At the end of a season, the remaining products may suggest where the assortment did not meet the demand. However, the data must be interpreted: a late delivery, an unbalanced size or an unclear presentation can produce results other than an unsuitable model choice.
I propose to reconstruct for the relevant categories initial quantities, availability period and sales under different conditions. The final total can hide very different margins between products sold immediately and products released only after price reductions.
A hypothetical example is a model that sells well in two colors and stays in a third. Indiscriminately reducing the next order could penalize even the required variants. The level at which the problem occurred must be recognised.
The review should produce few usable decisions: quantities to be adapted, information to be collected or supplies to be reviewed. Forecast errors should also be read with the information available at the time of purchase. The end result does not prove that a choice was unreasonable when it was made. Learning is about improving future assumptions while keeping the conditions that influenced the outcome of the season visible.