What is a good forecast accuracy? Why the headline number lies, what wMAPE really tells you, and the targets that actually move cash and service.
Ask five planners what counts as a good forecast and you will get five numbers. That is because accuracy only means something next to its context: which items, measured how, over what period, against what. A single percentage with none of that attached is a comfort blanket, not a measure.
A portfolio can post a respectable overall error while being badly wrong on the items that drive revenue and cash. Averages hide the misses, because the easy, high-volume items pull the number down and flatter the whole book. You do not run a business on the average item; you run it on the specific ones, which is why accuracy has to be read SKU by SKU.
wMAPE, weighted mean absolute percentage error, weights each item's error by its volume, so the products that matter most count most. It is a far more honest headline than a plain average because it stops a swarm of tiny, easy items from hiding big misses on the ones that count. Pair it with the per-SKU distribution and you finally see the truth.
Good varies by business, but for many mixed portfolios a weighted error in the low single digits overall, with a sensible median per SKU, is strong. The point is not to chase a vanity number; it is that even a 10 to 20% improvement in accuracy can cut inventory cost by several percent and shrink the safety stock you carry. Accuracy is not a scoreboard; it is cash and service.
There is no single universal number. Good accuracy depends on the demand pattern, the forecast horizon, the level of aggregation and the metric used. A low aggregate error can still hide poor SKU-level forecasts, so accuracy should be read as a distribution, not one headline figure.
Weighted mean absolute percentage error weights each item's error by its volume, so high-volume items count more and low-volume items do not distort the average. It is more robust than plain MAPE for mixed portfolios, but it still needs to be read alongside the SKU-level spread.
Because an aggregate figure averages over items. A portfolio can post a strong overall number while a handful of important SKUs are badly forecast, and those are exactly where the stockouts and write-downs happen. The distribution and the bias matter more than the headline.
Not automatically. The link depends on demand variability, lead times and how safety stock is set. Accuracy gains help most when buffers are tuned to the improved error; without that, they do not translate into cash. The relationship is conditional, not guaranteed.
A forecast number without its context is not a measure. It is a mood.
Picture your next S&OP meeting: one number on screen that nobody argues with. See it free, on your own data, in 45 minutes; and if it does not hold up, we part as friends.