Fleet managers transitioning from diesel to electric delivery vans face a calculation problem they did not have before: estimating how much battery charge a specific vehicle will consume on a specific route today. With diesel, the consequences of a bad fuel economy estimate are a stop at a gas station. With electric, the consequences of a bad energy estimate are a vehicle stranded mid-route or, if the driver turns around early, undelivered stops and a disrupted schedule.
The most common approach we see in early-stage EV fleets is to use the vehicle manufacturer's rated range or a fleet-wide average wh/mile figure to determine which routes are safe to run. That approach carries more risk than it appears to.
Why averages fail at the vehicle level
Take a fleet of 15 cargo vans of the same make and model. The manufacturer rates them at 200 miles of range under standard load at 70 degrees. In practice, across those 15 vehicles on a given January day, real-world range will vary by 30 to 50 miles between the best-performing and worst-performing units, before you account for route differences. Individual battery health, tire inflation, HVAC draw, and driver behavior each introduce variance that compounds.
Now add route-level variables. A route with heavy stop-and-go density in an urban core consumes meaningfully more energy per mile than a suburban route with long stretches between stops. Payload weight matters too: a van loaded to 90 percent of capacity uses noticeably more kWh per mile than one running at 30 percent capacity, particularly in hilly terrain. Cold weather amplifies all of this because battery efficiency drops and cabin heating pulls from the same pack.
If your planning assumption is "this vehicle model gets 200 miles of range," you are ignoring every one of these factors. You are treating a probability distribution as a single point. The mean of that distribution might be 200 miles. The 5th percentile on a cold day with a heavy payload on a stop-dense route might be 155 miles. Scheduling a 170-mile route for that vehicle on that day puts you very close to the edge.
What per-vehicle, per-route prediction actually requires
Accurate energy prediction at the route level requires combining at least four categories of data: vehicle-specific historical consumption, route characteristics (distance, stops, elevation profile), current conditions (temperature, payload estimate), and real-time telemetry as the route progresses.
Vehicle-specific history matters because individual units diverge in efficiency over time. A van with 40,000 miles on it has measurably different consumption characteristics than a van of the same model with 8,000 miles. Lumping them together into a fleet average masks that variance.
Route characteristics need to be derived from actual GPS trace data rather than straight-line distance or even mapped road distance. A route that covers 85 road miles but includes 120 stops in a dense grid pattern will consume significantly more energy than a route that covers 85 road miles with 30 stops on arterial roads. Stop density creates more acceleration cycles, more idle HVAC draw, and more time in low-efficiency creep-speed driving.
Temperature adjustments should be applied at the vehicle and battery level, not at a generic fleet level. Battery thermal management systems vary by vehicle generation and configuration, and a cold-soak effect from a vehicle sitting outside overnight at 18 degrees Fahrenheit affects a newer liquid-cooled pack less than an older air-cooled one.
The second-order problem: what bad estimates cost in charging decisions
Inaccurate energy prediction does not only risk mid-route breakdowns. It also makes charging decisions unnecessarily conservative in ways that cost money.
When fleet managers do not trust their range estimates, the natural response is to charge every vehicle to 100 percent SOC every night regardless of what tomorrow's routes actually require. That is operationally safe but expensive. Charging all vehicles to full when half of them only need 60 to 70 percent for tomorrow's routes means more simultaneous charging demand, higher peak power draw, and a harder demand charge problem on the electricity bill. It also compresses the scheduling flexibility that makes cost optimization possible.
If you know with reasonable confidence that Van 7 is running a 62-mile suburban route tomorrow with a light payload forecast, and its current SOC is 55 percent, it needs maybe 25 kWh of charge rather than a full top-up to 100. That vehicle can wait several hours before it needs to start charging. That scheduling flexibility is only available if you trust the prediction.
How Fleetvolts approaches this
We maintain a per-vehicle energy model that updates with each completed route. The model carries forward historical consumption rate distributions segmented by route type (stop density, distance class), temperature range, and estimated payload band. Each day, when a dispatch plan comes in, the model computes an expected energy draw for each vehicle-route combination along with a conservative bound that we use for charge planning.
The goal is not a single point estimate. It is a distribution tight enough to make confident scheduling decisions. If the 10th-percentile consumption for a particular vehicle on a particular route class puts the vehicle back at 18 percent SOC and the next dispatch is not until 7:00 AM the following day, we can confidently delay charging for that vehicle by several hours. If the same vehicle is doing an unusual long-distance route with payload data suggesting a heavy load and temperatures are forecast to drop below freezing overnight, we prioritize it higher in the charging queue regardless of its current SOC.
We are not claiming this is a solved problem. Route conditions change, dispatch plans shift, drivers take detours, and payloads are sometimes different from what was logged. The system is continuously recalibrating as it receives real-world outcomes. The accuracy goal is not perfection; it is good enough to make the 5:30 PM charging queue decisions reliably, without requiring a dispatcher to manually override the schedule every day.
The planning shift that matters
Moving from fleet-average range estimates to per-vehicle, per-route energy predictions is not a purely technical change. It also changes how fleet managers think about their role. When each vehicle has an individually modeled energy budget, you stop managing a homogeneous pool of "electric vans" and start managing individual assets with known characteristics. That is a more accurate representation of what you actually have, and it is what makes scheduling optimization at scale both possible and credible to the people doing the work.