can engineers please go take a proper stats and design of experiments class. The first anecdote he used about the push-notification is garbage. like pure garbage. If you didn't design the experiment upfront and specify what you were going to measure, doing some SQL'ing after the fact doesn't make it valid.
There are quasi-experimental techniques you can use to infer causality, but most are still inferior to a designed experiment.<p>The problem with post-hoc approach he used is that you can after the fact, always cherry-pick stats (after trying multiple different paths) and eventually find one that shows you it was a success, which is what this so called "growth engineer" did. How about this for a novel idea, engineers stick to engineering but if you want to cosplay as an analyst/data scientist, go learn some statistics and design of experiments.