Calculator
Enter periods with actual sales and the forecast. Add rows as needed. Blank or invalid rows are ignored.
Example data table
Use this sample to see typical behavior. Positive tracking signal suggests consistent under-forecasting if you use Actual - Forecast.
| Period | Actual | Forecast | Error (A - F) |
|---|---|---|---|
| 1 | 120 | 110 | 10 |
| 2 | 98 | 105 | -7 |
| 3 | 135 | 130 | 5 |
| 4 | 115 | 120 | -5 |
| 5 | 140 | 128 | 12 |
Formula used
Step 1: Compute signed error for each period.
If you choose Actual − Forecast, then: Error = Actual − Forecast.
If you choose Forecast − Actual, then: Error = Forecast − Actual.
Step 2: Compute RSFE (Running Sum of Forecast Errors).
RSFE = Σ(Error)
Step 3: Compute MAD (Mean Absolute Deviation).
MAD = Σ(|Error|) / n, where n is the number of valid periods.
Tracking Signal: TS = RSFE / MAD
Interpretation: values near zero indicate low bias. Many teams flag issues when |TS| exceeds about 4, but choose a threshold that fits your sales volatility.
How to use this calculator
- Pick an error definition that matches your reporting.
- Enter each period’s actual sales and forecast value.
- Click Calculate to see RSFE, MAD, and TS.
- If TS is outside your threshold, investigate bias.
- Export CSV or PDF to share with stakeholders.
FAQs
1) What does tracking signal measure?
It measures persistent forecast bias by comparing cumulative signed error to typical error size. A large positive or negative value indicates your forecasts consistently miss in one direction.
2) What is a “good” tracking signal value?
Closer to zero is better. Many teams treat values within ±4 as acceptable, but high-variance products may require a wider band and low-variance products may require a tighter band.
3) Why use MAD instead of standard deviation?
MAD is simple, robust, and easy to explain to sales teams. It scales errors into an interpretable average magnitude and works well when you want a quick bias signal.
4) What if MAD is zero?
If all valid periods have zero error, MAD becomes zero and TS is undefined. That usually means forecasts perfectly matched actuals or the dataset is too small to assess bias.
5) How many periods should I include?
Use enough periods to reflect your sales cycle. For weekly sales, 8–12 weeks is often a starting point. For monthly forecasting, 6–12 months can better capture seasonality.
6) Can I use revenue and units together?
Compute tracking signal separately for each metric. Mixing units and revenue in one table obscures meaning. Create one run for units and another for revenue to keep the interpretation clean.
7) Why does the sign change when I switch error definition?
The numerator uses signed errors, so reversing the subtraction flips the sign. The magnitude still reflects bias strength, but the direction represents opposite conventions.
8) When should I take action based on the result?
Act when the signal stays outside your threshold for multiple refreshes or when the bias aligns with known changes, like pricing, pipeline quality, channel mix, or seasonality shifts.