Definition
Correlation between AI-generated coaching intervention frequency and subsequent driver voluntary turnover, controlling for pay, tenure, and driving record, measuring whether the coaching method retains or repels drivers.
The Measurement Gap
Fleet safety programs measure coaching volume and safety scores but almost never measure whether the coaching itself is causing drivers to quit. The blind metrics that hide this relationship include:
Voluntary Turnover Rate (overall): Fleet-wide turnover is tracked as an HR metric, disconnected from safety program data. Nobody asks whether drivers who left received disproportionate coaching: the two datasets live in different systems owned by different departments.
Coaching Completion Rate: Measures whether drivers completed their assigned coaching sessions but not whether those sessions made the driver more likely to stay or leave. A 95% coaching completion rate could coexist with coaching-driven attrition if drivers complete the session and then submit their resignation.
Safety Score Improvement Post-Coaching: Tracks whether scores improve after coaching, which confirms the coaching changes behavior scores, but not whether the driver resents the process enough to leave for a competitor who does not use camera-based coaching.
Decision Supported:
Is the AI safety coaching program improving fleet safety, or is it just filtering out drivers who refuse to be surveilled? Are experienced, safe drivers leaving because the coaching method feels adversarial? CACI answers whether coaching frequency is a hidden driver of voluntary turnover, and whether the fleet is inadvertently selecting for compliance tolerance rather than driving skill.
Formula
Plain language: For each driver, count AI coaching interventions per 10,000 miles over a trailing six-month window. Then check whether the driver voluntarily departed in the subsequent six months. Calculate the partial correlation between coaching frequency and departure, controlling for pay, tenure, and preventable accident history. A positive correlation means more coaching predicts more departures, a warning signal.
Notation:
CACI = corr(CoachingFrequency, DepartureIndicator) | controlling for Pay, Tenure, SafetyRecord
Variables:
| Variable | Description | Source |
|---|---|---|
| CoachingFrequency | Number of AI-generated coaching intervention events per 10,000 miles driven in the trailing 6-month measurement window | Safety scoring platform (NATIVE) + telematics/ELD for mileage (NATIVE) |
| DepartureIndicator | Binary variable: 1 if the driver voluntarily departed within the subsequent 6-month observation window, 0 if retained | HRIS / payroll system (NATIVE): requires reason code distinguishing voluntary from involuntary |
| Pay | Driver compensation level: used as a control variable to isolate coaching effect from pay-driven turnover | HRIS / payroll system (NATIVE) |
| Tenure | Time in role: used as a control variable since newer drivers have inherently higher baseline turnover | HRIS (NATIVE) |
| SafetyRecord | Number of preventable accidents in the measurement period: used as a control variable to distinguish coaching effect from underlying safety risk | Safety management platform (NATIVE) |
| Miles per driver | Total miles driven per driver in the measurement period: denominator for normalizing coaching frequency | Telematics / ELD (NATIVE) |
A positive CACI indicates that coaching frequency correlates with voluntary departure after controlling for the variables that normally explain turnover. A negative CACI or CACI near zero indicates coaching is not a meaningful turnover predictor.
Proposed formula, pending validation pilots.