Forecast Accuracy Analyzer

Overlay forecast versus actual call volume with configurable standard deviation bands to distinguish real errors from normal variance. Calculates WMIV to show the theoretical accuracy floor at your volume level.

This tool applies the Minimum Interval Variance (MIV) formula to distinguish genuine forecast errors from unavoidable statistical noise in interval-level contact volumes. The core formula is MIV = sqrt(2 / (pi * n)), where n is the forecasted volume for a given interval. This represents the theoretical minimum percentage error achievable even with a perfect forecast, derived from the Poisson-like arrival properties of telephone calls. The tool then constructs sigma bands at 1, 2, or 3 standard deviations around the forecast using the MIV as the base deviation width, corresponding to approximately 68%, 95%, and 99% of expected random variation.

The key assumption is that contact arrivals follow a random process where variance scales inversely with volume — low-volume intervals will always exhibit higher percentage errors regardless of forecast quality. When actuals fall within the sigma bands, the deviation is attributable to normal stochastic variation, not a forecasting failure. Points outside the bands represent genuine forecast misses that warrant investigation. The tool also computes WAPE (volume-weighted absolute error) and directional bias (over/under-forecast tendency) as complementary diagnostics.

Practitioners should use the "within bands" percentage as a quality indicator: if 68% or more of intervals fall within 1-sigma bands, the forecast is performing at or near the mathematical optimum. Intervals consistently outside bands — especially those showing directional clustering — indicate systematic model failures (wrong day-of-week patterns, missed events, trend drift) rather than random noise. Focus improvement efforts exclusively on out-of-band intervals; attempting to reduce in-band error is chasing randomness.