At the NY R conference today, the Meta team presented their MMM package for R:
They have built a feature where you calibrate the MMM model selection using plug-in estimates from experiments or causal inference.
Seems like an interesting approach, wasn’t enough time in the 15m talk to explain how it worked.