Most visibility programmes are judged on whether the integration works. The integration usually works. What fails is the data coming through it, and it fails quietly: positions that are four hours old and look current, a truck parked in the same coordinate for two days because the box died rather than because it stopped, a plate on the map that is not the plate that took the load. None of that shows up as an error in freight visibility software. It shows up as a dispatcher who stopped believing the screen.
Telematics data quality is measurable, and the measurements are worth more than the feed count every vendor reports.
Where the ping comes from
Four sources, and they are not interchangeable.
The telematics box. Hardwired, ignition-aware, reporting every one to fifteen minutes while the engine runs and usually falling silent or dropping to a very low rate when it stops. This is the best source and the one most European carriers already have, because the fleet management contract and the tachograph download put it there. Detail on what that estate looks like is in the guide to fleet telematics for European hauliers.
The tachograph. Legally mandated, excellent for driving and rest time, and not a position feed. It tells you where a driver started and stopped a period, on a download cadence measured in weeks. Useful for compliance, useless for a dispatcher.
The driver smartphone app. The fallback for carriers with no box, and the source with the widest quality spread. It depends on the driver installing it, granting background location permission, keeping the phone charged and the operating system not suspending it to save battery. Coverage of sixty per cent on an app-based carrier is a normal result, not a bad one.
The trailer tracker. Battery or solar powered, and therefore deliberately frugal: one to four pings a day is a common configuration. Excellent for asset location and trailer pool management, misleading if anybody treats it as a live shipment feed.
The first quality question about any feed is which of these four it is, per carrier. A coverage number that mixes them means nothing.
Frequency against accuracy
These get conflated constantly and they are separate properties.
Accuracy is a position’s error radius. Open-sky GPS lands within roughly five to ten metres, enough for any geofence larger than a yard. Accuracy degrades in the places that matter most: multi-storey structures, dense urban delivery, covered docks and tunnels, where reflected signal can put a truck on the wrong side of a building.
Frequency is how often you hear. A fifteen-minute interval at motorway speed leaves a gap of twenty kilometres or more, which is fine for a predicted arrival and useless for confirming a geofence crossing at a specific gate. This is the reason arrival detection built purely on position sampling misses events that a one-minute feed would catch, and why the interval belongs in the carrier contract rather than in an assumption.
The rule worth applying: geofences must be sized against the reporting interval, not against the site. A two hundred metre geofence with a ten-minute feed will miss arrivals routinely, and every missed arrival becomes a manual status update somebody types later.
Gaps, jumps and frozen positions
Three failure shapes, each needing a different response.
A gap is silence. Normal during a rest period and abnormal mid-lane. The system has to distinguish them, which requires an ignition or motion flag rather than position alone. Without it, a legal nine-hour rest and a failed box look identical.
A jump is a position that cannot be reached from the previous one in the elapsed time. Usually a coordinate error, occasionally a timezone or UTC offset mistake in the provider’s mapping, which produces a truck that appears to move backwards by an hour of travel. Any sane pipeline rejects a position implying a speed above roughly 130 km/h rather than storing it.
A frozen position is the dangerous one, because it looks like data. The same coordinate arrives repeatedly with a fresh timestamp. It means either a genuinely parked vehicle or a box that has stopped updating its fix while still reporting. Only ignition state and odometer movement separate those two cases, and a feed that does not carry either cannot be trusted for dwell measurement at all.
Any position older than the expected interval should be displayed with its age, not as a current location. The single cheapest credibility fix available to most operations is putting “last seen 4h 12m ago” next to a truck icon.
Matching the vehicle to the load
This is the expensive failure in European road freight, and it is not a data quality problem in the usual sense: the positions are perfect, they belong to the wrong truck.
Subcontracting is normal here. The carrier who accepted the tender is often not the carrier who moved the load, and the plate on the transport order becomes wrong somewhere between acceptance and departure. Trailer swaps at a hub break the link again. The result is a load tracked against a vehicle that went somewhere else, which is worse than no tracking because it is confidently wrong.
Three practical controls. Require the plate to be confirmed at the point of loading rather than at tender, since that is when it becomes true. Make the reported position consistency-checked against the planned route, so a vehicle heading the wrong way raises an exception rather than a map pin. And treat a subcontracting notification as a mandatory event in the transport order, not as a courtesy, because that is the moment the link has to be rebuilt.
What belongs in the carrier contract
Data quality is a commercial term, and it costs nothing to write down at tender time. Five clauses cover it:
- A minimum reporting interval, stated in minutes, while the vehicle is in motion on the contracted lane.
- A coverage target, expressed as the share of loads with a usable position feed from loading to unloading. Eighty-five to ninety-five per cent is achievable with box-equipped carriers; set it against what the carrier actually runs.
- A notification obligation when a unit’s hardware fails or a vehicle is substituted, within a defined window.
- Confirmation of the executing vehicle at loading, including after subcontracting.
- Review at the quarterly business review, using your measurements rather than the carrier’s, alongside the service numbers set out in the guide to road freight KPIs.
Measuring the feed
Five numbers, per carrier, monthly. They take an afternoon to build and they change the conversation from impressions to evidence.
- Position coverage. Share of loads with at least one position in every hour between departure and arrival. The headline number, and the one carriers dispute least once it is defined this way.
- Median reporting interval. Median rather than mean, because a handful of long gaps drags an average into meaninglessness.
- Stale share. Percentage of active loads whose most recent position is older than twice the contracted interval, sampled through the day.
- Plate match rate. Share of loads where the vehicle reporting position matches the vehicle on the transport order. The number that exposes undeclared subcontracting.
- Automatic event share. Share of arrival and departure events generated by geofence rather than keyed by a human. This is the one that determines whether detention and dwell figures are defensible, and it is the foundation under any predicted arrival time worth showing a customer.
Run those five for one quarter and the carrier base separates into three groups: carriers whose data is good, carriers whose data is fixable with a configuration change at their provider, and carriers who are not actually sending what they agreed to send. Each group needs a different conversation, and none of those conversations is possible from a feed count. The broader framing, and why coverage rather than technology is what determines whether a programme works, is set out in the guide to freight visibility in European road freight.
