randomized controlled experiments are the gold standard for establishing causality 👉 "compared to control, metrics in treatment moved in this directions because of this change we made".
that said, it's not always possible, feasible, or ethical to randomly assign units into treatment and control. in those cases, quasi-experimental designs ("quasi" means "as if" or "almost" in latin) can help data scientists get closer to causality. see this article for an awesome summary.