Forecast model selection
In Demand Planning we select forecast model based on nature of the product or based on the error after running the forecast with all the models?
Both, depending upon the circumstances.
Selecting a model based on a statistical error is merely using numeric algorithms to determine the best model of reality. In the end, it is still 'guessing' what the reality will be.
Your organization may already have a better handle on how the market for each material should behave, and how it should be forecasted. If this expertise does not exist, and if you have adequate history to use, then by all means use statistical error methods to select the most appropriate forecast model.
In some organizations, they use a mix. The marketing experts manually forecast certain important materials, or select a model manually for these materials, and then the rest are set to be forecasted using models selected which have the best statistical 'fit' to history.