Interpret your first KPIs
Read your first accuracy indicators —BIAS, MAE, MAPE, RMSE, Accuracy— and understand what they tell you.
By the end of this tutorial you will know how to read a SKU's KPI panel, identify whether your forecast has a systematic bias, and recognize whether the errors are acceptable or require action.
For the mathematical definitions of each indicator, see the Forecast accuracy metrics reference.
What you will need
- At least one closed month (closed version) with actual demand recorded; without that, the KPIs have nothing to compare against.
- Access to the detail panel of any SKU in the Forecast module.
Step 1 — Open the KPI panel for a SKU
- In the Forecast module, click the name of a SKU in the grid. Its detail panel opens.
- Locate the KPIs or Accuracy metrics section. You will see summary cards with cumulative averages and, below them, the period-by-period detail.
TODO: screenshot of a SKU KPI panel with visible cards
Step 2 — Read BIAS first
BIAS measures the direction of systematic error:
| BIAS value | Meaning |
|---|---|
| Close to 0 | Your forecast has no appreciable bias. Ideal. |
| Positive BIAS | You are systematically overestimating (forecast > actual demand). |
| Negative BIAS | You are systematically underestimating (forecast < actual demand). |
A large error with no consistent direction (BIAS ≈ 0) may be market noise. A persistent BIAS in the same direction signals a structural problem: the model or the team is systematically off in one direction. Fix the bias before trying to reduce error magnitude.
Example: if a SKU's cumulative BIAS is −120 units/month, you are ordering too much every month. Check whether there is a demand driver the model is not capturing (unmodeled seasonality, product end-of-life, etc.).
Step 3 — Read error magnitude: MAE and MAPE
Once you know the direction (BIAS), evaluate the size of the error:
-
MAE (Mean Absolute Error): average error in the same units as the SKU. Useful for comparing SKUs in the same product or category. An MAE of 50 units is very different if the SKU sells 60 or 6,000 per month.
-
MAPE (Mean Absolute Percentage Error): error as a percentage. Allows comparison of SKUs at different scales.
| MAPE | Interpretation |
|---|---|
| < 5 % | Excellent. The model is very accurate. |
| 5 % – 10 % | Good. Acceptable for most categories. |
| 10 % – 20 % | Improvable. Review outliers, seasonality or uncaptured promotions. |
| > 20 % | Problematic. Action required: review the model, the data or the adjustment methodology. |
When actual demand is very low (1 or 2 units), a small absolute error becomes a huge MAPE. In those cases, give more weight to the absolute MAE and BIAS than to MAPE.
Step 4 — Read Accuracy
Accuracy is the complement of MAPE:
Accuracy = 100 % − MAPE
The recommended minimum target is 80 % (equivalent to MAPE ≤ 20 %). High-rotation SKUs without seasonal behavior typically reach 90 %–95 %.
The summary cards in the panel show the cumulative average Accuracy across all closed periods. Use it as a general health indicator for that SKU's forecast.
Step 5 — Read RMSE as a large-error signal
RMSE (Root Mean Square Error) penalizes large errors more than MAE because it squares each error before averaging.
- If RMSE is much larger than MAE, one or more periods had very large errors that are distorting the average (possible outliers or unmodeled events).
- If RMSE is similar to MAE, errors are fairly consistent: there is no catastrophic month hiding behind the average.
Always compare MAE and RMSE. An RMSE/MAE ratio close to 1 means consistent errors. A ratio above 1.5 or 2 calls for investigating the periods with the highest error.
Step 6 — Read the KPIs together
The correct diagnosis uses all five indicators as a system:
- Is there bias? → BIAS. If significant, fix it first.
- How large is the typical error? → MAE (in units) and MAPE (as a %).
- Am I above the acceptable threshold? → Accuracy ≥ 80 %.
- Are there catastrophic error periods? → Compare RMSE vs MAE.
A good forecast has BIAS ≈ 0, MAPE < 10 % and RMSE not much higher than MAE.
Result
You know how to read a SKU's accuracy KPI panel, identify systematic bias and assess whether the error magnitude is acceptable.
Next step: take this analysis into inventory optimization in Your first inventory simulation.