Definition
Mean absolute percentage error between AI-predicted consumable consumption by zone and actual consumption measured at restocking.
The Measurement Gap
Measurement Gap and Decision
The Measurement Gap
Traditional consumable management tracks reorder points and safety stock levels. When you run out, you reorder. When you have too much, it sits in storage. Neither approach measures forecast quality because there was no forecast to evaluate: just rules of thumb and buffer quantities.
AI-driven consumable forecasting changes the operating model from reactive to predictive. But predictive only helps if the predictions are accurate. Without CFA, procurement teams cannot distinguish between a model that genuinely predicts demand and one that generates forecasts within an acceptable range by defaulting to historical averages. The cost of bad forecasts compounds: stockouts create emergency orders at premium prices, while overstock ties up budget and creates waste from expired products.
The Decision It Supports
CFA tells procurement and FM leadership whether the AI forecasting model adds value over simple historical averaging. High accuracy justifies reliance on the model for purchasing decisions. Low accuracy means the model needs retraining, better input data (occupancy patterns, event schedules, seasonal factors), or replacement. The metric also identifies which consumable categories forecast well and which resist prediction: cleaning chemicals may track occupancy closely while specialty supplies may not.
Formula
Plain Language
For each consumable category, compare the AI-predicted quantity for the period against actual consumption. Compute the absolute percentage error. Average across categories for the composite CFA score.
Formal Notation
$$CFA = 1 - \frac{1}{N} \sum_{c=1}^{N} \frac{|Forecast_c - Actual_c|}{Actual_c}$$
Variable Definitions
| Variable | Definition | Unit |
|---|---|---|
| Forecast_c | AI-predicted consumption for category c in the period | Units (category-specific) |
| Actual_c | Measured actual consumption for category c in the period | Units (category-specific) |
| N | Number of consumable categories included | Count |
Units: 0 to 1. A value of 1.0 means perfect forecast. A value of 0.8 means 20% average error across categories. Values can go negative if errors exceed 100%, indicating the forecast is worse than useless.
Proposed formula, pending validation pilots.
At a Glance
Decision Supported
Is the AI accurately predicting consumable demand, or systematically over-stocking (waste) or under-stocking (stockouts)?