Definition
The economically weighted fraction of AI recommendations that fail to become timely, correctly executed work: capturing late execution, cancellations, and material alterations relative to the useful prognostic window.
The Measurement Gap
Legacy maintenance metrics were designed for work orders generated by inspections, PMs, or operator requests. Sources with no time-sensitive prediction window attached. When the source is a predictive model with a finite prognostic horizon, every day of delay between alert and execution erodes value. Schedule compliance treats an AI-triggered alert and a routine PM identically; if both slide a week, both score the same, but the AI alert may have burned its entire useful window. MTTR starts the clock at first wrench; it cannot see the three weeks between alert and first touch. Precision and recall measure model accuracy assuming the recommendation actually became work: they do not measure organizational translation fidelity.
Decision Supported:
Which specific handoff layer (alert-to-WO, WO-to-schedule, schedule slippage, materials, dispatch, execution) is leaking value, and which process owner can fix it.
Formula
Plain language: For every AI recommendation in the period, decide whether it was executed in time, as recommended, and weight each recommendation by economic importance. The percentage of weighted value that did not survive the journey from alert to wrench is the translation loss.
Notation:
RETL = Σ_i (w_i × L_i) / Σ_i w_i
Variables:
| Variable | Definition | Unit |
|---|---|---|
| L_i | 1 if recommendation i was not executed within the useful window OR was materially altered; 0 otherwise | Binary |
| w_i | Economic weight: asset criticality × failure consequence | $ or weighted score |
| Useful Window | Prognostic horizon minus minimum organizational response time | Time |
| Material Alteration | Scope, timing, or method changed enough that the AI's intent was not fulfilled | Boolean |
Useful Window connects directly to Actionable Prognostic Horizon (APH): APH minus standard response time defines the denominator for "in time." Sub-metric decomposition (alert-to-WO, WO-to-schedule, schedule slippage, materials, dispatch, execution) reveals where the leak actually occurs.
Proposed formula, pending validation pilots.