Predict-then-optimize versus joint learning

Forecast uncertainty needs to be measured and carried into downstream decisions. This technical note frames predict-then-optimize versus joint learning as part of that full planning chain.

Practitioner lens

A rigorous evaluation of predict-then-optimize versus joint learning reports loss by horizon and segment, then checks whether uncertainty estimates achieve their intended coverage on unseen periods.

First experiment

Report uncertainty, compute cost and maintenance burden alongside point accuracy.

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