AllConnects is a Flemish specialist in fleet and asset management. With FLEET.connect, CHECK.connect, and LIVE.connect, companies in the construction, transportation, engineering, and rental industries collect reliable data from their vehicles, machinery, and teams—the foundation on which any form of analysis or AI must be based.
What the research says
A European study on the state of fleet management this year revealed a striking figure: 73 percent of the fleet managers surveyed are currently exploring whether AI solutions can be implemented in their own organizations. Only 23 percent rule that out in the short term.
At the same time, the researchers note that effective implementation is still in its early stages at most organizations. So there is a lot of interest, but few concrete investments. That is not a contradiction—it is a symptom.
Where do administrators see the most opportunities? The four areas of application that are mentioned most often are:
| Scope | Mentioned by |
|---|---|
| Predictive maintenance | 38% |
| Automatic Detection of Anomalies | 24% |
| Predictive dashboards | 19% |
| Analysis and Optimization of Consumption | 14% |
What stands out is that all four revolve around the same thing. Not smarter software, but recognizing patterns in vehicle data. And that's where the problem begins.
Why It Stops at Just Thinking About It
The explanation you usually hear for the gap between interest and implementation is that the technology isn't mature enough yet. That's not true. The models exist, they work, and they've become affordable.
The real problem: most fleets don't have their data infrastructure in order.
Take predictive maintenance, which, at 38 percent, is by far the leading contender. To predict that a vehicle will break down within three weeks, you need:
- Error codes generated by the vehicle itself—and that are received somewhere.
- Actual engine hours and fuel consumption, not estimated based on kilometers traveled.
- Maintenance history linked to the vehicle, not to an Excel file kept by a single colleague.
- There's enough history to identify a pattern.
If even one of those four is missing, no model can make a meaningful prediction. You do get a result—that’s true. A model always gives an answer. But an answer based on estimates is just an estimate with a graph drawn around it.
The Risk of a Nice Dashboard
An AI dashboard that runs on unreliable input is more dangerous than no dashboard at all. Without a dashboard, you know you don’t know. With a dashboard, you think you do, and you adjust your maintenance schedule based on figures that aren’t backed by anything.
AI isn't a switch you can just flip on. It's an additional layer built on top of your data. If the foundation is crooked, you'll just keep building crookedly at a faster pace.
Calculated or Read: The Difference That Makes All the Difference
There is one technical distinction that determines whether your fleet is ready for this step, and it is rarely explained: the difference between calculated and read data.
The calculated data is based on an estimate. A tracker records that a delivery van has traveled 180 kilometers and multiplies that by the average fuel consumption. The result is a fuel efficiency figure. This provides a general overview, as it averages out across an entire fleet.
The data is retrieved directly from the vehicle via the CAN bus—the internal network through which the on-board computer, the engine control unit, and the sensors communicate with each other. This allows you to view the actual fuel consumption, the actual engine hours, the odometer reading, the fuel level, and the error codes.
For a monthly report, that difference doesn't matter much. For a prediction, it makes all the difference. A model that looks for the moment when an engine's fuel consumption begins to deviate cannot possibly identify that from a figure calculated based on kilometers.
Would you like to discuss this in more detail?
We wrote a separate article about just how accurate CAN bus data actually is, what you can and cannot glean from it, and why there is such a big difference compared to calculated data. Read about the accuracy of CAN bus data →
The four applications, adapted to a Belgian vehicle fleet
The study focused on large and medium-sized fleets. Our customers are different: a construction company with eighteen delivery vans and four cranes, a bus company with forty vehicles, and a technical service provider with thirty service vehicles. What do those four applications mean in practice for them?
Predictive maintenance 38%
Not: “The garage will call you sooner.” Instead: You can tell from the fuel consumption pattern and the error codes that something is off, so you schedule the repair for a week when that vehicle doesn’t need to be in the shop anyway. For a fleet of twenty vehicles, that means avoiding one or two downtimes per year. That downtime rarely costs you just an invoice—it costs you a day’s worth of crew time.
Identifying anomalies: 24%
A machine that suddenly idles twice as long. A delivery van that takes a detour every Thursday. A cooling unit that keeps running at night while it’s on the job site. You don’t see things like this in a monthly report—you only notice them when someone points them out. That’s exactly what automatic detection does.
Predictive dashboards 19%
When do you need an additional vehicle? Which vehicle is nearing the point where replacement is cheaper than repair? Where is your cost per kilometer starting to get out of hand? This is the use case that most closely resembles a budget discussion—and it’s also the one that requires the most historical data.
Optimize consumption by 14%
This is the area with the lowest score, which is surprising. This is probably because many companies here already use driving behavior scores and therefore feel they’ve already covered this ground. Still, there’s room for improvement here, especially when it comes to idling machinery.
