Definition
Net economic value of AI-driven storm/outage predictions after accounting for correct pre-positioning savings, incorrect pre-positioning waste, and delayed restoration costs from misallocated crews.
The Measurement Gap
Traditional storm performance tracking measures outcomes: SAIDI by storm category, restoration hours by crew district, mutual aid utilization as a headcount. These are retrospective measures. They tell you what happened. They do not tell you whether the AI prediction that drove the operational response was worth the cost of acting on it.
The specific failure point is pre-positioning economics. Before a storm, the AI model produces a damage forecast: predicted customer outages by geographic zone, predicted crew requirements by district, predicted mutual aid need. The operations team mobilizes crews based on that forecast. If the storm tracks differently than predicted, crews are in the wrong district. Customers in the actual impact zone wait for crews to reposition. Mutual aid crews sitting idle in an unaffected district incur lodging, standby, and redeployment costs. None of this appears in any metric as "prediction error cost." It appears as "storm cost", an undifferentiated line in the emergency response budget.
OPEY enables a specific operational decision: did the AI prediction system earn its keep this storm season, and should the utility adjust its reliance on the model's spatial predictions for crew pre-positioning? That decision requires connecting prediction quality to economic consequence at the event level, not in aggregate annual statistics.
What goes wrong without this measurement: utilities continue investing in prediction models that produce accurate-looking MAPE statistics while generating negative economic value from the operational decisions they drive. The model looks good on paper. The storm budget runs over every season.
AI and ML make this measurement possible by logging prediction outputs against actual outcomes at the zone level, enabling the cost attribution that manual storm tracking cannot perform.
Formula
OPEY = Σ (Value_correct_positioning − Cost_incorrect_positioning − Cost_mutual_aid_excess) / N_events
Variables:
- Value_correct_positioning: estimated restoration cost savings from crews pre-positioned in actual impact areas (reduced travel time × cost per crew-hour × incremental customers restored faster ($/event)) Source: OMS restoration logs, crew dispatch records (WFM/CMMS)
- Cost_incorrect_positioning: mobilization, travel, standby, and redeployment costs for crews pre-positioned in non-impact areas ($/event). Source: crew expense accounting, WFM
- Cost_mutual_aid_excess: per-crew-day rate × excess crews mobilized but not utilized × days ($/event). Source: mutual aid invoices, OMS crew deployment records
- N_events: count of storm or major outage events where AI predictions drove pre-positioning decisions. Source: OMS event log
Calculation Frequency: Per event, aggregated to seasonal and annual
Output: Dollars per storm event. Positive values indicate the prediction system created net economic value. Negative values indicate the operational decisions based on prediction cost more than they saved. Track OPEY by storm type (hurricane, ice storm, severe convective) to identify where prediction accuracy varies by phenomenon type.
Proposed formula, pending validation pilots.
At a Glance
Decision Supported
Did the AI storm prediction make the utility money or lose it, not "was the prediction accurate" in statistical terms, but "were the operational decisions based on the prediction economically justified?"