Blog · Forecast and pipeline · Supply chain
A baseline demand forecast estimates normal demand before temporary events and availability losses. Work through a product-month example that separates baseline, promotion lift, one-time demand and stockouts, then reconciles the result to actual sales.
A baseline demand forecast estimates normal customer demand before temporary lifts and stockout losses. It can include recurring seasonality and trend, so normal demand need not be flat. To compare the baseline with actual sales, show each adjustment separately and reconcile the remaining difference.
Set the baseline at a specific grain: product, location or channel, and month. State the unit, forecast issue date and treatment of recurring events. A baseline for cases shipped from a distributor is different from one for units purchased at a store.
The baseline definition gives the general comparison point. Here, the comparison point is demand under stated normal trading conditions. The broader demand forecasting workflow covers how that estimate fits into planning.
Seasonality belongs in the baseline when it represents repeatable normal behavior. Hyndman and Athanasopoulos distinguish seasonal patterns tied to a fixed calendar period from changes without a fixed frequency in Forecasting: Principles and Practice. A regularly stronger summer month may therefore have a higher baseline without needing a separate promotional lift.
Decide how to treat recurring holiday campaigns. One business may include ordinary holiday trading in the baseline and add only the extra effect of a special campaign. Another may estimate a nonpromotional baseline throughout. Either convention can support a useful comparison; switching between them cannot.
Keep historical sales observations separate from assumptions. The sales export should contain product, location, period and units sold, with returns identified. The assumption file adds the issued baseline, promotion dates, estimated incremental units, exceptional orders and availability restrictions.
Give each adjustment an owner and supporting evidence. A promotion calendar proves that an event happened; it does not prove the number of incremental units attributed to it. A stockout log establishes unavailable time; the associated missed demand is still an estimate.
Use these columns for the comparison:
| Field | Meaning |
|---|---|
| Baseline units | Normal demand under the agreed trading conditions |
| Promotion lift units | Estimated demand above baseline attributable to promotions |
| Exceptional demand units | Identified temporary demand outside ordinary trading |
| Lost demand units | Estimated demand not served because of availability restrictions |
| Actual sales units | Recorded sales for the same product, place and period |
| Forecast version | The assumptions available when the forecast was issued |
Document why an adjustment was accepted. The authors' guidance on judgmental adjustments emphasizes documenting and justifying overrides to statistical forecasts. The same discipline keeps a baseline review from becoming a collection of unexplained corrections.
A US beverage distributor reviews one product across its domestic accounts. The baseline rises with the expected seasonal pattern. February has a promotion; March has a one-time customer order and an estimated stockout loss. All quantities below are units.
Expected sales = Baseline + Promotion lift + Exceptional demand − Estimated lost demand
Unexplained difference = Actual sales − Expected sales
| Month | Baseline | Promotion lift | Exceptional demand | Estimated lost demand | Expected sales | Actual sales | Unexplained difference |
|---|---|---|---|---|---|---|---|
| January | 1,000 | 0 | 0 | 0 | 1,000 | 980 | -20 |
| February | 1,100 | 300 | 0 | 0 | 1,400 | 1,420 | 20 |
| March | 1,200 | 0 | 200 | 150 | 1,250 | 1,230 | -20 |
| Total | 3,300 | 300 | 200 | 150 | 3,650 | 3,630 | -20 |
February's expected sales are 1,100 + 300 = 1,400. March's are 1,200 + 200 − 150 = 1,250. The unexplained differences are therefore 1,420 − 1,400 = 20 and 1,230 − 1,250 = -20, respectively.
The total bridge proves the comparison:
3,300 + 300 + 200 − 150 − 20 = 3,630 actual units
Actual sales exceed baseline by 3,630 − 3,300 = 330 units. That excess is not the promotion estimate. It comprises 300 promotional units, 200 exceptional units, an estimated 150-unit availability loss and a negative 20-unit unexplained difference.
The worked promotion lift example explains how to test a promotion separately. Keep that estimate as a supported input here rather than deriving it from the overall gap.
You can also remove the stated event effects from actual sales and add back estimated lost demand. Call the result adjusted sales, with an explicit warning that it combines observations and assumptions.
Adjusted sales = Actual sales − Promotion lift − Exceptional demand + Estimated lost demand
| Month | Actual sales | Calculation | Adjusted sales | Adjusted sales − baseline |
|---|---|---|---|---|
| January | 980 | 980 − 0 − 0 + 0 | 980 | -20 |
| February | 1,420 | 1,420 − 300 − 0 + 0 | 1,120 | 20 |
| March | 1,230 | 1,230 − 0 − 200 + 150 | 1,180 | -20 |
| Total | 3,630 | 3,630 − 300 − 200 + 150 | 3,280 | -20 |
Adjusted sales minus baseline must equal the unexplained difference in the first table, both by month and in total. Here, 3,280 − 3,300 = -20. If the two views disagree, check signs, missing rows or duplicate adjustments before discussing performance.
These adjusted sales are not directly observed demand. For a fuller treatment of availability restrictions, use the unconstrained demand forecast.
The demand planner owns the baseline definition and version history. Sales or marketing supports event assumptions, operations supplies availability evidence, and finance checks the commercial interpretation. Keep units and revenue separate until a price assumption has been stated.
Review repeated unexplained differences by product and location. Persistent negative differences could indicate an overstated baseline, overstated lift or understated availability losses. The arithmetic identifies what needs investigation; it does not choose the explanation. Use forecast bias by SKU and site for the detailed diagnostic.
Freeze the baseline before the month begins, then evaluate it after the month closes. Forecast evaluation guidance distinguishes forecasts assessed on new observations from fit to observations already used. Rewriting January's issued baseline after January closes would conceal the error you need to learn from.
Bring a sample of your own product-month sales export, baseline assumptions and event or availability history to Covirage. 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, which takes about a minute, then set up your business and bring your file.
See Covirage for supply chain teams for the supporting business context.
Yes, when seasonality is part of normal demand. Recurring holiday or weather patterns can belong in the baseline. Record the convention explicitly so the same seasonal effect is not also added as a separate event adjustment.
No. Last month's sales may include discounts, shortages, unusual orders or a different selling calendar. They are evidence for a baseline, but they become a usable comparison only after those differences are identified and assessed.
Review it on the planning cycle used by the business, but preserve each issued version. Change assumptions when there is evidence of a persistent shift, and keep temporary events separate so an exceptional month does not automatically reset normal demand.
No. Actual sales minus baseline also contains availability effects, exceptional orders and ordinary forecast error. Estimating promotion lift needs a defensible comparison and checks for timing shifts or substitution, rather than assigning the entire difference to the promotion.