Definition
The realized dollar value per AI-issued maintenance alert, calculated as the failure cost prevented minus every cost the alert triggered (intervention, production loss, and false-alarm investigation), divided by total alerts issued.
The Measurement Gap
AI-generated alerts create a new economic decision (whether acting on each alert produces net positive value) that data-science metrics (precision, recall, F1, AUC) cannot answer because they treat every correct prediction as equally valuable and ignore intervention cost entirely. A true positive that catches a $200K catastrophic failure scores identically to one that flags a $500 seal replacement, and a false alarm that burns eight hours of investigation registers as zero cost. Legacy operational metrics inherit the same blindness: OEE measures output against capacity without asking whether the maintenance protecting that output is economically justified, PM Compliance reports schedule adherence rather than work value, and MTBF tracks failures rather than alert economics. The gap is structural: legacy metrics were designed before AI-generated alerts existed.
Decision Supported:
Is the predictive model generating positive economic value per alert, or are intervention costs and false-alarm overhead destroying value the model technically detects?
Formula
Plain language: For every alert the model issues in a period, take the failure cost that was avoided, subtract the cost of the intervention triggered, the production loss during planned work, and the cost of false alarms generated in the same window. Divide the total by the number of alerts to get net economic yield per alert.
Notation:
ANEY = Σ (C_avoided − C_intervention − C_production_loss − C_false_alarm) / N_alerts
Variables:
| Variable | Definition | Unit |
|---|---|---|
| C_avoided | Estimated cost of the failure had no action been taken (downtime + emergency repair + secondary damage + safety/environmental penalties) | $ |
| C_intervention | Actual cost of the planned work order triggered by the alert (labor + materials + services + tools) | $ |
| C_production_loss | Production value lost during the planned intervention window | $ |
| C_false_alarm | Fully loaded cost of investigating alerts that turned out to be false positives | $ |
| N_alerts | Total alerts issued by the model in the evaluation period | count |
Proposed formula, pending validation pilots.