A practical method for finding an early behaviour that predicts value without mistaking correlation for causation.
Begin with returned value
Write down the value a successful user receives and the value the business receives in return. Then identify the earliest observable evidence that the exchange has begun. For a budgeting app, opening the app three times is activity; connecting an account and reviewing categorised spending is closer to experienced value.
Test candidates, do not crown them
Compare candidate behaviours across acquisition cohorts and time periods. Look for a consistent relationship with downstream retention or revenue, while controlling for obvious selection effects. A strong candidate is frequent enough to guide teams, early enough to influence, and specific enough to design around.
Pair prediction with an experiment
Historical analysis can identify a promising marker, but it cannot prove that forcing the behaviour creates retention. Design an intervention that helps suitable users reach the milestone, then measure both the milestone and downstream value with guardrails for frustration or low-quality use.