MAEMean Absolute Error

MAE stands for Mean Absolute Error. It's the average size of your forecast errors, measured in actual units — how far off the forecast was on average, regardless of whether it was over or under.

A MAE of 50 means: on average, the forecast missed by 50 units each period. It doesn't tell you which direction (see forecast bias for that) — only how big the typical miss was.

The thing that makes MAE useful is that it's in units, not percentages. That keeps it concrete: an error you can hold up against a product's actual demand and reason about directly.

How it's calculated

For each period, you take the absolute difference between actual and forecast — the size of the miss, ignoring the sign — then average those across all the periods.

MAE = average of |actual − forecast|, in units

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

MAE = (10 + 10 + 30 + 5) / 4 = 13.75 units

So on this product, the forecast was off by about 14 units a week, on average — a number you can act on directly.

Why MAE — error you can reason about

Because MAE is in units, it converts straight into consequences. A forecast that's 100 units off on a £5 product is £500 of error per cycle. You can see what the inaccuracy is costing you and what to do about it — adjust the reorder by 100 units, or carry a bit more cover on that line.

A percentage is abstract by comparison. "12% MAPE" doesn't tell you whether that's £50 or £50,000 of exposure — it depends entirely on the volume and value of the product underneath it. And a tidy headline percentage can mask big problems on exactly the products that matter most, because a single high-volume line buried in an average looks fine until you read it in units.

MAE keeps the conversation grounded in what an error actually means for the business: this many units, this much cash, this reorder.

MAE vs MAPE

The two metrics answer the same question — how accurate is the forecast — but in different currencies, and each breaks where the other holds up. Cross-reference MAPE for the percentage view.

MAPE expresses error as a percentage of actuals. That makes it handy for communicating across products of different sizes — "we're 88% accurate" travels well across a portfolio. But it breaks on zero-sales periods, where dividing by zero leaves it undefined, and it inflates badly on low-volume weeks, where a small absolute miss becomes a huge percentage.

MAE stays robust at low volumes and for intermittent demand — there's no division by actuals to blow up. The trade-off is that it isn't directly comparable across products of different scale. An MAE of 50 means something very different on a product selling 100 a week than on one selling 10,000 a week: a disaster in the first case, a rounding error in the second.

So neither is "better" — they're suited to different jobs.

When to use which

A simple division of labour:

• MAE for reasoning about individual products, comparing forecasting models against each other on the same product, and anything with low or intermittent volume where MAPE falls apart.
• MAPE for communicating accuracy across a range to the wider business, where a percentage is more intuitive than a pile of unit figures.

In practice most planning teams use both: MAE to do the thinking and choose the model, MAPE to report the headline.