Demand Forecasting: Definition & Methods

TL;DR: Demand forecasting estimates how much of a product customers will want during a specified future period. It combines available evidence with stated assumptions to support purchasing, staffing and inventory decisions. A forecast is an estimate, not a guarantee or a purchase order.

Last updated: September 2026

What is demand forecasting?

Demand forecasting estimates future customer demand for a defined product, location and time period. A useful forecast specifies what is being measured: units of a particular SKU next week, for example, rather than an unspecified expectation that sales will grow.

Observed sales and demand can differ. When a product is unavailable, recorded sales do not reveal all the orders customers might have placed. Keep stock availability, promotions, returns and changes in product range alongside the sales history so that a shortage or one-off event is not silently treated as normal demand.

A simple demand-forecasting example

Suppose a shop sold 80, 100, 90 and 110 units of one product over four comparable weeks when it was available throughout. The four-week average is 95 units per week: (80 + 100 + 90 + 110) / 4. Using that average for next week is a simple baseline forecast.

This is an illustrative calculation, not a merchant result. A planned promotion, a seasonal change or a new competitor could make those weeks less representative. Record any adjustment and its reason, then compare the estimate with what happens. The 95-unit forecast is not automatically the order quantity: purchasing also depends on usable stock, incoming deliveries, lead time and the chosen inventory buffer.

Does demand forecasting require AI?

No. A recent average or the corresponding period from a previous season can provide a starting point. Statistical methods and machine-learning models offer other ways to estimate demand. Compare candidate methods with a simple baseline before adding complexity.

Forecasting and automation serve different purposes. A forecast estimates future demand; an automated workflow can move data, schedule a report or act on an approved rule. A business can use either or both. Automating a forecast does not establish that it is accurate.

How do you assess a forecast?

Test a method on later observations that it did not use for training, using a forecast horizon relevant to the decision. Check both the size of errors and whether forecasts repeatedly run too high or too low. More historical data does not guarantee better results if the demand pattern has changed.

Choose an error measure appropriate to the data. Mean absolute error expresses average error in the units being forecast. Mean absolute percentage error expresses it as a percentage, but is undefined when actual demand is zero and can be unstable near zero. A lower percentage error alone does not determine the safety stock required.

How does a forecast support inventory planning?

A demand forecast is one input to inventory planning. Combine it with the stock position, expected replenishment timing and uncertainty, then decide how much inventory to hold or order. Review whether the resulting decisions meet the business's service and cost objectives. Forecasts can inform efforts to reduce shortages and excess stock; they cannot eliminate those risks.

Further reading

• Simple forecasting methods and baselines
• Evaluating forecast accuracy