Definition
Estimated percentage of telematics-scored policyholders whose driving behavior patterns show statistical signatures of score optimization (avoiding measured risk behaviors) rather than genuine risk reduction.
The Measurement Gap
Telematics programs assume that measured driving behavior reflects actual risk. But when policyholders learn which behaviors are scored, some optimize the measurement rather than genuinely reducing risk. They avoid driving during monitored periods, exhibit careful behavior near measurement windows while reverting to normal habits otherwise, or improve on scored dimensions while worsening on unscored ones. The telematics score improves, the discount increases, but the actual risk may not change. This is a Goodhart's Law failure: when the measure becomes the target, it ceases to be a good measure. Legacy metrics cannot detect this. Loss ratio captures aggregate claim outcomes but cannot distinguish genuine risk reduction from strategic score optimization. Claims frequency measures how often losses occur without connecting frequency changes to behavioral authenticity. Telematics adoption rate measures program participation, not measurement integrity. Discount penetration tracks financial impact of the program without assessing whether discounts are earned. Customer satisfaction with telematics programs measures experience, not whether the behavioral data is trustworthy.
Decision Supported:
Is the telematics program measuring real risk reduction or learned gaming behavior, and should the scoring algorithm be modified to detect and discount strategic behavior patterns?
Formula
Plain language: Identify behavioral signatures consistent with gaming: abrupt driving pattern changes immediately after telematics enrollment, avoidance of driving during monitored periods, and driving behavior that improves on scored dimensions while worsening on unscored dimensions. Score each policyholder on a gaming probability scale.
Notation:
BGDR = Policyholders_with_gaming_score_above_threshold / Total_telematics_policyholders x 100
Gaming Score per policyholder: GS_i = w_1 x enrollment_discontinuity_i + w_2 x period_avoidance_i + w_3 x dimension_divergence_i
Variables:
| Variable | Definition | Unit |
|---|---|---|
| enrollment_discontinuity_i | Magnitude of driving behavior change within 30 days of telematics activation vs. prior 90 days | Z-score (dimensionless) |
| period_avoidance_i | Ratio of unmonitored trips or trip-start-avoidance to total expected trips | Dimensionless |
| dimension_divergence_i | Correlation between scored-dimension improvement and unscored-dimension worsening | Dimensionless (-1 to 1) |
| w_1, w_2, w_3 | Weights (initially equal at 0.33 each, tuned during pilot) | Dimensionless |
| Threshold | Gaming score above 0.6 (proposed, validate during pilot) | Dimensionless |
Proposed formula, pending validation pilots.