How accurate should a demand forecast be?

Accuracy targets depend on what you are forecasting. Aggregate forecasts at category level are routinely within 10%, while individual SKU-week forecasts on a long tail can be far worse and still be useful. Judge a forecast by whether its errors would have changed a buying decision, not by the percentage alone.

Measure bias separately from error. A forecast that is wrong by 15% in both directions is noisy; one wrong by 15% consistently high is systematically overbuying, and only the second has an obvious fix.

More on Demand Planning, or browse every commerce operations answer.