Supply Chain

How Do You Calculate Forecast Accuracy? MAPE, WAPE, and the Trap

MAPE is the standard forecast accuracy metric and it breaks precisely where most manufacturers live: a portfolio mixing high runners with slow-moving parts.

Mean absolute percentage error is the default forecast accuracy metric: for each item, take the absolute difference between forecast and actual, divide by actual, then average those percentages. It is intuitive, it is widely understood, and on a portfolio with mixed volumes it produces answers that are actively misleading. The failure is structural and worth understanding before you report a number built on it.

Where MAPE breaks, worked

Two SKUs. The first has actual demand of 1,000 and a forecast of 1,100, an absolute percentage error of 10%. The second has actual demand of 2 and a forecast of 4, an error of two units and an absolute percentage error of 100%. Averaged, MAPE is 55%, which suggests a forecasting catastrophe. Weighted by volume, WAPE is 10.2%. The business missed by 102 units on 1,002, and only one of these two metrics conveys that.

• MAPE across the two SKUs: 55%
• WAPE, volume-weighted: 10.2%
• Actual miss on 1,002 units of demand: 102 units

The three failure modes of MAPE

Any one of these should send you to a different metric

• It is undefined when actual demand is zero, which is common for intermittent parts and forces an arbitrary exclusion rule that itself biases the result.
• It penalizes over-forecasting more heavily than under-forecasting, because the denominator is actual, so it quietly encourages a biased forecast.
• It weights every SKU equally regardless of value or volume, so your slow movers dominate a metric that ought to be about business impact.

What to use instead

WAPE, sometimes called MAD over mean, sums absolute errors and divides by total actual demand. It is volume-weighted by construction, it handles zeros without special cases, and it answers the question a planner actually has: what share of total demand did we get wrong? For intermittent demand specifically, mean absolute scaled error compares your forecast against a naive baseline, which is the honest test for slow movers where any percentage metric struggles.

A metric that lets two units of error on a slow mover outweigh a hundred on a high runner is not measuring forecast quality. It is measuring your product mix.

Accuracy and bias are different questions

Every metric here uses absolute values and is therefore blind to direction. A forecast can post excellent WAPE while being consistently biased in one direction, and the bias is usually the more expensive problem. Report both, always. Accuracy tells you how tight the forecast is; bias tells you whether it is centred. A process that monitors only the first will run a systematically skewed forecast for years without noticing.

Use the forecast accuracy calculator to compare MAPE and WAPE on your own portfolio. Calculate accuracy properly

Published 2026-08-08.