When not to remove an outlier

Before choosing a model, define the demand being estimated and the action it informs. This guide places when not to remove an outlier in that end-to-end context.

Practitioner lens

For when not to remove an outlier, inspect the history for stockouts, promotions, calendar changes and missing periods. Observed sales can differ from the demand customers actually wanted.

First experiment

Record the forecast, override and outcome separately so the next review can distinguish model from judgment.

This is an original educational brief. It is a starting point for discussion, not a claim that one technique fits every supply chain.