Inspection Classification Equity Index (ICEI)

GMI-WT-008Water TreatmentIntegration-Required

Definition

Measures whether an AI pipe inspection system's accuracy varies systematically across asset classes (pipe material, diameter, defect type), reported as the ratio of the worst-class accuracy to the best-class accuracy.

The Measurement Gap

AI pipe inspection may systematically misclassify defects on certain pipe materials (clay versus PVC versus concrete versus ductile iron), certain diameters, or certain defect types. A model might be excellent at identifying corrosion in large concrete mains but systematically underrate joint defects in small clay laterals. PACP scores are the output of the inspection process: no legacy metric measures per-condition-class accuracy. Rehabilitation prioritization absorbs the bias silently: money flows to what the AI sees well, not necessarily to what needs repair most. The invisible failure mode is systematic underinvestment in asset classes the AI reads poorly, compounding over years of capital programming.

Decision Supported:

Is the AI inspection system equally reliable across all pipe types in your system, and which asset classes should receive mandatory human quality review?

Formula

Plain language: Take a sample of AI-inspected pipes across every material type, diameter range, and major defect category. Have a certified human coder review the same footage. For each asset class, calculate what percentage of AI classifications match the human expert. Divide the worst-performing class's accuracy by the best-performing class's accuracy.

Notation:

ICEI = min(Acc_c) / max(Acc_c)

Where Acc_c = Σ Match(AI_j, Human_j) / N_c for all inspections j in class c.

Variables:

Variable Definition Unit
Acc_c AI classification accuracy for asset class c Fraction (0–1)
Match(AI_j, Human_j) 1 if AI and human agree on defect presence, type, and severity grade for inspection j; 0 if they disagree Binary
N_c Number of inspections in class c with human ground truth (minimum 30 per class proposed) Count
c Asset class defined by combination of pipe material, diameter band, and defect type group n/a

Proposed formula, pending validation pilots.

At a Glance

Unit
Ratio (0 to 1)
Trend Direction
Toward 1.0
Reporting Frequency
Quarterly (requires accumulation of human-verified samples)
Category
AI Decision Quality
← Digital Twin Fidelity Index (DTFI)Model Accuracy Trend (MAT) →

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