses footage from the specific events that drove the score to have a targeted conversation. The score declines when the coaching is working — that is the outcome to track, not the score itself.
Linking improved scores to positive outcomes — recognition in team meetings, shift preferences, reduced monitoring frequency — reinforces that the score is a feedback mechanism rather than a trap. Experienced drivers who understand that a good score reduces the frequency with which they are asked to attend coaching sessions have a direct, self-interested reason to engage with the metric.
Monthly scoring gives enough data to be statistically meaningful (particularly for drivers who do not cover high mileage every week) while being frequent enough to generate coaching conversations before a pattern becomes entrenched. Weekly scores can be useful for high-mileage drivers or during the early months of introducing a scoring system, when drivers are still calibrating their behaviour to the metrics being measured.
Drivers should always be able to see their own score. Whether scores are visible across the whole fleet (as a leaderboard) is a separate question. Leaderboards can generate competitive motivation for some drivers; they can generate resentment or gaming behaviour in others. A middle ground — driver sees their own score and their percentile position in the fleet without seeing individual scores of named colleagues — gives context without the social pressure of direct comparison.
This is a signal that the score is measuring something the driver cannot control — route characteristics, vehicle issues, or AI false positives. Investigate the specific events driving the low score before concluding it reflects driver behaviour. If route-specific events are the cause, adjust the normalisation. If AI false positives are being counted, implement the human review stage before events reach the score. A consistently low score for a demonstrably competent driver is a problem with the scoring framework, not with the driver.
This is legally and practically complex. Using safety scores in pay decisions creates an incentive to game the score rather than improve the underlying behaviour. If safety performance is to be factored into compensation, using verified coaching session records (which demonstrate genuine engagement with improvement) is more defensible than using a score alone. Legal advice specific to your employment contracts and HR framework is strongly recommended before linking scores to pay.
The sequence matters. Introduce the score on a no-consequences basis for the first two months — collect data, run coaching conversations, but make clear that scores are not being used for any formal purpose during that period. Drivers see that the system works fairly before any consequences attach to it. Introduce the formal scoring period only after drivers have experienced the coaching framework and have seen that footage is used in their favour, not just against them.
Five steps to build a fair, driver-accepted safety score from camera and telematics data.
4 August 2026
Monday to Friday:
8.30am-5.30pm
