A stakeholder says users who use feature X retain better, so we should push everyone to use it. How do you respond?

That is a correlation, and the most likely explanation is selection: engaged users adopt more features, so the feature may be a symptom of retention rather than a cause. Say so without shutting the idea down, then propose a way to test it, ideally a randomized experiment, or failing that a matched cohort comparison or a difference in differences analysis around a release.

I would say it is a promising signal and then explain why I cannot conclude it yet. The most likely story is that engaged users try more features, so feature X adoption is a marker of engagement rather than the cause of it, and pushing an unengaged user into it may do nothing at all. There is also a plain reverse causation reading: people who were going to stay are the ones still around to discover the feature. The clean answer is a randomized test, so prompt a random subset into the feature and compare retention against a holdout, which tells us the causal effect on exactly the population we would target. If we cannot run that, I would look at a difference in differences around the feature's launch, or match users on prior activity, tenure, and plan and compare like with like. When I did the matched version of this on an onboarding feature, about two thirds of the apparent lift disappeared, which changed the roadmap decision.