Forecast bias is the tendency to forecast too high or too low, again and again. This workbook calculates bias (forecast minus actual) for every SKU each month, turns it into a tracking signal by region, lists the SKUs whose bias had the same direction for three months, and plots error against value so you fix the items that cost the most first.
What's inside the workbook
| Sheet | What it does |
|---|---|
| Summary | Monthly demand planner and consensus forecast against actuals for each country, with forecast accuracy for both, and charts. |
| FA Summary | Month-on-month accuracy of the demand planner and consensus forecasts by country and region, and how much the consensus changed the planner's number. |
| SKU Accuracy – Consensus | Monthly accuracy of the consensus forecast for every SKU. |
| Tracking Signal & BIAS Summary | The tracking signal by region and month, plus a count of SKUs with consistent over- or under-forecasting over the last three months. |
| BIAS | Monthly bias in value by SKU and the average bias of SKUs that were biased in the same direction for three months while accuracy was below 95%. |
| Error Quadrant | WAPE against sales value by product group for a chosen region, so high-value, high-error groups stand out. |
| Input sheets | Master (the SKU list, hierarchy and net realisable value), then actuals, demand planner forecast and consensus forecast in value. These are the sheets you fill in; they are shaded light brown. |
Preview
| Tracking signal | Oct 21 | Nov 21 | Dec 21 | Jan 22 | Feb 22 |
|---|---|---|---|---|---|
| Country-1 · Region-1 | -0.07 | -0.19 | -0.44 | 0.48 | 0.30 |
| Country-1 · Region-3 | 0.87 | 0.45 | 0.15 | -0.10 | 0.32 |
| Country-1 · Region-2 | -0.44 | -0.87 | 0.85 | 0.81 | 0.97 |
| Country-2 · Region-5 | -0.43 | 0.36 | 0.14 | 0.39 | 0.56 |
| Country-2 · Region-4 | 0.29 | 0.47 | 0.44 | 0.58 | 0.33 |
The tracking signal with the workbook's sample data. Values run from −1 (every miss was under-forecast) to +1 (every miss was over-forecast). Region-2 has been strongly over-forecast for three months in a row, the pattern this report is built to catch. Red highlighting added here for readability.
How to use the forecast bias template
- Fill in the Master sheet with your SKU list, hierarchy (country, region, two category levels) and each SKU's net realisable value.
- Paste monthly values into Actuals-Value, DP-Value (demand planner forecast) and Consensus-Value, in the same SKU order as Master. Everything else calculates from these.
- Start with the tracking signal. Look for regions that stay above about +0.5 or below −0.5 for several months. One month is noise; three months in the same direction is bias.
- Drill into the BIAS sheet to find the SKUs behind it. The three-month average is only filled in when all three months had the same sign and accuracy was under 95%, so the list is already filtered to real, persistent bias.
- Use the Error Quadrant to decide what to fix first: high value and high error before low value and low error.
How the template calculates bias and the tracking signal
A positive bias means over-forecasting (too much stock); a negative bias means under-forecasting (lost sales, expediting). Because the tracking signal divides net error by total absolute error, it always sits between −1 and +1, whatever the size of the region. It is a normalised version of the classic tracking signal, which divides cumulative error by MAD and is read against limits of about ±4.
For a full explanation of bias, its causes and how to correct it, read our guide What is forecast bias?
What to do when you find bias
- Over-forecasting that comes from targets. If the consensus is regularly above the planner's number, sales targets are probably leaking into the forecast. Keep the target and the forecast as two separate numbers.
- Under-forecasting after stockouts. If you forecast from sales and you ran out of stock, the history understates demand, and the next forecast will be too low again. Correct the history for stockout months.
- Bias on declining or new products. Models are slow to react to a product going into decline or a launch taking off. Review these separately.
Planamind calculates bias % as Σ(forecast − actual) ÷ Σ actual and shows it in accuracy diagnostics (labelled over or under), the forecast accuracy report and its Excel download, and the tracking sheet by forecast horizon. Its FVA report flags overrides that made the forecast worse, including items with positive bias and more than three months of stock. Planamind does not correct bias automatically: the planner decides. See demand planning software.
Frequently asked questions
Is the forecast bias template free?
Yes. Fill in the short form and the Excel file downloads immediately.
What is a good tracking signal?
With the normalised version in this template (−1 to +1), values close to zero mean errors cancel out. Values that stay beyond about ±0.5 for several months point to bias. With the classic version (cumulative error ÷ MAD), the usual limits are about ±4.
What is the difference between forecast bias and forecast accuracy?
Accuracy measures how big the misses are; bias measures whether they lean one way. A forecast can be 80% accurate and still be too high almost every month, which builds excess stock.
Why is bias measured in value?
So that a bias on expensive items counts for more than the same unit bias on cheap ones. The workbook uses each SKU's net realisable value from the Master sheet.
Can I use it with a single forecast?
Yes. If you only have one forecast, paste it into both the demand planner and consensus sheets. The demand planner vs consensus comparison will then show no change, and the bias reports still work.