Definition
Degree to which AI-driven maintenance resource allocation across asset classes is proportional to each class's actual risk contribution, revealing classes systematically starved of attention.
The Measurement Gap
Measurement Gap and Decision
What PPES Measures: PPES quantifies the degree to which the AI prediction portfolio's maintenance resource allocation is proportional to the economic risk across all equipment classes. It reveals whether rare-but-critical equipment is systematically under-served by the predictive model.
Decision Supported:
Is the predictive maintenance queue allocating proportional attention to all equipment classes, or is it biased toward well-represented equipment types while critical outliers are neglected? Without PPES, a prediction program can show excellent aggregate performance while completely ignoring equipment classes that represent the highest consequence of failure.
Why This Is Unmeasured:
Predictive maintenance reporting focuses on aggregate metrics. Total alerts, total interventions, model accuracy across the fleet. No standard metric compares the share of predictive attention each equipment class receives against the share of total economic risk it represents. ML models inherently favor equipment types with abundant training data (common, frequently-failing components) and under-serve rare-but-critical assets (e.g., DWDM amplifiers, microwave radios in remote locations) where training data is sparse. The result is a prediction portfolio that looks productive on average but has systematic blind spots on the equipment whose failure carries the highest business impact (Gartner, "AI Bias in Predictive Maintenance Portfolios," 2024; IEEE Reliability Engineering, "Coverage Equity in Industrial ML Systems," 2023).
Formula
Plain Language: For each equipment class, compare the share of predictive maintenance resources it receives to the share of total equipment economic risk it represents. A class-level ratio of 1.0 means proportional allocation. Below 1.0 means under-served. The portfolio score normalizes the sum of all class-level deviations from proportionality.
Notation:
PPES_c = Maintenance_share_c / Risk_share_c
PPES_portfolio = 1 − Σ |PPES_c − 1.0| / (2 × N_classes)
Variable Definitions:
- Maintenance_share_c = percentage of total predictive maintenance interventions allocated to equipment class c (percentage, 0–100)
- Risk_share_c = percentage of total portfolio economic risk represented by equipment class c, computed as failure probability × failure cost × installed base count for class c divided by the sum across all classes (percentage, 0–100)
- N_classes = number of equipment classes in the portfolio (count, dimensionless)
- PPES_c = class-level equity ratio: 1.0 is proportional, below 1.0 is under-served, above 1.0 is over-served
- PPES_portfolio = normalized portfolio score, 0 to 1, where 1.0 = perfectly proportional allocation across all classes
Units: PPES_c is a dimensionless ratio per class. PPES_portfolio is a normalized score from 0 to 1. Higher is more equitable.
Sign Convention:
Higher PPES_portfolio is better. Individual PPES_c values near 1.0 are ideal; values approaching zero for any class indicate the prediction system has a critical blind spot.
Edge Cases:
- Classes with zero predictions (model has no training data): flag as unmeasured risk with PPES_c = 0. The model is blind to this class, not making an allocation decision.
- Risk estimation requires failure probability data that may not exist for newer equipment types. Use manufacturer MTBF estimates as a starting point, flagged as estimated rather than observed.
- Minimum sample: requires at least 5 equipment classes with 10 or more predictions each for the portfolio score to be meaningful.
Proposed formula, pending validation pilots.