Fuel losses are rarely one dramatic event
When a fleet manager suspects fuel theft, the picture in their head is usually a hose and a jerrycan at midnight. That happens, but it is not where most money goes. The larger losses are slow, ordinary and spread across dozens of vehicles: twenty litres here, a padded receipt there, a card used on a vehicle that was parked at the time. Nobody notices because no single event is large enough to notice.
That is exactly why telemetry works so well on this problem. A human reviewing fuel invoices cannot correlate every litre against every kilometre. A system that samples tank level every minute alongside position, ignition and odometer can, and it does it for every vehicle, every day, without getting tired.
How fuel actually disappears
Understanding the methods matters, because each leaves a different signature in the data.
- Siphoning from the tank while the vehicle is parked overnight or at a remote site. Produces a sharp level drop with the engine off.
- Short filling at the pump, where the attendant dispenses less than the receipt shows and splits the difference. The tank rises by less than the invoice claims.
- Fuel card misuse, filling a private vehicle or a container on the company card. The transaction exists but no matching tank rise appears on the vehicle at all.
- Sale of fuel in transit, common on long-haul routes, often disguised by an unscheduled stop at an unfamiliar location.
- Excessive idling and unauthorised trips, which are not theft in the legal sense but consume real fuel and are usually the largest single recoverable amount.
- Genuine leaks and injector faults, which look like theft until you notice the loss is continuous rather than sudden.
A good fleet tracking platform should be able to separate these categories automatically, because the operational response to each one is completely different.
Capacitive fuel level sensors versus CAN bus data
There are two practical ways to know how much fuel is in a tank, and they are not equivalent.
CAN bus and OBD data
Modern vehicles already report fuel level and, on many commercial models, cumulative fuel consumed. Reading it over CAN or FMS is attractive because there is nothing to install in the tank, no warranty concerns and no drilling. The catch is resolution. Factory senders are designed to move a dashboard needle, not to detect a twenty-litre change. Readings are often coarse, damped and slow to settle, and the value can jump when the vehicle is on a slope. CAN data is excellent for consumption trends, engine hours and driver behaviour, and it is frequently good enough to spot large or repeated losses. It is rarely precise enough to build a disciplinary case on.
Capacitive fuel level sensors
A capacitive probe is a rod installed into the tank that measures the dielectric change as fuel rises and falls around it. Properly installed and calibrated, it gives resolution an order of magnitude better than a factory sender, updates continuously, and is far less affected by damping. This is the standard for heavy trucks, generators, buses, tankers and construction plant, and it is what you install when the objective is evidence rather than a rough estimate.
Two things determine whether a probe earns its cost. The first is calibration: tanks are irregular shapes, so the sensor must be mapped by adding known volumes and recording the reading at each step, producing a lookup table rather than a straight line. Skip this and every number afterwards is wrong. The second is installation quality: probe length cut correctly, sealed properly, wired to a clean supply and, ideally, tamper-protected, because a sensor that can be unplugged without raising an alert is a sensor that will be unplugged.
Where budget allows, use both sources. CAN gives you consumption, the probe gives you level, and disagreement between them is itself a useful signal. Combining sensor streams like this is routine IoT integration work rather than anything exotic.
What a drain event looks like in the data
Fuel level graphs have a characteristic shape. During normal driving the line slopes down gently and steadily. A refuelling event is a steep rise over a few minutes. Sloshing while the vehicle moves produces noise around the trend, which is why good software applies filtering and ignores readings taken during heavy acceleration or on gradients.
A drain looks different from all of these. You see a sudden fall, typically a meaningful volume over five to twenty minutes, while the engine is off and the vehicle is stationary. Because consumption cannot explain a drop with the ignition off, that pattern alone is close to conclusive once sensor error is ruled out. The subtler variant happens with the engine running at a stop, where the drop is much faster than the idle burn rate would produce. Comparing the observed rate of loss against the known idle consumption of that engine separates the two cases cleanly.
Context turns a data point into a finding. Overlay the drop with location history and you know precisely where the vehicle was. Overlay ignition, door and, where fitted, camera events and you know what else happened at that moment. Overlay the driver assignment and you know who was responsible for the vehicle. That is the difference between "we think fuel is going missing" and a timestamped, located, attributable event.
