Definition
Mean absolute percentage error between AI-predicted state of charge at route completion and actual state of charge upon arrival, measuring whether the fleet's EV route assignment model is accurate enough to prevent service failures.
The Measurement Gap
Fleets entering EV adoption track metrics that confirm commitment without validating execution.
EV utilization rate
shows how often electric vehicles are dispatched without revealing whether assignments matched actual range capability.
Total fleet emissions reduction
aggregates the environmental benefit without accounting for diesel rescue dispatches that partially offset it.
Charging infrastructure uptime
measures whether chargers work at the depot without tracking whether en-route charging stations were available when planned routes depended on them.
These metrics create a false confidence loop. Leadership sees high EV utilization and celebrates adoption progress while operations quietly manages range failures, emergency reroutes, and late deliveries caused by EVs assigned to routes they could not complete. The prediction model that assigns EVs to routes is never held accountable for its accuracy because no metric feeds back the actual outcome: how much charge remained at arrival versus how much was predicted.
EVARG closes this loop by treating EV assignment as a prediction problem with measurable accuracy, supplemented by failure-rate metrics that capture real-world consequences when predictions are wrong.
Decision Supported:
Is the fleet assigning EVs to routes they can actually complete, or is the assignment model creating service failures that undermine both operational reliability and the business case for electrification? Should the fleet tighten assignment margins, invest in better prediction models, or restrict EV assignments to proven route profiles?
Formula
Plain language: For every EV route assignment, EVARG compares what the model predicted the battery state of charge would be at arrival versus what it actually was. The average size of this prediction error (as a percentage of the predicted value) is the core metric. Two supplemental metrics track consequences: how often EVs arrive critically low or need emergency charging, and how often planned charging stops were unavailable or slower than expected.
Notation:
EVARG = (1/N) × Σ |PredictedSOC_arrival_i − ActualSOC_arrival_i| / PredictedSOC_arrival_i × 100
Supplemental metrics:
RangeFailureRate = (Trips where ActualSOC < 10% at arrival OR emergency charge required) / Total EV trips × 100
ChargingStationReliabilityGap = (Trips where planned en-route charging was unavailable or significantly slower than expected) / Total EV trips with planned en-route charging × 100
Variables:
| Variable | Description | Source |
|---|---|---|
| PredictedSOC_arrival | AI-predicted battery state of charge at route destination (%) | Fleet EV management / TMS |
| ActualSOC_arrival | Actual battery state of charge at route completion (%) | Vehicle telematics (CAN bus) |
| Emergency charge event | Unplanned charging stop required to complete route | Route logs / telematics |
| Planned charging availability | Whether en-route charging station was available and functional at planned time | Charging network API + route logs |
| Charging speed differential | Actual vs. expected charging rate at planned stops (kW) | Charging session records |
| N | Total completed EV route assignments in measurement period | TMS / dispatch records |
EVARG is expressed as MAPE, a 20% EVARG means the model's predictions are off by 20 percentage points of SOC on average. A model predicting 30% SOC at arrival that consistently delivers 10% SOC has a dangerous positive bias that EVARG will surface.
Proposed formula, pending validation pilots.