What is demand forecasting?

Demand forecasting tries to answer how much of each SKU the next week, month, or season will consume, before it happens. The simplest methods work from history: a moving average of recent usage, adjusted for trend and seasonality. More involved methods layer in known future events, like a signed contract that will consume 400 units in March, a promotion, or a product launch that will cannibalize an older SKU. For most operations, the honest starting point is humble: a rolling average of the last 8 to 12 weeks of usage per SKU, with a manual override for items where someone knows something the history doesn't show. That alone beats reordering off gut feel, and it's only possible if usage history exists, which is the quiet prerequisite. A business that doesn't record consumption can't forecast it. Where teams trip: forecasting the aggregate instead of the item. "We'll sell 20% more this year" doesn't tell purchasing whether to buy more of SKU A or SKU B. The forecast has to land at the level where POs are written. The second trap is ignoring forecast error. Comparing forecast to actual each month, even crudely, shows which items are predictable enough to automate and which need a human watching them.