Blog · Forecast and pipeline · Supply chain
Demand forecasting turns expected customer demand into a planning input for supply chain teams. Compare methods, work a SKU-and-site example and check forecast errors before using the result for supply decisions.
Demand forecasting in supply chain estimates future customer demand by SKU, site and period before purchasing, production and distribution decisions are made. Its value comes from the decisions it informs and the errors it makes visible. Compare an issued forecast with later actual orders at the same horizon, then check the rows underneath the company total.
A demand forecast answers what customers are expected to request. A supply plan answers how the company intends to meet that demand, given inventory, purchase commitments, production capacity and delivery constraints. Those are connected questions, but they are not interchangeable.
If the company cannot ship an item, shipment history may understate customer requests. The Census Bureau's manufacturing definitions distinguish shipments from new orders and define orders net of cancellations. Inside your own files, record the event and date basis with equal care.
Use the demand forecasting overview for the broad definition. Use supply forecasting when the question is about what suppliers or capacity can provide. The demand review should make the gap between those views visible without quietly changing expected customer demand to fit available supply.
Start with a simple comparison method the business can reproduce. Then test whether another method improves the result over the same periods and forecast horizons. Complexity alone does not establish that it will improve purchasing decisions.
NIST explains that time series analysis accounts for trend, seasonal variation and dependence across observations. Those patterns are a useful starting point for choosing what to test.
| Method or approach | Business interpretation | Main check before use |
|---|---|---|
| Last observed demand | The latest period is the baseline for the next | Was the latest period unusual or constrained? |
| Moving average | Recent periods are averaged to smooth short-term movement | Does smoothing lag a real change in demand? |
| Seasonal comparison | A comparable prior season informs the target period | Are the calendar and selling conditions comparable? |
| Smoothing or trend methods | Recent observations and their direction inform the projection | Does performance hold outside the fitted history? |
| Driver-based scenario | Supplied price, promotion or business assumptions change demand | Are the assumptions documented and available at issue time? |
Different products may need different approaches. A steady replenishment item and an item ordered only occasionally should not be forced into the same interpretation. Specialized workflows such as CPG demand forecasting add category context without changing the need for a reproducible baseline.
The minimum review file has SKU, site, demand period, actual units and units of measure. Add a forecast file with the same keys, a forecast issue date and forecast units. Preserve the issue date: a forecast created before a decision is different from a revision made after demand starts arriving.
Choose whether the period follows order date or requested delivery date. Review cancellations and partial deliveries consistently. Keep availability exceptions beside the actuals so the reader can distinguish quiet demand from demand the company could not fulfill.
Match forecasts and actuals on the complete key. If a forecast is monthly by distribution center, do not join it directly to individual daily shipment lines and repeat the forecast on every line. Aggregate the actuals to the agreed review level before comparing them.
The planner should reconcile input quantities to accepted quantities and account for excluded rows. After that check, calculate each row's signed and absolute error before rolling up the results.
A US distributor makes a September forecast from July and August demand. For illustration, it uses a simple average of those completed months. All quantities below are individual units, and all demand uses requested delivery month after cancellations. No September actuals enter the forecast calculation.
September forecast = (July demand + August demand) / 2
| SKU | Site | July units | August units | September forecast | Calculation |
|---|---|---|---|---|---|
| A | East | 100 | 140 | 120 | (100 + 140) / 2 |
| A | West | 60 | 100 | 80 | (60 + 100) / 2 |
| B | East | 40 | 80 | 60 | (40 + 80) / 2 |
| B | West | 20 | 60 | 40 | (20 + 60) / 2 |
| Total | 220 | 380 | 300 | (220 + 380) / 2 |
The forecast rows sum to 300 units. Averaging the historical totals also returns 300 units, which checks the aggregation because every row uses the same method and periods.
After September closes, compare the frozen forecast with actual demand. Signed error is forecast minus actual; a positive figure means the forecast was high.
| SKU | Site | September forecast | September actual units | Signed error | Absolute error |
|---|---|---|---|---|---|
| A | East | 120 | 100 | 20 | 20 |
| A | West | 80 | 100 | -20 | 20 |
| B | East | 60 | 50 | 10 | 10 |
| B | West | 40 | 50 | -10 | 10 |
| Total | 300 | 300 | 0 | 60 |
The company total is correct, yet the locations are wrong. East was forecast at 120 + 60 = 180 units against 100 + 50 = 150 actual units. West was forecast at 80 + 40 = 120 units against 100 + 50 = 150 actual units. The 30-unit excess at East offsets the 30-unit shortfall at West.
Weighted absolute percentage error = Sum of absolute row errors / Sum of actual units × 100
Here the result is 60 / 300 × 100 = 20.0%, rounded to one decimal place. Net bias is (300 − 300) / 300 × 100 = 0.0%. Both figures are needed: the total shows no directional error, while absolute row errors show that the forecast misallocated demand.
In this example, procurement should not infer a company-wide demand shortage from West's miss alone. The planner first checks whether demand moved between sites, whether customer assignments changed or whether the recent history was unusual. Distribution then reviews whether inventory placement reflects the same pattern, alongside movement constraints and costs.
The forecast does not by itself prescribe a transfer or an order quantity. Those decisions need inventory and supply facts. Its contribution is a clear statement of expected demand and where that expectation differed from the outcome.
Keep the evaluation tied to information available when the forecast was issued. The authors' guidance on forecast accuracy evaluation distinguishes performance on new observations from fit to data already used in the forecasting method. Repeat the review across relevant horizons before treating one good month as evidence that a method works.
Demand planning owns the forecast and its assumptions. Sales confirms customer changes; operations checks timing and capacity; procurement checks supplier commitments; finance connects the agreed quantities to financial assumptions. A supply chain dashboard can present the agreed measures once those definitions and checks are settled.
Bring forecast and actual-order exports with SKU, site, period, units and forecast issue dates. Covirage maps the columns, and its tools compute and check the comparisons; the external AI model explains the results with every figure traced to its rows. Create a free Covirage account in about a minute, then set up your business and bring a sample of your own data.
See Covirage for supply chain teams for the supporting context.
It gives purchasing, production and distribution teams an estimate of customer demand before they commit resources. Its usefulness depends on the decision horizon, the product and location detail, and the quality of the underlying records. It does not remove supplier, capacity or availability constraints.
Start with a simple baseline and compare suitable alternatives on historical periods that were not available when the forecasts were created. Consider whether demand is stable, seasonal, changing or intermittent. Evaluate performance at the horizons and locations where the operating team makes commitments.
Define it for the decision. Requested-delivery orders may better reflect when customers wanted units, while shipments reflect what the business fulfilled. Stock shortages and delayed deliveries can separate them. Preserve both when available, and avoid scoring an order-based forecast against shipments without explaining the difference.
The demand planning owner maintains the forecast definition, assumptions and issued versions. Sales helps explain customer changes; operations and procurement specify the decision horizons and constraints. Finance checks the bridge to financial plans, while each receiving team owns the action taken from the forecast.