MAPEMean Absolute Percentage Error

MAPE stands for Mean Absolute Percentage Error. It's the most widely-used forecast accuracy metric — averaging the percentage error of each individual forecast, ignoring whether the error was positive or negative.

A MAPE of 20% means: on average, the forecast was off by 20% of actual demand, in either direction. Lower MAPE = more accurate forecast.

It's the metric most planning conversations default to because it's intuitive ("we're 80% accurate" = MAPE of 20%) and easy to compare across SKUs.

The MAPE formula

MAPE = (Σ |Forecast − Actual| / Actual) / n × 100

Worked example. Forecast vs actual for one SKU over four weeks:

MAPE = (11.1 + 7.7 + 37.5 + 4.8) / 4 = 15.3%

That tells you the forecast for this SKU was, on average, 15.3% off the actual demand week to week.

What's a "good" MAPE

There isn't a universal benchmark. It depends on the SKU's demand variability and the planning horizon.

Rules of thumb that travel:

• Stable, high-volume SKUs (CV < 0.25): MAPE under 10% is achievable
• Normal consumer SKUs (CV 0.25–0.5): MAPE 15–25% is realistic
• High-variability SKUs (CV > 0.5): MAPE 30–50% is normal — consider whether forecasting is even the right tool for these
• New launches: MAPE > 50% in the first 3 months is the rule, not the exception

A MAPE that's worse than just predicting the prior period's demand (the "naïve forecast") means your forecast model is destroying value. That's the floor.

Where MAPE breaks

MAPE has known weaknesses:

• Asymmetric. It penalises over-forecasting more than under-forecasting on a percentage basis. Forecasting 200 when actual is 100 is a 100% error. Forecasting 0 when actual is 100 is a 100% error. Forecasting 100 when actual is 0 is infinite error (division by zero).
• Breaks at low or zero demand. Any week with actual demand of zero or near-zero blows up the MAPE for that SKU. SKUs with intermittent demand (B2B, niche items) need a different metric.
• Doesn't tell you direction. A symmetric MAPE of 20% can hide chronic over- or under-forecasting — see forecast bias for the directional read.
• Hides catastrophic individual errors. A single 200% error week can be averaged away by three good weeks. Look at the distribution of errors, not just the mean.

MAPE vs other accuracy metrics

Common alternatives, when MAPE doesn't fit:

• WAPE (Weighted Absolute Percentage Error) — weights errors by volume, so big SKUs count more. Better for portfolio-level reporting.
• MAE (Mean Absolute Error) — absolute units, not percentage. Useful when SKUs have similar volume, and for low or intermittent demand where MAPE breaks down.
• RMSE (Root Mean Squared Error) — penalises big errors disproportionately. Useful when occasional huge misses are costly.
• sMAPE (symmetric MAPE) — fixes the asymmetry problem above. Less intuitive but mathematically cleaner.
• MASE (Mean Absolute Scaled Error) — compares against the naïve forecast. Best for academic comparisons.

For most operational planning, MAPE at the SKU level + WAPE at the category level is enough.

Using MAPE in practice

The MAPE number itself is less interesting than what you do with it.

• Trend over time. Is your overall MAPE improving quarter-over-quarter? If not, the planning process isn't learning.
• Distribution across SKUs. Are the bad SKUs concentrated (one or two terrible) or spread out? Concentrated = fixable; spread out = systemic.
• Bucketed by category. New launches will always be high; mature core lines should be much lower. Holding them to the same MAPE target is unfair.
• Action threshold. Define a MAPE level above which the SKU goes on a watchlist for forecast review. Below that, leave it alone.