Data & drivers

Intermittent demand

Intermittent demand is a series that is zero most of the time and jumps when it is not — spare parts, slow SKUs, B2B orders. Average error metrics lie; you often care about hit rate and size when a hit happens.

Croston's method and its cousins split 'when' from 'how much.' A standard neural net on raw zeros will learn to predict zero and look clever on MAE.

Grain is a lever. Daily intermittent can become weekly almost-continuous. Sometimes that is the right product. Sometimes ops needs the day the part will move.

WAPE at this grain is often ugly. Compare to a slow-mover baseline, and report fill rate if that is the actual pain.