Geofences and refill correlation
The second half of the picture is refuelling. Draw geofences around your depot, your own tanks and the stations your drivers are authorised to use. Now every rise in tank level can be tested against three questions: did it happen inside an approved location, does the volume match the invoiced litres, and does the timestamp match the transaction? Answering those questions reliably depends on keeping minute-level history for months rather than days, so plan your storage and retention accordingly.
This is where short filling and card misuse surface. An invoice for eighty litres against a tank rise of sixty-two is a discrepancy that repeats if it is systematic. A fuel card transaction with no corresponding rise on any vehicle is a straightforward exception. Importing fuel card statements and matching them automatically to sensor events is one of the highest-value integrations you can add, and it typically pays for itself faster than the sensors do. Where that data has to flow back into an ERP or accounting system, the same connector work applies.
Geofences also catch the in-transit sale. A long stop at an unapproved location, combined with any level drop, is worth reviewing even when the volume is small, because that pattern usually repeats along the same route.
Setting alert thresholds without drowning in noise
The most common failure in fuel monitoring projects is not technical, it is alert fatigue. Set the sensitivity too high and the operations team receives forty notifications a day, stops reading them within a fortnight, and the project quietly dies.
Tune thresholds per vehicle class rather than globally. A drop that is significant on a small van is noise on a tanker with a nine-hundred-litre capacity, so express thresholds as a volume appropriate to the tank, not a single fleet-wide percentage. Require the drop to persist across several consecutive readings before an alert fires, so a single bad sample cannot trigger it. Suppress alerts during known refuelling windows and inside approved geofences. Add separate, lower-priority alerts for sensor disconnection and for readings that stop changing entirely, since both are classic tampering indicators.
Start conservative. Run the system in observation mode for two to four weeks, review what it would have flagged, and tighten from there. You are aiming for a handful of high-quality alerts a week that someone genuinely investigates, not a stream nobody reads. If you are unsure where to set the initial figures for your vehicle types, ask us for a starting configuration rather than guessing.
Working out the return
The business case is simple to construct and you should build it with your own numbers rather than a vendor's claims. Take your annual fuel spend for the vehicles you intend to fit. Estimate a conservative recovery percentage from the combination of theft prevention, idle reduction and invoice accuracy, and be deliberately modest, because a credible small number survives scrutiny better than an optimistic large one. Against that, set the cost of the probe, its installation and calibration, plus the incremental subscription for the fuel module.
On heavy vehicles with high consumption, the payback period is usually measured in months rather than years, and much of the benefit arrives before you catch anybody. Losses tend to fall as soon as drivers know the tank is monitored, which is a good reason to announce the rollout openly rather than run it covertly. Deterrence is cheaper than investigation, and it avoids the employee-relations problems that come with surveillance nobody was told about.
Reporting closes the loop
Alerts catch incidents. Reports change behaviour. A weekly fuel report per vehicle and per driver showing litres consumed, kilometres travelled, litres per hundred kilometres, idle hours and any flagged events makes patterns unmissable, and it lets you rank vehicles by efficiency so attention goes where it is worth spending.
Feed the same data into your maintenance and finance systems and it stops being a fuel project and starts being an operations one. Rising consumption on a specific vehicle points at injectors, tyre pressure or a driver who needs coaching, long before it becomes a breakdown. For fleets with compliance or audit obligations, retaining that history matters, which is why data retention and backup should be part of the conversation from the start and why we discuss hosting and retention policy with every fuel-monitoring deployment.
None of this requires exotic technology. It requires the right sensor, honest calibration, sensible thresholds and someone who actually looks at the report each week. Fleets that do those four things consistently see their fuel line stop drifting upwards, and that is the whole point.
Want to see fuel monitoring running on your own vehicles before you commit? We fit and calibrate probes, integrate CAN data, and set up the alerts and reports on our tracking platform or your existing system. Ask for a pilot on five vehicles or message us on WhatsApp — we will review the first month of data with you.