How to measure demand forecast accuracy

The forecast accuracy formula (MAPE)

MAPE = Mean of (|Actual − Forecast| ÷ Actual) × 100 Forecast accuracy % = 100 − MAPE

Actual

units actually sold in the period

Forecast

units you forecast for that period

| |

absolute value: ignore whether the forecast was over or under

Worked example

Worked example: six months of forecasts for one productExample numbers

Month	Actual	Forecast	Error	Absolute % error
1	400	380	−20	5.0%
2	420	450	+30	7.1%
3	380	400	+20	5.3%
4	500	430	−70	14.0%
5	450	460	+10	2.2%
6	520	480	−40	7.7%

1. Add up the absolute percentage errors: 5.0 + 7.1 + 5.3 + 14.0 + 2.2 + 7.7 ≈ 41.3%
2. MAPE = 41.3% ÷ 6 ≈ 6.9%
3. Forecast accuracy = 100% − 6.9% = 93.1%

Month 4 is the outlier. A 14% miss is worth investigating: was there a promotion, a press mention or a competitor stockout that the forecast didn't include?

WAPE: a better measure for slow sellers

MAPE divides by actual sales, so it can't handle months with zero sales and exaggerates errors on small numbers. WAPE (weighted absolute percentage error) divides total error by total sales instead:

WAPE = Σ |Actual − Forecast| ÷ Σ Actual × 100

For the same six months: total absolute error = 20 + 30 + 20 + 70 + 10 + 40 = 190 units; total actual sales = 2,670 units. WAPE = 190 ÷ 2,670 = 7.1%.

How forecast error should size safety stock

Safety stock exists because forecasts are wrong. The bigger and more variable your errors, the more buffer you need. The service-level formula makes the link explicit:

Safety stock = Z × σ(daily demand) × √(Lead time in days)

In practice, you can use the variability of your forecast errors in place of demand variability: if your forecasts are usually close, the standard deviation is small and so is the safety stock. If accuracy improves, you can safely hold less stock.

Safety Stock CalculatorHow much buffer stock you need to cover demand and lead-time swings.Open calculator Want this handled automatically?Forecast demand and know exactly when to reorder, before you stock out or tie cash up in overstock.Try Verve AI

How to improve forecast accuracy

• Clean your sales history. Remove or adjust stockout periods and one-off events before forecasting from them.
• Add what you know about planned promotions, launches and price changes.
• Forecast at the right level. Forecast A products individually; group the long tail by category.
• Use the right method for each product. Moving averages for steady sellers, seasonal methods for seasonal ones. See demand planning vs supply planning.
• Track accuracy monthly and look for patterns, such as always over-forecasting new products.

Peak seasons need extra stock on top of everyday safety stock. The seasonal lead-time buffer calculator covers both higher demand and slower suppliers.