1. Clarify the question by drawing the causal diagram (controlled and uncontrolled factors, points of leverage)
2. To understand the effect of [thing] on [metric], remove [thing] and recalculate (Customer groups driving metric, features in ML model)

3. ML/Regression is all about E[y

   X]; causal inference is a clever usage of this, and so is learning to rank, and so is…other stuff? In general look for options to use E[y

   X]

4. Never ever forget the decision at hand, return to it when you are stuck
5. do the shoe leather work, often that’s what makes the difference

links to: Peng, Shalizi, Gelman