Condition Forecast Calibration Error (CFCE)

GMI-FAC-008Facilities ManagementIntegration-Required

Definition

Weighted deviation between AI-predicted condition deterioration trajectories and actual observed condition changes: measuring whether the capital plan's timing assumptions correspond to physical reality.

The Measurement Gap

There is no metric that measures whether AI condition forecast confidence corresponds to actual prediction accuracy, which means capital plans treat point estimates as facts while uncertainty goes unreported.

Capital planning in facilities management increasingly relies on AI-driven condition forecasting to time major system replacements: roofs, HVAC systems, elevators, facades, electrical distribution. The AI produces a remaining useful life estimate, and the capital plan schedules the replacement accordingly. But the estimate carries an uncertainty band that rarely appears in the capital plan. If the AI predicts 8 years of remaining life with a confidence interval of plus or minus 3 years, failure as early as Year 5 is within the uncertainty band. The capital plan that defers replacement to Year 7 is betting on the point estimate, not the distribution.

When the forecast is wrong, the consequences are severe. Emergency replacements of major building systems cost 30-50% more than planned replacements due to expedited procurement, overtime labor, temporary equipment rental, and occupant displacement. A roof failure that was predicted for Year 8 but occurs in Year 5 doesn't just cost the replacement: it costs the water damage, the business interruption, and the tenant relationship.

Traditional condition metrics (Facility Condition Index, deferred maintenance backlog) measure the current state. They don't measure whether the AI's projection of future state is trustworthy. CFCE fills that gap by tracking prediction accuracy over time, building a track record that capital planners can use to calibrate their confidence in future forecasts.

Decision Supported

When the AI predicts remaining useful life for a major building system, should the capital planner schedule replacement based on the point estimate, add a safety margin, or require engineering review? CFCE provides the historical accuracy data needed to make that calibration decision by system type.

Formula

Plain Language

For assets where the AI predicted condition at a future date and that date has now arrived or passed, compare the predicted condition to the actual assessed condition. Weight by asset replacement cost so that expensive assets receive proportionally more attention.

Notation

CFCE = Σ (w_a × |Condition_predicted_a − Condition_actual_a|) / Σ w_a

Variables

Variable Definition Unit
w_a Replacement cost of asset a, used to weight expensive assets more heavily Dollars
Condition_predicted_a AI-predicted condition score for asset a at the assessment date Condition scale (0-5 or 0-100)
Condition_actual_a Actual assessed condition score for asset a at the assessment date, from independent inspection Condition scale (0-5 or 0-100)

Units: Condition-score units matching the organization's scale. Lower is better. Zero means the model perfectly predicted every asset's condition trajectory. Direction matters: systematic overestimation of remaining life (dangerous (deferrals are riskier than planned) versus systematic underestimation (expensive) replacing too early).

Proposed formula, pending validation pilots.

At a Glance

Category
Digital Twin & Capital Planning

Decision Supported

When the AI says "this roof has 8 years of remaining life," how much confidence should the capital planner place in that number, and what's the model's track record on similar predictions?

← Capital Impact Ranking Delta (CIRD)ADA Compliance Continuity Rate (ACCR) →

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