What 21.6% empty running actually costs your fleet

A fleet director reads that 21.6% of EU road-freight vehicle-kilometres ran empty in 2024, multiplies it by their own annual mileage, and takes the answer into a board meeting.

Two things are wrong before anyone has looked at the fleet: that percentage does not describe their operation, and the cost it implies is not the cost they would save.

The EU average is two different numbers averaged into a useless one

Eurostat’s 2024 figure splits 25.8% empty on national journeys against 12.6% on international ones. The country spread is wider still: Cyprus 43.7%, Ireland 34.7%, Greece 34.4% and Austria 34.0% at one end; Denmark 7.5%, Lithuania 11.0% and Belgium 11.6% at the other.

The mechanism is distance. The average national journey moves a tonne 97.3 km; international, 604.3 km; cross-trade, 722.0 km. A repositioning leg of roughly fixed length is a large fraction of a 97 km cycle and a small one of a 722 km cycle. Empty running is mostly geometry, not discipline.

That decides whose number you should be quoting. Over 77% of Bulgarian haulage activity is international and cross-trade alone exceeds 44% of it, so the honest reference point for a Bulgarian operator is 12.6%, not 21.6% — and against 12.6%, a supplier promising to halve your empty miles is not being ambitious, they are being arithmetically illiterate. Two to five percentage points on the addressed corridors over two quarters is what the published range supports.

A figure quoted without the national/international split was not read from the source.

Filling an empty kilometre and deleting one are financially unrelated

This gets reported wrong, and always in the direction that flatters the project.

There are exactly two ways the empty share falls. You find freight for a leg you were going to drive anyway — the numerator falls, the denominator does not. Or you eliminate the repositioning leg, and both fall because the truck never turns a wheel. Same headline percentage. Not the same money.

A converted empty leg earns contribution margin: the freight rate, minus the extra fuel a payload costs, minus any collection detour at its full marginal cost, minus waiting at the ramp. It saves no toll. Germany’s LKW-Maut charges 34.8 ct/km for a 5-plus-axle Euro VI CO2-class-1 combination over 18 t — 15.5 ct infrastructure, 2.3 ct air pollution, 1.2 ct noise, 15.8 ct CO2 at €200 per tonne — and since 1 July 2024 the assessment basis has been technically permissible maximum laden mass, not gross vehicle weight and certainly not the load on board. The vehicle pays the same rate full or empty.

An eliminated leg saves the whole marginal cost of a kilometre: toll, fuel, tyres and maintenance, plus the driver hours it would have consumed — which, at a 12–13% European driver vacancy rate, is the scarcest of the four.

Report them as two lines: contribution margin captured on legs previously run empty, in euros; and kilometres not driven, priced at a blended toll rate from your own EETS or Toll Collect billing plus fuel at the pump. Sum them and a matching project can claim a toll saving it never produced.

Do the scale honestly. A 100-truck international fleet at roughly 125,000 km per vehicle per year drives 12.5 million vehicle-kilometres; three points of empty running is 375,000 km. Convert all of it to loaded and the toll line does not move while the fuel line rises. Eliminate all of it and German toll alone would be over €130,000 — except that only part of that mileage runs on the German network, so the blended rate is the number you are allowed to use, and it is lower. Diesel went from EUR 1.56/l at end-Q4 2025 to EUR 1.96/l at end-Q1 2026, moving the second line further in one quarter than most optimisation projects move it in a year.

Most fleets cannot measure their own empty running at all

The TMS records planned trips; Eurostat measures realised vehicle-kilometres. The gap between them is where the interesting kilometres live: repositioning nobody raised an order for, the detour to a wash bay, the run back from a refused delivery.

The measurable version is reconstructed from the vehicle, not the plan — rFMS positions plus digital tachograph mass-memory downloads, differenced across odometer readings at load and unload boundaries, each segment classified laden or empty against consignment records.

State the instrument’s limits while you do it. ACEA’s rFMS 5.0.0, published 25 July 2025, guarantees vehicle position only once every 15 minutes and vehicle status only once every 60, with two weeks’ minimum storage. At motorway speed a 15-minute floor is roughly 20 km of unobserved path, so segments are classified at event boundaries and measured on odometer deltas, never by drawing a line between two pings. Smart Tachograph 2 helps here: it records border crossings automatically via GNSS with Galileo OSNMA authentication, operational since 24 July 2025, so kilometres arrive already attributed by country.

Until that baseline exists there is nothing to improve, and a percentage-point claim spanning the moment the measurement changed is not a result.

Planner acceptance below 70% is a master-data readout, not an attitude problem

Route optimisation bought and quietly abandoned is the most common story in European dispatch, and the pattern is documented: planners over-constrain the solver until it reproduces existing routes, hand-correct its output at 5–15 minutes an iteration, and settle into using the licence as a kilometre calculator.

Planner acceptance rate — the share of produced routes dispatched unmodified — is what exposes it. Below roughly 70%, no savings will materialise whatever the solver reports internally.

Read a low rate as a diagnostic, not a verdict: log every planner edit with the field that caused it. What comes back is almost always master data, and usually the same four fields. Service time, the modelled fixed duration at a stop, is the most commonly wrong field in any route model. Appointment lead times living in someone’s inbox instead of the site record. Geocodes drifted to the street centreline rather than the gate, which puts the driver on the right street at the wrong entrance. And a cases-per-pallet figure still reading 48 after packaging moved to 42, producing a load plan that fails physically on the dock.

One class of override should never be counted against the planner. A route that pushes a driver past 4.5 hours’ driving with no verified parking for the 45-minute break, moves a vehicle in week seven of its eight-week return-to-base clock, or triggers a posting declaration on a cross-trade leg under Directive (EU) 2020/1057 deserves to be rejected. Those constraints belong inside the solver; if they are checked afterwards, the acceptance rate is measuring the model’s ignorance and the planner is the last line of defence.

The same discipline applies to the backhaul list. Rank candidate return loads by contribution margin after toll and fuel rather than rate per kilometre, test each against remaining duty time and the cabotage counter — three operations in seven days, then a four-day cooling-off — and show the constraint that killed every rejected option. A list a dispatcher cannot interrogate is ignored by week two.

Where this stops being the right lever

Closed-loop fleets — hub-and-spoke distribution, fixed milk runs, dedicated contracts — carry structural empty running no matcher removes. The return leg is empty because the network says so; the lever is network design or trailer swaps, not freight matching.

The market is moving against the easy version too. The Ti/Upply/IRU Q1 2026 benchmark puts contract rates at index 140.1, up 8.9% year on year, while spot fell to 132.3, down 2.0% — a return load covered on TimoCom or Trans.eu earns a thinner margin than the same kilometre under contract. Bulgarian road freight fell 18.6% in 2024, the sharpest contraction of any member state.

And if driver turnover is above 25%, fix that first: extra return loads compete for duty hours you cannot crew.

One boundary about us. Palamed has not delivered a logistics engagement — our four projects are a European car marketplace with over 300,000 listings, a platform for an AI automation agency, the Ministry of Education and Science dictionary at beron.mon.bg, and email-marketing automation for a beauty brand. Everything above is the published pattern and the specification we would hold a fleet project to: the split baseline, the two cost lines, the acceptance-rate gate. If a supplier quotes you one empty-miles percentage, ask which Eurostat column it came out of.