Most teams report a single accuracy number every month. The problem is that one number can look healthy while the plan is quietly failing on the items that matter most. Choosing the right metrics is the first step to a forecast people trust.
MAPE: easy to read, easy to mislead
The mean absolute percentage error averages the percentage error of each item or period. It is intuitive, which is why it is popular. But it breaks down when actual sales are zero or very small, it gives a slow-moving item the same weight as a best seller, and it penalises over-forecasts and under-forecasts differently.
WMAPE: weight the error by what matters
The weighted MAPE divides the total absolute error by total actual sales. Large items count more than small ones, and zero-sales periods no longer cause problems. For most portfolio-level reporting, WMAPE is a better headline number than MAPE.
Bias: the error that builds up
Accuracy tells you how big the errors are. Bias tells you their direction. A forecast that is consistently too high builds excess stock, and one that is consistently too low causes stock-outs, even if the average error looks acceptable. Track bias as the sum of forecast minus actual, divided by total actual, for every segment.
MASE: compare against a simple benchmark
The mean absolute scaled error, proposed by Hyndman and Koehler, compares your forecast error with the error of a naive forecast. A value below 1 means your method beats the naive approach. It works for intermittent demand and lets you compare items with very different volumes.
A practical scorecard
- Report WMAPE and bias together, never accuracy alone.
- Measure at the level where decisions are made, such as item by warehouse, and at the lead time that matters for ordering.
- Break results down by ABC-XYZ segment, because a good total can hide poor accuracy on A items.
- Use MASE to check whether a model adds value over a simple benchmark.
- Track forecast value added: did each manual override improve or worsen the result?
In OPTIAI, forecast accuracy and bias are tracked by segment and by planner, so the team can see which models and which overrides actually improve the plan.
See it on your own data
Book a demo and we'll show you how OPTIAI or OPTIFLOW applies these practices to your products, locations and supply chain.
Book a Demo