Definition
The fraction of items rationalized out of inventory based on AI obsolescence prediction that are subsequently needed (purchased emergency, found in scrap, demanded by maintenance) within a defined regret window.
The Measurement Gap
Legacy MRO supply chain metrics reward rationalization without measuring regret outcomes. Working capital released rewards the disposition. Slow-mover inventory percentage rewards it again. Obsolescence prediction accuracy is rarely measured at outcome: the AI's 95% confidence that a part will not be needed again produces no feedback when the part is subsequently needed, because nobody tracks regret events. When one of 218 rationalized parts is urgently needed after an unusual failure on an aging asset, the part is no longer in inventory, the OEM no longer makes it, and the emergency replacement through a broker costs multiples of original price while the asset sits down for days. The realized cost of this single regret event substantially exceeds the working-capital savings the entire disposition batch produced. Without regret-rate measurement, the obsolescence prediction model learns to recommend disposition without ever seeing its own failure rate. The decision RRR supports: whether to tighten AI rationalization recommendations, require engineering review by criticality, or extend the buffer-hold period before disposal.
Formula
RRR = needed_after_rationalization / total_rationalized_items
Variables:
- needed_after_rationalization = emergency PO placed for the rationalized part, OR substitute used because original was rationalized, within the regret window
- total_rationalized_items = items disposed based on AI obsolescence prediction
- regret_window = 24 months default, longer for long-service-life assets (adjust by asset expected remaining life)
Worked example: 218 items rationalized based on AI obsolescence prediction. Within 24 months, 11 were subsequently needed: 7 via emergency PO, 4 via substitute because original was gone. RRR = 11 / 218 = 0.050, or 5.0%. By criticality class: critical-class RRR = 8.3% (3 of 36 critical items (operationally dangerous rate); commodity-class RRR = 4.4% (8 of 182) within tolerance).
Edge cases: Long-tail need. Extend window for long-service-life assets; a 24-month window may miss regret events on assets with 10-year overhaul cycles. Regret events on critical parts weighted by cost impact, not just count. Some items are rationalized into broker channels rather than scrapped: recovery cost from brokers is the regret impact.
Data sources: Rationalization decisions from inventory disposition log. Subsequent demand from emergency PO events and substitution events for rationalized items. Predicted obsolescence from obsolescence AI platform. Item criticality from item master and reliability engineering classifications.
Proposed formula, pending validation pilots.