Restoration Routing Deviation Attribution (RRDA)

GMI-EU-035Energy & UtilitiesIntegration-Required

Definition

Measures what percentage of deviation between the AI restoration routing plan and actual restoration execution is attributable to model errors (wrong assumptions about access, travel time, crew capability) vs. unknowable conditions (road closures discovered on site, new damage, safety hazards).

The Measurement Gap

Post-storm performance analysis traditionally measures restoration hours, crew productivity (customers restored per crew-hour), and SAIDI contribution by storm category. These metrics do not evaluate whether the restoration routing plan itself contributed to extended restoration times. A crew that traveled 45 minutes to a location that turned out to be already restored (because the model had stale damage information) consumed 90 minutes of round-trip time that appeared in the productivity metrics as reduced throughput. The model's error is invisible.

The specific failure point is data integration gaps that the routing model could address but hasn't. Common model-attributable deviations include: the model assumed road access based on pre-storm GIS data when county road closure feeds were available and not integrated; the model assumed crew travel time based on highway speeds when local geography makes that assumption consistently wrong for certain districts; the model used damage estimates from the storm prediction AI that systematically underestimate damage in specific terrain types. Each of these is a correctable data or methodology problem. Field-attributable deviations (newly fallen trees discovered mid-restoration, hazardous materials at a damage site, power lines blocking a road the crew expected to use) are not correctable by model improvement.

RRDA enables the specific decision of which data integration investments would most reduce model-attributable routing deviations. If RRDA analysis shows that 60% of model-attributable deviations involve road access assumptions, integrating a real-time DOT road closure feed is the highest-ROI improvement. If the majority involve damage estimate errors, the storm prediction model is the target.

Without this measurement, every post-storm report concludes with vague recommendations to "improve storm prediction" without specifying which component of the routing model failed and which data would fix it.

OMS restoration sequence logging and workforce management GPS data make the plan-vs-actual comparison precise enough to support deviation attribution.

Formula

RRDA = Model_attributable_deviations / Total_deviations

Variables:

  • Model_attributable_deviations: count of significant deviations from the AI restoration routing plan where the cause was a model assumption error (wrong road access assumption when better data was available, incorrect crew travel time based on model geography, wrong damage extent due to storm prediction model error, wrong equipment availability assumption based on stale CMMS data (count)) Source: crew debrief records, post-storm structured review
  • Total_deviations: all significant deviations from the AI restoration routing plan (defined as deviations resulting in 30+ minute delay or one or more routing changes with documented cause (count)) Source: OMS restoration sequence logs compared to AI routing plan
  • Field-attributable deviations (1 − RRDA): deviations caused by conditions genuinely unknowable at plan time. New damage during restoration, undiscovered hazards on arrival, road conditions not captured in any available data source

Calculation Frequency: Per major storm event; sample-based for smaller events (50+ deviation classifications per event)

Output:

Percentage, 0% to 100%. RRDA below 30% means most deviations are field-condition surprises the model could not have known: the model is doing what it can. RRDA above 50% means the model's assumptions are failing more than field conditions are surprising: specific model data improvements are warranted.

Proposed formula, pending validation pilots.

At a Glance

Category
Forecast Accountability, Storm Response, and Capital Planning

Decision Supported

Can the restoration routing AI be improved, or is the deviation inherent in storm uncertainty?

← Renewable Forecast Dispatch Cost (RFDC)Congestion Forecast Calibration Score (CFCS) →

Assess Your Organization

The AI-GPS Assessment evaluates your AI governance maturity and recommends KPIs like this one for your implementation roadmap.

Take the Free AI-GPS Assessment