Sales, stock and finance can use the same starting point while seeing which extra sales still depend on an assumption. This example records what was known at each review, what was added and what was held back.
Synthetic demonstration, not a client result. Every number, order, review action and role below is invented. This is a learning example, not a validated forecast for a real business. Review the method and business assumptions before using it.
30 September 2026 · Version 1 snapshot. The 1 October review below later holds the order adjustment.
Baseline
140 unitsRepeat September's sales.
Conditional forecast
160 unitsOnly if the extra 20-unit order belongs in October.
Promotion adjustment
HeldThe proposed extra 15 units lack support.
1. Keep the history unchanged
The example covers one product, one sales channel and units sold and delivered in each month of 2026. Assume all recorded sales were available by the last day of each month. Returns and cancelled orders are excluded. Assume no stock shortages in this invented history; real shortages would need investigation because sales can understate demand.
Invented sales history: January–September 2026
Month
Units sold
January
100
February
110
March
105
April
120
May
115
June
120
July
130
August
125
September
140
These nine months do not establish a recurring annual pattern. We do not estimate the effects of time of year, price changes or potential deals from them. Units are not revenue, cash received or all customer demand when stock is limited.
2. Test the baseline on later months
Fix the rule before the July test: forecast next month's sales using the last completed month's sales. At each month-end, use only sales already available. Keep the next month aside for comparison, then move forward one month. The rule is unchanged across all three tests.
For the direction of each miss, subtract the forecast from actual sales. A positive number means the forecast was too low; a negative number means it was too high.
Three one-month-ahead tests, all in units
Forecast made / target
Forecast
Actual
Gap
30 June / July
120
130
10 low
31 July / August
130
125
5 high
31 August / September
125
140
15 low
Ignore the direction when averaging the gaps: (10 + 5 + 15) / 3 = 10 units. This is the mean absolute error: the average size of the miss. Three invented test months demonstrate the calculation, not future business accuracy. This does not mean October sales will be within 10 units of the forecast. It also does not tell you how much extra stock to hold.
The differences with their signs are +10, -5, +15. Their average is (10 - 5 + 15) / 3 = +6.67 units, rounded to two decimals. Actual sales averaged 6.67 units above the forecasts in these three tests. This average direction, sometimes called bias, can hide misses that cancel each other out; it is different from the 10-unit average miss size.
Before using this rule, compare it with other suitable methods using more months that reflect your usual sales. If you change a method after seeing a month's result, that month can no longer be a fresh test of it.
3. Keep October assumptions beside the prediction
As of 30 September 2026, the baseline for October is 140 units. October actual sales are unknown. No October error is calculated. The adjustments below are explicit assumptions, not facts extracted from the history.
October actual sales: unknown. Await the closed month's records; unknown does not mean zero.
Order A: conditional addition of 20 units
Assume an order for 20 extra units has an agreed October fulfilment date. Assume it is extra business beyond the regular sales covered by the baseline. It is not already counted and does not replace an expected order.
Assumption record: sales lead; recorded 30 September; applies to October only; source is an invented order record for this example. Inclusion remains conditional on the stated checks.
Calculation:140 + 20 = 160 units. This is a conditional forecast, not a confirmed sales total. Check real records to confirm when the order will be delivered and whether all 20 units are extra sales.
Promotion: proposed addition of 15 units, held
The sales lead proposes 15 extra units without supporting evidence. Exclude it from the working forecast. The sales lead records the proposal on 30 September for October only; its source is an unsupported estimate. 140 + 20 + 15 = 175 is only the unsupported proposal, not an adopted forecast.
Resolve before inclusion: evidence for the extra sales, their timing, and whether they are already counted in regular sales or Order A.
If Order A moves to November
Remove its 20 units from October: 160 - 20 = 140 units. Record a new version; keep the original 160. Do not automatically add the order to a November forecast that has not been prepared.
4. Preserve the review log
All entries below are invented review actions. A role identifies who would check the input; it is not evidence of a real person's approval.
30 September · Version 1
Sales lead: records the assumed 20-unit October order and why it is extra business. Conditional forecast: 160. Promotion remains excluded; baseline remains 140.
Stock planner: checks timing and availability before making a purchasing commitment. A sales forecast alone does not determine a stock order.
1 October · Version 2 · Missing date
Exception: the delivery month for Order A can no longer be confirmed. Hold its adjustment; do not silently set its quantity to zero or guess the month.
Sales lead: owns the date check. Conditional forecast: pending. Reference baseline: 140, with the unresolved order visible. Keep Version 1 for later comparison.
After October closes · Review pending
Finance reviewer: compares October fulfilled units with the baseline and the dated adjusted forecast, once the actual is available. No October error or improvement can be reported yet.
Record whether each adjustment helped and why it changed. If source sales are corrected later, retain the original record and date the correction.
5. Checks before using this pattern
A missing month, unknown sales value, duplicate product/month row or inconsistent unit must be resolved before producing the baseline. Missing is not zero. Zero recorded fulfilled units is valid when checked.
A negative fulfilled-unit count is invalid under this example's definition. Check returns separately instead of silently changing the number.
An added order needs a quantity, a target month, an owner and a reason it is not already represented. A missing date holds the adjustment, as Version 2 shows.
The additional-order rule has no historical comparison here. Do not say it outperforms the baseline. More complex methods need their own evidence.