Demand forecasting
Demand forecasting is the estimate of how many units of each product a store will sell in a future period, built from sales history, trend and seasonality. An ecommerce operator uses the forecast to size purchase orders, time reorders and plan cash - buying to a number instead of a gut feel.
| Variable | What it covers |
|---|---|
| Baseline demand | Recent average unit sales per period (for example the trailing three months), cleaned of stockout gaps and one-off spikes. |
| Trend factor | The period-on-period growth or decline running through that baseline. |
| Seasonality index | How the target period historically compares to an average period - 1.0 means average, 1.3 means 30% above. |
This is the simplest standard decomposition. More sophisticated models change the machinery, not the question: units, per SKU, per period.
Worked example
Example numbers. The daily-demand line is where the forecast meets the reorder point.
What is a good demand forecast?
Forecasts are judged by forecast error - actual units against forecast units - and there is no universal accuracy number to hit. Achievable error depends on how volatile the category is, how deep and clean the sales history runs, how many SKUs spread the volume thin, and how heavily promos distort the pattern. Two tests matter regardless: the forecast must beat the naive "same as last period" baseline, or it is adding ceremony rather than information; and the error must be measured every cycle so the model gets corrected instead of trusted. Wide uncertainty is fine - the honest response is smaller, more frequent orders, not a more confident number.
Demand forecasting vs related metrics
| Metric | What it measures | How it differs from demand forecasting |
|---|---|---|
| Reorder point | The stock level that triggers a new order. | A consumer of the forecast: expected daily demand times lead time, plus safety stock. |
| Days of cover | Stock on hand divided by daily demand. | Reads the forecast backwards - how long today's inventory lasts if the forecast holds. |
| Sell-through rate | Share of received units actually sold in a period. | Backward-looking - what happened. The forecast is the forward view it feeds. |
| Scenario analysis | The plan re-run under different assumption sets. | The forecast supplies the base case; scenario analysis stresses it up and down. |
Common mistakes
- Forecasting dollars instead of units. Purchase orders are placed in units per SKU - a revenue forecast cannot size a buy.
- Reading stockouts as low demand. A month at zero stock is censored demand, not absent demand - leaving it in drags the baseline down and guarantees the next stockout.
- Baking promos into the baseline. A discount-driven spike repeats only if the discount does. Separate promo lift from organic demand before extrapolating.
- One aggregate forecast for the whole catalog. Total units can hold steady while the mix shifts underneath - and the mix is what you have to buy.
- Never measuring forecast error. An unmeasured forecast never improves, and nobody learns whether to trust it with real purchase-order money.
Demand forecasting FAQ
Related
Blufire S11 Planning & Forecasting builds the demand forecast from your own order history and lets you stress the assumptions in the Scenario Lab before the purchase order is placed.
Updated July 2026