Every telematics platform can detect harsh braking. Almost none of them turn that into something a fleet manager does anything about. The gap is not detection; it is that a list of ten thousand events is not actionable and a ranked list of forty drivers is.
Events are the raw material, not the product
A month of a hundred vehicles produces tens of thousands of behavioural events. Nobody reads them. They get exported to a spreadsheet once, discussed in a meeting, and never opened again.
What changes behaviour is a comparison: this driver against that driver, this month against last. That requires three things the events do not have on their own — normalisation, weighting, and a trend.
Normalise, or you are just measuring who drives most
The first mistake is ranking by event count. A driver doing 4,000 km a month will out-brake one doing 800 km, and telling them so is telling them they work more.
Everything has to be per unit of exposure — per hundred kilometres, or per hour of driving. Once you do that, the ranking changes completely, and it becomes defensible in the conversation that follows, which is the actual test.
What goes into the score, and in what proportion
Weighting is where most implementations get vague. Routly’s eco score is explicit about it:
| Component | Weight | What it measures |
|---|---|---|
| Fuel efficiency | 40% | Consumption against a baseline for the vehicle |
| Idling | 20% | Engine running, vehicle stationary |
| Harsh events | 20% | Acceleration, braking, cornering |
| Speeding | 15% | Over the limit, by margin and duration |
| Over-revving | 5% | RPM above the efficient band |
Fuel is the largest because it is the outcome, and the rest are largely causes of it. A score dominated by harsh braking measures a driving style; a score dominated by fuel measures the thing the fleet is paying for. That choice is arguable and you should make it deliberately — a fleet whose priority is accident reduction rather than cost would reasonably raise harsh events and lower fuel.
One implementation detail with a real consequence: not every vehicle reports RPM. When it is absent, its 5% is redistributed proportionally across the other four components rather than treated as a zero. Treating a missing signal as a perfect score flatters drivers of older vehicles; treating it as a failure punishes them. Redistribution is the only version that compares like with like.
Violations get their own ladder
Alongside the eco score, violations map to a 1–10 scale through a penalty ladder that is deliberately non-linear:
| Total penalty | Score |
|---|---|
| 0 | 10.0 |
| ≤ 17 | 9.0 |
| ≤ 38 | 8.0 |
| ≤ 67 | 7.0 |
| ≤ 107 | 6.0 |
| ≤ 167 | 5.0 |
| ≤ 267 | 4.0 |
| ≤ 467 | 3.0 |
| ≤ 1067 | 2.0 |
| above | 1.0 |
The bands widen as they go, which is the point. The difference between a driver with zero violations and one with seventeen points is worth a full grade; the difference between 500 and 900 is not, because both are already at the end of the conversation. A linear scale wastes all of its resolution on the tail where nothing changes.
What a score must never be
Three rules, and the third is not optional.
Not a league table on the wall. Public ranking produces gaming — coasting to a stop from further back, avoiding difficult routes, taking the flat way round. You get a better score and worse operations.
Not a comparison across incomparable work. Urban delivery against motorway trunking is not the same job. Compare within a cohort or normalise for route type; a driver who scores badly because their round is all roundabouts has learned only that the system is unfair.
Not an automatic consequence. Behavioural scoring of a named person is personal data with weight of its own, and an automated decision producing legal or similarly significant effects engages Art. 22 of the GDPR. Scoring used for coaching is defensible; scoring that automatically triggers discipline is a compliance problem and a labour-relations one. Human in the loop.
What actually moves the number
From fleets that get a result, the pattern is consistent and unglamorous:
- Show the driver their own score, before anyone else discusses it. Most improvement happens here, without a conversation.
- Talk to the bottom decile, not everyone. A monthly five-minute conversation with four drivers beats a memo to forty.
- Pick one component at a time. “Your idling is the highest on the fleet” is a thing a person can change this week. “Your score is 6.2” is not.
- Show the trend, not the snapshot. Improving from 4 to 6 deserves more recognition than sitting at 7, and recognising it is what keeps the exercise alive past month two.
The measurement that matters
The score is not the outcome. Fuel consumed per hundred kilometres across the fleet is the outcome, and it is what you should be watching to decide whether any of this is working. If scores improve and consumption does not, the score is measuring the wrong thing or people are gaming it — either way the answer is in the fuel number, not in the score.
The KPI board exists to put those side by side: distance by vehicle, harsh braking by driver, consumption by month. The point is the correlation, not any single chart.
Frequently asked
How many events do I need before a score means anything?
Enough distance for the normalisation to be stable — a few hundred kilometres a month, minimum. Scoring a driver on 40 km of data produces noise that will be argued with, correctly.
Can I change the weights?
They are constants in the scoring engine, so yes, and the redistribution logic handles a missing component. Decide once, then leave them alone: a weighting that changes every quarter makes trends meaningless, which destroys the only view that drives improvement.
What about accidents rather than style?
Harsh events correlate with risk but do not predict individual accidents well enough to act on. Treat behavioural scoring as a fuel and wear programme that has a safety side effect, and be honest about that framing internally.
Do drivers have to be told?
Yes. Transparency is an obligation, not a courtesy, and the programme works better when they can see their own numbers anyway.
Can I try it?
The demo has a KPI board with harsh braking by driver on generated data. The weights above are what the software ships with.
