S&OP: forecast bias, error and the balance of demand and supply
For the monthly sales and operations planning cycle: reconciles the forecast with actual (bias, WAPE, tracking signal) and the supply plan with the demand plan. Shows whether the team overstates the forecast, where the error is large and when supply does not cover demand.
At a glance
- What question does it answer?
- Does the team overstate the forecast, is the error large and is supply enough for demand.
- What data is needed?
- By month: the demand forecast and actual sales in the same units.
- How to read the result?
- Over six months the forecast is 10,700 pieces against an actual 10,650: bias +0.5%, WAPE 9.1%, tracking signal 0.3. All months are within the 25% tolerance. But the supply plan has a shortfall of 300 pieces against the forecast in some months and an excess of 300 in others: the balance does not reconcile by month although it is even in total. Example on the default values: Forecast bias — 0.5% · overstates; WAPE — 9.1%; Tracking signal — 0.3; Supply shortfall against the forecast — 300 units.
Why and how to reason
A forecast can be wrong in two ways: randomly (sometimes above, sometimes below) and systematically (always above or always below). The second is more dangerous because it accumulates. So three numbers are computed. Bias = (sum of forecast − sum of actual) ÷ sum of actual: shows direction. WAPE = sum of absolute deviations ÷ sum of actual: shows size. Tracking signal = cumulative error ÷ mean absolute error: grows if the forecast is biased. Example over 6 months: forecast 10,700, actual 10,650, bias +0.47%, neutral. But absolute deviations 120 + 110 + 160 + 350 + 150 + 80 = 970, so WAPE is 9.1%: monthly errors are large, they just offset. The signal of 0.31 is far from an alarm. The second block checks the supply plan against the forecast: a shortage of 300 units, an excess of 300.
When to use it
It comes in handy at a monthly sales and operations planning (S&OP) meeting: when the forecast is compared with actuals and the purchase plan with the demand plan. You need at least 6 months of data. Assumption: forecast and actual are comparable (the same level: category, not SKU). It does not fit new goods with no history and data where actuals are distorted by lack of stock: then sales are below demand and forecast error is invisible.
How to read the result
Look at bias and WAPE together: a small bias with a large WAPE means random errors (improve the model), a large bias means a systematic error (find who overstates or understates and why). A tracking signal beyond ±4 is an alarm. A month with error above tolerance (you set it) hints where to look for the cause. Supply shortage against forecast is lost sales, excess is stock. Decision: correct the forecast if bias exists, and reorder or move the purchase. It is worth recording at the meeting not only the numbers but also explanations of large deviations; after a few cycles repeating causes show up and can be closed with a rule or a plan change.
Common mistakes
Computing error on the net sum without absolute values (positives and negatives cancel); mixing levels; not setting tolerances; counting shortage and excess independently of months. A forecast that is right on average does not serve purchasing: it happens month by month. Do not compare a forecast at one level with an actual at another.
What data is needed?
- By month: the demand forecast and actual sales in the same units.
- Optional: the supply plan; tolerances for bias and WAPE.
Inputs
| Field | Unit | Value in the example | Meaning |
|---|---|---|---|
| Acceptable WAPE | % | — | Your tolerance: wider for new styles and narrower for basics. There is no single standard; empty: no color and no count of months above tolerance. |
| Acceptable bias (absolute) | % | — | Your tolerance for a systematic skew of the forecast up or down. Empty: no color and no tolerance-based conclusion about the skew. |
How it is calculated
| Step | Formula |
|---|---|
| Bias | (Σ forecast − Σ actual) ÷ Σ actual |
| WAPE | Σ |forecast − actual| ÷ Σ actual |
| Tracking signal | cumulative error ÷ mean absolute error |
| Balance | supply − demand forecast by month; shortfall and excess are summed separately |
Worked numbers
Default values from the app; in the interactive version you enter your own.
- Forecast bias
- 0.5 % · overstates(Σ forecast − Σ actual) ÷ Σ actual.
- WAPE
- 9.1 %Σ |forecast − actual| ÷ Σ actual.
- Tracking signal
- 0.3Cumulative error ÷ mean absolute error. A value outside ±4 is read as a systematic skew (a guide).
- Months with an error above tolerance
- no WAPE tolerance set
- Supply shortfall against the forecast
- 300 units
- Supply excess against the forecast
- 300 units
Forecast, actual and balance
| Month | Forecast | Actual | Error | Error, % | Cumulative error | Supply − forecast |
|---|---|---|---|---|---|---|
| Month 1 | 1,200 | 1,080 | 120 | 11.1 % | 120 | 100 |
| Month 2 | 1,400 | 1,510 | −110 | −7.3 % | 10 | 100 |
| Month 3 | 1,800 | 1,640 | 160 | 9.8 % | 170 | −100 |
| Month 4 | 2,600 | 2,950 | −350 | −11.9 % | −180 | −200 |
| Month 5 | 2,200 | 2,050 | 150 | 7.3 % | −30 | 100 |
| Month 6 | 1,500 | 1,420 | 80 | 5.6 % | 50 | 0 |
Once a month: data (actuals and stock), demand (the forecast), supply (production and deliveries), money (budget and cash flow), a decision (what we change and who owns it). The cycle works if decisions are written down and the forecast error is discussed without looking for someone to blame.
Nuances and limits
- The total hides skews by month: look at the cumulative error too.
- Tolerances are a guide: they are wider for new models and narrower for basics.
- The cycle works if decisions are written down and the error is discussed without looking for someone to blame.