Definition
Percentage of AI-scored hard braking events that, upon review, were actually defensive driving responses to external hazards rather than indicators of aggressive or inattentive driving.
The Measurement Gap
Fleet safety programs universally track hard braking events as negative indicators, but almost none distinguish between braking caused by poor driving and braking caused by excellent hazard response. The blind metrics that mask this distinction include:
Hard Braking Event Count: Treats every deceleration exceeding a g-force threshold as equally bad. A driver who brakes hard to avoid hitting a child in a crosswalk receives the same negative score as a driver who brakes hard because they were tailgating and not paying attention.
Driver Safety Score (composite): Aggregated safety scores that incorporate hard braking events without classification inherit the same false positive contamination. The composite score looks authoritative but contains systematic measurement error.
Coaching Intervention Volume: Organizations that coach based on raw event counts will direct coaching resources toward drivers who may already be the fleet's best defensive operators.
Decision Supported:
Is the safety scoring system penalizing good drivers for braking correctly? Are coaching resources being wasted on drivers who do not need coaching. While genuinely risky drivers receive less attention because the coaching queue is filled with false positives? DBFPR answers whether the AI safety scorer is measuring what the organization thinks it is measuring, and whether the coaching program built on that data is targeting the right drivers.
Formula
Plain language: Review a sample of hard braking events flagged by the AI safety scorer. For each event, classify whether the braking was a defensive response to an external hazard or an indicator of poor driving. Divide the defensive events by the total reviewed events to get the false positive rate. Separately, calculate how many coaching interventions were triggered by events that turned out to be defensive, that is the coaching waste rate.
Notation:
DBFPR = (DefensiveEvents / ReviewedEvents) × 100
Coaching Waste Rate = (CoachingFromDefensive / TotalCoachingInterventions) × 100
Variables:
| Variable | Description | Source |
|---|---|---|
| DefensiveEvents | Hard braking events confirmed upon review as responses to external hazards: vehicle cut-offs, pedestrian incursions, road debris, animal crossings, red-light runners | Safety manager review (MANUAL initially) / AI camera classification (DERIVED if trained) |
| ReviewedEvents | Total hard braking events reviewed in the measurement period, excluding events classified as Ambiguous due to insufficient evidence (e.g., no camera footage available) | Telematics platform (NATIVE) + review classification |
| CoachingFromDefensive | Coaching interventions that were triggered by events subsequently classified as defensive | Safety management platform (NATIVE) + review classification |
| TotalCoachingInterventions | All coaching interventions triggered by hard braking events in the measurement period | Safety management platform (NATIVE) |
| Hard braking events | Deceleration events exceeding the fleet's g-force threshold, with timestamp, location, and speed data | Telematics platform (NATIVE) |
| Camera footage | Video evidence from forward-facing and driver-facing cameras corresponding to each hard braking event | Dashcam / AI camera system (NATIVE) |
Proposed formula, pending validation pilots.