Demand forecasting

Demand forecasting predicts how much of a product or service customers will want in a future period, at whatever level of detail, such as by SKU, store, or region, the planning decision requires. It is a specialized application of time series forecasting, focused on the supply chain decisions that depend on the answer.

Simpler approaches project forward from historical demand using moving averages or exponential smoothing; more sophisticated ones add regression analysis on causal drivers such as price, promotions, or weather, and increasingly blend many signals with machine-learning models. Forecasts are usually generated at multiple time horizons at once, since a factory scheduling decision and a raw-material order need different lead times.

Demand forecasting matters because it sits upstream of inventory levels, production schedules, and staffing plans: forecast too low and a business runs out of stock, forecast too high and it ties up cash in excess inventory. It is closely related to sales forecasting but focused on units of physical demand rather than revenue. Common pitfalls include forecasting new products with no sales history, demand shocks from promotions or events that break normal seasonal patterns, and forecasting at a level of granularity too fine for the data available to support it reliably.