Predictive Deferral Validation Rate (PDVR)

GMI-HC-023HealthcarePilot-Ready

Definition

PDVR measures the percentage of predictive-maintenance-justified PM deferrals where the equipment operated without failure through the deferred interval. When a predictive model recommends deferring a calendar-based PM because the equipment's predicted remaining useful life extends beyond the next scheduled maintenance window, that recommendation is a bet. PDVR tracks whether those bets pay off, providing the closed-loop validation that turns predictive maintenance from a theory into a measured practice.

The Measurement Gap

Measurement Gap and Decision

Decision Question: When the predictive model recommends deferring a scheduled PM, does anyone track whether the deferral was safe. Whether the equipment actually lasted as long as the model predicted?

What Goes Unmeasured Today:

Predictive maintenance programs generate remaining useful life (RUL) estimates that justify deferring calendar-based PMs. A chiller scheduled for quarterly maintenance gets deferred to six months because the ML model predicts 180 days of remaining useful life. The PM Completion Rate records the deferral as justified. Equipment Uptime shows 100%: right up until it doesn't. Nobody closes the loop. Nobody checks whether the equipment actually survived to the predicted RUL. If the chiller fails at day 90 of a 180-day prediction, the prediction was wrong by 50%, but no metric captures that miss. The planned-to-reactive ratio may actually improve (one fewer planned PM recorded) while reliability degrades. Organizations running predictive maintenance programs are accumulating a portfolio of unvalidated predictions. Some are accurate. Some are not. Without PDVR, there is no way to know the portfolio's batting average, and no way to identify which equipment types or system categories produce unreliable predictions.

Invisible Failure Mode:

Predictive deferral decisions compound. Each unvalidated deferral extends the interval without confirming the model's accuracy. Across dozens of building systems with hundreds of PMs, the aggregate deferral portfolio represents a hidden reliability bet that only surfaces when a failure occurs in a deferred window, often at the worst possible time (Jardine et al., 2006).

Formula

Primary Formula:

PDVR = Validated_Deferrals / Total_Deferrals × 100%

Variable Definitions:

Variable Definition Source
Validated_Deferrals Prediction-justified PM deferrals where the equipment operated without unplanned failure through the entire deferred interval CMMS/EAM + predictive maintenance platform
Total_Deferrals All PM deferrals where the justification was a predictive model's remaining useful life estimate CMMS/EAM deferral records
Unplanned failure Any corrective work order, emergency dispatch, or equipment shutdown on the deferred asset during the deferred interval CMMS/EAM work order records

Data Requirements:

PM deferral records with deferral justification (must be tagged as prediction-based vs. resource-based or other reasons). Predicted RUL or predicted safe interval for each deferred PM. Work order history for each deferred asset during the deferred interval to identify unplanned failures.

Edge Cases and Limitations:

Deferrals justified by resource constraints rather than predictions should be excluded. PDVR measures prediction quality, not scheduling convenience. Equipment that received interim condition-based intervention during the deferred period (not a failure, but a planned action) needs classification rules. Very low deferral volumes per system type produce unstable rates: aggregate by system category until volumes support individual system tracking.

Proposed formula, pending validation pilots.

At a Glance

Category
Predictive Building Maintenance Decision Quality
← Building Engineer Skill Atrophy Index (BESAI)System-Type Prediction Accuracy Variance (STPAV) →

Assess Your Organization

The AI-GPS Assessment evaluates your AI governance maturity and recommends KPIs like this one for your implementation roadmap.

Take the Free AI-GPS Assessment