A driver-scoring report we reviewed with a customer this summer ranked the fleet's best driver first, with a perfect score. That driver had covered 112 km in the month. The driver in second place had covered 2,400 km, spent five days a week on the road, and had two harsh-braking events to show for it.
Nobody who runs a fleet believes the first driver is better. But the report said so, in a large font, at the top of the page. And once a ranking is printed, it gets used: in the Monday meeting, in the bonus conversation, in the quiet decision about who gets the new van.
Small denominators make loud numbers
Almost every fleet metric is a ratio. Litres per 100 km. Idle time as a share of engine time. Events per 1,000 km. Ratios behave badly when the denominator is small. Two trips with the engine left running at a delivery door will give one driver a 40% idle share. A single motorway run will give another a fuel figure no one else can match. Neither number is wrong. Neither means anything yet.
The same problem hides inside fleet averages. If a third of your vehicles have no CAN-bus connection, they report distance but zero fuel. Divide total fuel by total kilometres across the whole fleet and you get a consumption figure that looks impressive and is simply false. Every vehicle that does report fuel then looks like an outlier against it.
What we do instead
Emris computes driver and vehicle scores over a trailing 30-day window, and applies two rules before anything is published.
A minimum sample. No score is published under 5 trips or 100 km in the window. Below that, the driver or vehicle is shown as provisional: visible on the page, never ranked, never coloured green or red. When the sample grows, the score appears on its own.
An itemised ledger. Every score is a sum of visible parts: fuel against the fleet average, idle share, elevated speeds, driving events for drivers, weekday utilisation for vehicles. Each part shows the points it contributed and the data it was computed from. If a vehicle reports no fuel, the fuel component is absent rather than invented, and the remaining components are re-weighted so the total still reads on a 0 to 100 scale.
The fleet average itself is computed only over the vehicles that actually report fuel, and the answer says so.
Why this matters more than it sounds
A score is not an analytics feature. It is the start of a conversation with a person. A manager who puts a ranking in front of a driver is spending trust, and a ranking that turns out to rest on 112 km spends it all at once. The next ranking, even a correct one, is argued with.
A score that can be opened up and explained is one a manager can defend, and one a driver can act on. "Your idle share is 18%, mostly at the depot between 07:00 and 07:30" is a conversation. "You're seventh" is not.
Three questions to ask any scoring report
- What is the minimum sample before a score is published, and what happens below it?
- Can I see the points behind the total, and the trips behind the points?
- When data is missing for a vehicle, is that component dropped, or is it silently scored as perfect?
If the answers are "none", "no" and "we're not sure", the ranking on page one is decoration. We would rather show you a smaller table you can trust.