Definition
Mean absolute percentage error between AI-predicted and actual traffic demand at the cell and cluster level, measuring whether capacity investment is being steered by a forecast that holds.
The Measurement Gap
Measurement Gap and Decision
What CFA Measures: CFA computes the mean absolute percentage error between AI-predicted traffic demand and actual traffic at the cell or site level, measured at the forecast horizon relevant to CAPEX decisions. A companion metric, Stranded CAPEX Rate (SCR), translates forecast errors into financial impact.
Decision Supported:
Are the demand forecasts driving CAPEX decisions accurate enough to justify the investment, or is the operator building capacity based on predictions whose accuracy nobody checks? Without CFA, operators approve hundreds of millions in network build-out based on forecast models whose accuracy is assumed rather than measured.
Why This Is Unmeasured:
Demand forecasting models produce projections that feed directly into CAPEX planning cycles, but no standard operational metric closes the loop by comparing predictions to actual demand after the forecast horizon passes. The forecast is consumed, the investment is approved, the capacity is built, and nobody goes back to check whether the model was right. This creates a structural blind spot: forecast models that consistently over-predict drive stranded capacity (infrastructure that sits underutilized), while models that under-predict create avoidable congestion. Both cost money, and neither is visible until CFA is measured (Heavy Reading, "Network CAPEX Forecasting: Closing the Accountability Gap," 2024; IEEE Communications Magazine, "Demand Forecasting Accuracy for 5G Capacity Planning," 2023).
Formula
Plain Language: For each cell or site where a demand forecast influenced a capacity decision, compare what the model predicted the traffic would be at the decision-relevant time horizon versus what actually materialized. Average the absolute errors. The companion Stranded CAPEX Rate measures how much capital was deployed on underperforming forecasts.
Notation:
CFA = 1 − (1/N) × Σ |Demand_predicted_i − Demand_actual_i| / Demand_actual_i
Companion metric:
SCR = CAPEX_on_underperforming_capacity / Total_forecast_driven_CAPEX
Variable Definitions:
- Demand_predicted_i = forecasted traffic demand at cell or site i for the target period (Gbps, Erlang, or sessions, depending on the planning metric used)
- Demand_actual_i = actual traffic demand at cell or site i for the same target period (same units)
- N = number of cell or site forecast-to-actual comparison points (count, dimensionless)
- CAPEX_on_underperforming_capacity = capital expenditure on capacity additions where actual utilization is below 30% of designed capacity 18 months post-deployment (dollars)
- Total_forecast_driven_CAPEX = all capital expenditure driven by the demand forecast model (dollars)
Units:
CFA is a percentage where 100% = perfect forecast accuracy. SCR is a percentage of total forecast-driven CAPEX.
Sign Convention:
Higher CFA is better. Lower SCR is better.
Edge Cases:
- Forecasts for different time horizons (1-year vs. 3-year) should be evaluated separately: accuracy expectations differ materially.
- New site builds where traffic growth depends on marketing campaigns and population growth: flag as higher-uncertainty forecasts and evaluate with wider acceptable bands.
- Minimum sample: 30 cell or site forecast-to-actual comparisons per measurement period.
Proposed formula, pending validation pilots.