Definition
The measured service-life outcome of AI repair-vs-replace recommendations: specifically whether repaired parts deliver the predicted remaining useful life, or fail earlier than a replacement would have at the moment the recommendation was made.
The Measurement Gap
The repair-vs-replace AI recommends repairing a gearbox rather than replacing it. The repair cost is $8,400 versus $34,000 for replacement, and the model predicts 18 months of remaining useful life post-repair. The gearbox is sent to a repair vendor, returned, reinstalled. Eleven months later it fails catastrophically, taking a coupling and a motor with it. Total cost: $8,400 repair plus $62,000 in consequential damage plus $180,000 in lost production. The failure is attributed to the gearbox, not to the repair decision. Nobody audits whether the AI's prediction of 18 months was reasonable given the part's condition at the time of the recommendation. Maintenance cost per unit measures total spend but does not separate repair-followed-by-failure from successful repair. Mean time between repairs measures the asset's reliability but does not attribute back to the recommendation that chose repair over replacement. The repair vendor's warranty covers the repair itself but not the production impact of a failed repair decision. The AI platform reports that 78% of repair recommendations were "accepted", a metric of compliance, not outcome. The fundamental question: did the repaired part actually last as long as the AI said it would?: goes unanswered across the entire fleet.
Decision Supported:
Whether to accept repair recommendations by criticality class, whether to retrain the repair-vs-replace model on outcome data, and when to override toward replacement despite repair-favorable economics.
Formula
Plain language: For every "repair" recommendation, track the repaired part's actual service life. Compare to what the AI predicted. The fraction meeting predicted remaining useful life is the outcome index.
Notation:
RROI = repairs_meeting_predicted_RUL / total_repair_recommendations_with_outcome
Variables:
| Variable | Definition | Unit |
|---|---|---|
| repairs_meeting_predicted_RUL | Repaired parts with actual service life greater than or equal to 0.9 times predicted RUL | Count |
| total_repair_recommendations_with_outcome | Repair recommendations where the part has reached failure or planned replacement within the outcome window | Count |
| RUL | Remaining useful life predicted by the AI at the time of the repair recommendation | Hours or days |
Proposed formula, pending validation pilots.