the problem is 1) sample sizes are very low for early startups 2) feedback loops (esp for higher ACV) are long, sometimes >1 year.

you still need principles to drive at the end of the day