Eco-driving: what is worth tracking and what is noise

In an eco-driving report we looked at recently, harsh braking had fallen by 99%
in four months. The kind of result you put on a slide.

Nothing had changed behind the wheel. The data feed had broken.

This article is about telling one from the other — because an eco-driving score
you cannot trust is worse than no score at all: it makes you take decisions about
people based on a chart that is measuring something else.

The problem with the eco score

Almost every telematics provider gives you a “score”. A number between 0 and 100,
built from several indicators, with weightings you usually cannot see.

That makes it good for exactly one thing: tracking the trend of the same fleet,
on the same provider, over time. And poor at nearly everything else. You cannot
compare your score to another company’s, you cannot compare across providers, and
you cannot tell from the score what needs to change.

The score is a thermometer. It is not a diagnosis.

What is worth tracking

Three things, in order of how much trust they deserve.

1. Consumption per 100 km — calculated properly

It is the only indicator that translates directly into money, and the only one
you can verify from an independent source: the fuel receipts.

The trap is in the arithmetic. If you add up each truck’s daily average and
divide by the number of days, you get an average of averages — which is not the
same as reality, because a day covering 800 km counts the same as one covering 40.

The correct form is distance-weighted: total litres divided by total
kilometres over the chosen period. The gap between the two methods is usually a
few tenths of a litre — which is precisely the size of the “improvements” people
report.

2. Idling share

An engine running without moving the truck burns fuel and produces nothing. Easy
to measure, hard to argue with, and it changes through habits rather than
investment.

It is also where differences between drivers are real and explainable: winter,
loading bays, traffic. A percentage that rises in winter is not a discipline
problem.

3. The trend of the composite score — not its value

If fleet score moves from 77 to 80 over twelve months, something got better. What
those three points are worth, nobody can tell you — and anyone who does has a
product to sell.

Use it as a direction signal, not as a figure to report onward.

What is noise

Event counters, when you do not know the coverage

Harsh braking, harsh acceleration, over-revving — all counted. The problem is not
the indicator; it is that the count depends on a data feed that can fail
silently.

The case from the top of this article, on a fleet of around 33 trucks tracked for
thirteen months: average braking events per 100 km fell from about 78 to 0.4.
The explanation is not driving. The share of truck-days reporting exactly zero
braking events rose from 4.5% in April to 63% in August. The trucks were not
reporting events, and the zeros went into the average as though they were
performance.

Same period, same provider, same fleet. The chart showed spectacular progress. It
did not exist.

Indicators whose coverage is shrinking

On the same fleet: cruise-control usage appeared to climb nicely, from 55.7% to
65.9%. Except the number of trucks still reporting it had fallen from 634
truck-days in May to zero in August.

A percentage calculated on a sample that is shrinking and selecting itself is not
a trend. It is a statistical illusion.

How to check in two minutes

Before believing any trend in an eco-driving report, ask two questions:

1. How many truck-days is each month calculated on? If that number is
   falling, the rest of the chart means nothing.
2. How many of the values are exactly zero? A healthy indicator has a
   distribution. One filling up with zeros is a feed dying.

If your report cannot answer those, it is not a performance report. It is a chart.

What a real improvement looks like

On that same fleet, over those same thirteen months, two things survived checking:

• the average score rose from 77.0 to 80.4;
• idling fell from 11.2% to 10.5% of engine time.

Both had complete coverage in every month — every truck-day reported, no gaps.
They are small improvements. They are also real.

The third finding is the most interesting: consumption per 100 km did not
move — 26.99 litres at the start, 26.97 at the end. Over the same period the
fleet grew from 23 trucks to 33.

A fleet growing by nearly half usually sees its average drift, because the office
runs out of attention before it runs out of capacity. Holding consumption flat
while the fleet grows does not look like a win on a chart. It is one.

What you can do on Monday morning

1. Recalculate consumption distance-weighted, not as an average of averages.
   Compare the result with what you report today.
2. Ask for the coverage next to every indicator — how many truck-days, per
   month. If your provider will not give it to you, that is your answer.
3. Check the share of zeros on the event counters.
4. Do not compare scores across providers. They are not measuring the same
   thing.
5. Talk to drivers about idling, not about the score. It is the one indicator
   they can influence directly and understand without explanation.
6. Reconcile telemetry litres against the receipts. If they disagree, fix that
   before analysing anything else.

Frequently asked questions

So eco-driving scores are useless?
They are useful as a trend signal on one fleet with one provider. They are not
useful for comparing companies, assessing an individual driver, or reporting to a
customer.

How often do these feeds break?
More often than people think, and almost always silently — no error, no message.
An indicator that suddenly becomes very good deserves more suspicion than one
that gets worse.

Can I use the score for driver bonuses?
Carefully, and never on its own. A composite score whose weightings you cannot
see, running on a feed that can degrade, is a fragile basis for somebody’s pay.
Idling and verified consumption are easier to defend.

What do I do if I find my indicators were broken?
Mark the period as non-comparable and move on. The temptation is to correct it
retroactively; it is more honest to state the date from which the figure is good.

If you want to check together which indicators in your report survive the
questions above, book a walkthrough and we will look at your own data.