Blog · Forecast and pipeline · Supply chain
Demand forecasting estimates what customers will request in a future period. Build a simple baseline, separate demand from recorded sales, and check forecast errors against the rows behind them.
Demand forecasting estimates what customers will request in a future period, by product, service, location, or channel. It informs purchasing, production, staffing, and inventory. Recorded sales do not capture every request a company could not fulfill.
A demand forecast is incomplete without its unit and horizon. A distributor deciding next week's warehouse replenishment needs item quantities by site. A service company scheduling employees needs expected workload by day. Finance may translate those quantities into dollars afterward.
Forecasting also differs from setting a target. Forecasting: Principles and Practice distinguishes forecasts, goals, and plans: forecasts estimate what is likely, goals state what management wants, and plans describe the response. Keep the approved operating target alongside the forecast rather than replacing the forecast with it; budget vs forecast covers that reporting distinction.
Ask how many cases of each item customers will request from the Ohio warehouse next week. That fixes the location, unit, and period before choosing a method.
Use order lines with customer request date, item, fulfillment location, requested quantity, shipped quantity, cancellation reason, and availability status. Keep cases separate from individual units. Retain the forecast publication date and forecasted period so that a later revision cannot overwrite the original comparison.
Demand needs an explicit definition. An order canceled because a customer changed plans is different from a documented request canceled because inventory was unavailable. A shipment delayed into another week still belongs to the original demand period if the question concerns when customers requested delivery.
Recorded revenue has another definition. The Census Bureau's retail sales definitions describe sales of merchandise, net of refunds and allowances, rather than all requests customers might have made. A revenue export therefore needs additional order or availability evidence before it can stand in for unconstrained demand.
Consider an illustrative US distributor planning three items at its Ohio warehouse. There are no known promotion changes in the next week. Historical quantities below are validated customer requirements in cases, already assigned to the requested delivery week.
Four-week average forecast = Sum of demand in the previous four weeks / 4
Forecast error = Actual demand − Forecast demand
There is no universal demand forecasting formula. This average is a starting benchmark for these rows, not a claim that every item should use the same method.
| Item | Week 1 cases | Week 2 cases | Week 3 cases | Week 4 cases | Four-week total cases | Next-week forecast cases |
|---|---|---|---|---|---|---|
| A | 1,100 | 1,200 | 1,150 | 1,350 | 4,800 | 1,200 |
| B | 750 | 800 | 850 | 800 | 3,200 | 800 |
| C | 450 | 500 | 550 | 500 | 2,000 | 500 |
| Total | 2,300 | 2,500 | 2,550 | 2,650 | 10,000 | 2,500 |
For A, add 1,100 + 1,200 + 1,150 + 1,350 = 4,800; divide by 4 to get 1,200 cases. Repeat for B and C. The check is that the item forecasts sum to 2,500, which also equals 10,000 / 4.
Stable, seasonal, and sporadic items should be tested separately. Intermittent demand forecasting handles the special case of long gaps between orders; retail demand forecasting addresses store availability and promotions.
The next week closes with these results. Unfilled cases are documented customer requirements due that week, excluded from shipments, and counted once. They include no duplicate backorders or unsupported estimates of lost demand.
| Item | Forecast cases | Shipped cases | Validated unfilled cases | Actual demand cases | Error cases | Absolute error cases |
|---|---|---|---|---|---|---|
| A | 1,200 | 1,000 | 150 | 1,150 | −50 | 50 |
| B | 800 | 900 | 0 | 900 | 100 | 100 |
| C | 500 | 400 | 60 | 460 | −40 | 40 |
| Total | 2,500 | 2,300 | 210 | 2,510 | 10 | 190 |
For A, actual demand is 1,000 + 150 = 1,150 cases; error is 1,150 − 1,200 = −50. The same steps produce the other rows. The reconciliation is 2,300 shipped + 210 unfilled = 2,510 actual cases.
The aggregate miss is only 10 cases, yet absolute item errors add to 190. Dividing 190 by 2,510 gives weighted absolute percentage error of 7.6%, rounded to one decimal place. A total-only review would miss B's underforecast and the opposite errors in A and C. Use SKU and site forecast bias to investigate whether those directions persist.
These are observable requirements, not a complete count of customers who left without placing an order. If lost demand is estimated separately, label the estimate and its assumptions instead of silently adding it to actuals.
The demand planner owns the forecast review. The warehouse owner investigates unfilled requests, and procurement checks whether replenishment timing or supplier availability explains them. A forecast revision alone will not resolve an item that cannot be supplied.
For B, investigate whether the increase came from a repeat customer, a promotion, or a one-time order before changing future weeks. For A and C, determine whether the excess forecast is persistent. Keep customer evidence with any override, including who approved it and which periods it affects.
Review exceptions at the level where someone can act, then consolidate them in a forecasting dashboard. The review should distinguish a changed demand expectation from a changed fulfillment constraint.
Bring an order export, the corresponding forecast snapshot, and the records identifying validated unfilled requests. Covirage maps the columns, and its tools compute and check forecast-versus-actual comparisons; the external AI model explains the results with every figure traced to its rows. Create a free Covirage account and bring a sample of your own data.
See Covirage for supply chain teams for the business context.
Demand forecasting estimates customer requirements. Sales forecasting estimates the sales a company expects to record, often reflecting capacity, availability, price, and sales execution. The figures can agree when supply is available, but shortages can prevent customer demand from becoming recorded sales.
A demand planner or supply chain planning team usually coordinates the forecast. Sales contributes customer changes, marketing supplies promotion assumptions, operations checks capacity, and finance reconciles financial implications. Assign one owner to the published version and keep the approvals attached to it.
A recent average can provide a transparent starting point for a stable item. Test it against future periods before relying on it. Strong seasonality, changing demand levels, promotions, or many periods without orders may require a different method and more careful business assumptions.
Update it at a frequency that supports the decision and the replenishment lead time. Weekly reviews may support short replenishment cycles; monthly reviews may support capacity and purchasing commitments. Keep earlier versions so that updates do not erase the forecast that informed a decision.