create user cohorts by levels of engagement, and analyze what differentiates high value users from lower value ones. These start out as correlations, so use A/B testing to prove causality—once the best levers are found, test many variations of these ideas.
