Has anyone used DoE or ML to cut process-development time — real results?

Lots of talk about ML-guided media/feed optimisation and digital twins. Has anyone seen a concrete reduction in DoE runs or time-to-lock a process — what did you use and what was the honest payoff?