Vehicles are almost all connected but fleet managers still lack truly usable and reliable data. A problem that physical AI is responsible for solving
Today, 85% of vehicle fleets in Europe rely on fuel card data and spreadsheets to track their vehicles’ mileage, with manual entry prone to errors. At the same time, 15% of fleets use computerized fleet monitoring systems, otherwise called telematics, which are no longer adapted to current challenges. For good reason: while telematics was born historically around a need for geolocation, thanks to boxes installed on vehicles, which rely on telecommunications technologies to collect, transmit and analyze data remotely, the reporting of this can now be done without additional on-board equipment. In particular thanks to natively connected vehicles, driven by the “software defined vehicle” SDV trend in the automotive industry.
New variables inevitable as the volume and diversity of real-world data increases
Without the adoption of practices that optimize data, the result is a lack of clear information on private use, road behavior or energy consumption, as well as discrepancies between obsolete means and modern engines. Like electric vehicles, which are experiencing increasing growth in fleets. Consequently, this lack of visibility and reliability of vehicle fleet data often leads to economic losses resulting from erroneous decisions for fleet managers. Who nevertheless have to juggle many constraints…
Among them is an increase in the total cost of vehicle ownership (TCO) of 30% since 2020, but also volatility in fuel and electricity prices, inflation impacting all services as well as tax compliance obligations driven by the weight of the Mobility Orientation Law in France and other European regulations, which can cause annual penalties of 4,000 euros per electric vehicle. So many points of friction which suggest that the loss of budgetary control is inexorable.
A technological turning point to be negotiated without delay
In this unprecedented context, where connected vehicles are becoming the norm since by 2030 nearly 9 out of 10 new vehicles will be delivered connected from the factory (i.e. without additional hardware installation), a paradox persists: the abundance of data does not guarantee their reliability. Data from vehicles are in fact, and above all, raw sensor signals. As in any system operating in real conditions, these signals inevitably contain anomalies, inaccuracies or aberrant values linked to multiple factors (cold starts, environmental variations, vibrations, temporary loss of connectivity, electronic interference or even calibration differences between vehicles for example). Added to this is the heterogeneity of formats and the fragmentation of data sources depending on the manufacturer.
So, having raw data is not enough. Their interpretation requires advanced quality control, contextualization and reconstruction mechanisms in order to transform these signals into reliable, coherent and actionable information. In this new environment, the ability to provide verified and ready-to-use data becomes a strategic issue for companies. Particularly when the teams using data and fleet tools diversify, now ranging from the finance department to HR, including sales, the safety division and operational departments.
Physical AI, the panacea for optimizing the uses of electric vehicles
But how to go about it? Thanks to physics-based AI. Combining the principles of physics and artificial intelligence techniques to develop models capable of predicting even the most sophisticated systems with increased precision, this approach integrates the constraints of the physical world instead of existing only in software or digital environments. In fact, physical AI makes it possible to transform raw data from connected vehicle sensors and exploit these reliable metrics by the combination of relevant LLM analyzes as well as by scoring and prediction models adding context and depth to the reports carried out in the same environment.
Capable of transforming signals from a vehicle’s sensors into reliable information and then timely recommendations for fleets, physical AI also has a significant role to play in the adoption of electromobility. Because the paradigm shift induced by electrification leads to an evolution in operational conduct such as the administration of electrical infrastructure, the production of more in-depth reporting or even a consolidation of the management of data reported from vehicles in the fleet.
By reporting vehicle data, independently of the hardware (i.e. without collecting data from charging stations or expensive cables), physical AI therefore provides a precise vision of home charging costs (session history, time, kwh price) allowing reactive and proactive budget administration. By providing alerts, targeted recommendations and user training, it also encourages a reduction in expensive charging behavior, such as excessive rapid charging or regular total discharges. This detailed and cyber-secure analysis then formulates the promise of saving user companies up to 1,500 euros per electric vehicle per year (calculation based on a vehicle representative of the market traveling 30,000 km per year and consuming 6,000 kWh per year in charging).
