Amazing post. Yes please.
I worked at a BigCo deeply diving into metrics (I was responsible, from the engineering side, for the metrics of a from-scratch rewrite of a product with a lot of users, so huge metrics implications), It reinforced something I already knew: the single best way to improve your software engineering team’s relationship to data is to ban deriving insights from data. If you didn’t ask the question (or form the hypothesis) before you saw the data, and you didn’t measure data through an experiment, your insights are almost certainly horseshit and you don’t have the answers to any questions.
As per the first box in the article’s flow chart, it is possible to gain insights from well understood data, but programmers think they understand all data and they don’t spend one second trying to disprove their assumptions.
If you want to take my cynicism to its logical extreme: since programmers are incapable of good data science, you should just ban data (instead of favoring good arguments) and operate on intuition and chaos instead. You’ll be better off than most software organizations.