Demand Planning Calculator

For SKUs with strong seasonality (a single method applied year-round will consistently misforecast around seasonal peaks and troughs), heavy promotional volatility (promotional periods need to be normalized out of the history first), or genuinely new SKUs with fewer than three periods of real sales history. Seasonal naive methods, exponential smoothing (ETS), or ML-based forecasting handle these cases better. See the Demand Forecasting and Sensing guide for the fuller framework, including when to layer real-time demand sensing on top of a baseline forecast like this one.

Yes. The 0.2/0.3/0.5 split is a common default, not a fixed rule. Any three weights that sum to 1.0 are valid; increase the most-recent-period weight for fast-moving, trend-sensitive categories, or flatten the weights toward equal (closer to a simple average) for stable, low-volatility categories where you want less reactivity to short-term noise.

Three periods is a reasonable minimum for a fast, manual weighted-average approach. More sophisticated forecasting (ETS, ARIMA, ML-based methods) typically uses 12 or more periods to capture seasonality and longer trend cycles properly. This calculator's three-period method is best suited to a quick baseline check, not a full annual demand plan.

Only after being normalized. Include the promotional period's baseline (non-promotional) demand estimate rather than the actual inflated sales figure, or the forecast will systematically overpredict a non-promotional next period. Clean input data matters more to forecast accuracy than the specific weighting method chosen.

The forecast from this calculator feeds directly into the expected-demand-during-lead-time term of both formulas. A forecast that's chasing noise (too-reactive weighting on a volatile SKU) will produce reorder points and safety stock levels that swing unnecessarily; a forecast that's too slow to react to genuine trend shifts will produce stockouts or excess inventory as real demand outpaces or undershoots the plan.

Accuracy (commonly measured as MAPE) captures how far off the forecast was on average, in either direction. Bias captures whether the forecast is systematically too high or too low over time. A weighted moving average with the wrong weight distribution for a given SKU's volatility pattern can be accurate on average while carrying a persistent bias in one direction; check both, not just accuracy alone.

Not on its own. This is a fast, tactical baseline for a single next-period forecast on a relatively stable SKU. A full annual demand plan needs seasonality curves, promotional calendars, and typically more sophisticated statistical or ML methods layered on top of a baseline like this one.

Recalculate every period (weekly or monthly, matching however "period" is defined for the SKU) using the newest three periods of actual data. A stale forecast, computed once and left unchanged for several periods, loses most of its value as actual demand data accumulates and the underlying trend potentially shifts.