Why a small fleet has relatively more to gain
The instinctive reaction is: AI is for large fleets, because they have the volume and the budget. That’s partly true—more vehicles mean more data and, therefore, faster pattern recognition.
But the impact per incident goes exactly the other way.
If one delivery van breaks down in a fleet of a thousand vehicles, it’s a scheduling issue. You just bring in a replacement, and the customer doesn’t even notice. If one of five crane trucks at a construction company breaks down on Monday morning, the entire job site comes to a standstill—along with the crew on site—and that results in either a fine or a three-week delay in the schedule.
Small fleets have no buffer. As a result, every instance of downtime avoided is worth significantly more. The business case for an SME fleet isn’t any smaller—it’s more compelling.
Don't forget the driver
There is another figure in the study that has received too little attention. When asked what benefits drivers themselves experience from telematics, 46 percent cited greater ease of use, and 23 percent pointed to features that provide operational support.
That's not a minor detail. That's the condition.
A system that doesn’t benefit the driver or worker will be circumvented. Trips aren’t closed out properly, statuses are incorrect, and entries are forgotten. And there goes your data quality—precisely the foundation we were discussing above.
A system that handles trip tracking, that streamlines check-in at the job site into a single step, and that automatically sends the correct hours to payroll—that’s the kind of system people actually use. Support from the people on the job site isn’t just a nice-to-have. It’s the quality control for your own data.
Curious to know what your vehicles are actually emitting today?
In a half-hour conversation, we’ll go over what data you’re currently collecting, which data is calculated and which is extracted—and what that means in terms of what you can do with it. No obligation, and with a concrete answer.
Get adviceWhat do the Belgian figures show?
The survey above focused on Italian fleets. We are currently surveying our own members and will publish the results here this fall for Belgian fleet managers, fleet administrators, and business owners in the construction, transportation, and technical services sectors.
Is your fleet data ready for AI?
Before you get started with AI, there’s a more important question: What are you really getting out of your vehicles today? Go through these six points.
Six “yes” answers? Then it makes sense to discuss predictive applications. Four “yes” answers or fewer? Then your profit this year won’t come from AI, but from closing those gaps. And that’s good news, because it’s much quicker to implement—and it delivers immediate results, even without AI.
Frequently Asked Questions
Do I need AI to manage my fleet more efficiently?
No. Most companies derive their greatest value from data they can already collect today: actual fuel consumption, idle hours, trip logs, and maintenance tracking. AI adds a predictive layer to that, but that layer is only useful if the underlying data is accurate and complete.
What exactly is predictive maintenance?
With predictive maintenance, you continuously analyze vehicle data to identify technical issues before a breakdown occurs. Instead of repairing a vehicle after a breakdown or scheduling maintenance at fixed intervals, you plan based on the vehicle’s actual condition. This can reduce repair costs and prevent unexpected downtime.
What is the difference between calculated and read vehicle data?
Calculated data is an estimate based on other data—for example, fuel consumption derived from the number of kilometers traveled. Readout data comes directly from the vehicle via the CAN bus, the internal network through which the onboard computer and sensors communicate. Calculated data is often sufficient for general reporting; readout data is necessary for predictive applications.
Is this only relevant for large fleets?
No. Large fleets have more data and therefore identify usable patterns more quickly, but the impact of a single unexpected outage is much greater for a small fleet because there is no reserve capacity. For a construction company with five crane trucks, a single breakdown immediately brings a job site to a standstill. Relatively speaking, the gains are therefore often greater for smaller fleets.
How much data do I need to make reliable predictions?
As a rule of thumb, at least twelve consecutive months of data per vehicle, so that seasonal factors such as winter consumption or construction holidays are taken into account. Quality is more important than quantity: data collected over a single year is more useful than calculated data covering three years.
Can I process my employees' vehicle information without any restrictions?
You may process data necessary for your business operations, but you must have a legal basis, clearly inform your employees in advance, and not collect more data than is necessary. For technical vehicle data—such as fuel consumption, engine hours, and error codes—this is generally simpler than for data that tracks the behavior of an individual driver. Involve your prevention advisor and the consultative bodies in a timely manner.
Continue reading
Steve Van Avermaet COO at AllConnects. As COO at AllConnects, Steve translates technology and data into practical solutions for fleet and asset management. He writes about efficiency, innovation, and the practical aspects of a smart, connected fleet of vehicles and equipment.
Source: Fleet Management Watch, conducted by the Connected Vehicle & Mobility Observatory at the School of Management of the Politecnico di Milano, in collaboration with Targa Telematics (July 2026). The study is based on data from participants in two Italian observatories for large and medium-sized fleets, which together account for nearly 60,000 vehicles.
