Prediction Source Accuracy Differential (PSAD)

GMI-AVI-025AviationIntegration-Required

Definition

Difference in prediction accuracy between the OEM's component health model and the operator's independent analysis, measured against actual component condition at shop induction: providing an empirical basis for the multi-million-dollar decision of which prediction source to trust for each major component type.

The Measurement Gap

The measurement gap PSAD addresses: airlines operate in a tension between OEM removal recommendations (which are conservative by design and protect the OEM's liability position) and their own reliability engineering analyses, which typically draw on larger samples of fleet-specific operating data. Both sides claim their predictions are more accurate. Without a systematic comparison against actual shop findings, the argument is unresolvable on evidence, and decisions default to institutional deference or contractual relationships rather than accuracy data.

The stakes are substantial. For a major engine component, each unnecessary removal cycle costs on-wing revenue while each overconfident extension risks an in-service failure. Airlines that can empirically demonstrate their operator model is 8 percentage points more accurate than the OEM model have grounds for negotiating extended intervals. Airlines that find their operator model is less accurate than the OEM's can stop overriding OEM recommendations and accept the operating cost. Without PSAD, neither conclusion is reachable.

The decision PSAD supports: when the OEM and operator predictions disagree on removal timing, which source should be followed for which component types? Should the operator invest in improving its independent analysis capability, or defer to the OEM? Should the airline pursue interval extensions with the OEM, and with what evidence? These decisions require an accuracy differential computed against actual outcomes, which is exactly what PSAD provides by matching predictions to shop findings.

Formula

Plain language: For each component where both OEM and operator predictions existed at the time of the removal decision, compare each prediction to the actual condition at shop induction. Compute prediction error for each source. The differential is the difference in accuracy.

Primary notation:
PSAD = MAPE_OEM − MAPE_operator

Where:
MAPE_source = (1/n) × Σ |Predicted_condition_source_i − Actual_condition_i| / Actual_condition_i

Variables:

  • MAPE_OEM = mean absolute percentage error of OEM's prediction vs. actual condition at shop induction (percentage)
  • MAPE_operator = mean absolute percentage error of operator's prediction vs. actual condition (percentage)
  • Predicted_condition_source_i = predicted remaining life or condition score from the respective source at the time the removal decision was made
  • Actual_condition_i = actual component condition at shop induction, quantified by shop findings (remaining margin, measured wear, etc.)

Units: Percentage points. Positive PSAD = OEM is less accurate (operator's prediction is better). Negative = OEM is more accurate. Zero = equivalent accuracy.

Critical data requirements for validity:

Contemporaneous recording: both OEM and operator predictions must be documented at the time of the removal decision, not reconstructed after the fact. Implement a pre-removal decision log that captures both prediction sources, the decision made, and the reasoning.

Standardized condition scoring: "actual condition" at shop induction may be qualitative in current systems. Develop and apply a standardized condition scoring rubric (e.g., 0–100 remaining margin score) before computing PSAD. Without consistent scoring, the comparison is invalid.

Compute PSAD separately by component type: the OEM may be more accurate for engines and less accurate for landing gear. Aggregate PSAD obscures type-specific differences that drive actual removal decisions.

Proposed formula, pending validation pilots.

At a Glance

Category
Predictive Life Management Accuracy
← Operating Environment Prediction Bias Index (OEPBI)Repair-vs-Buy Outcome Accuracy Rate (RBOAR) →

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