An install is a delivery event, not proof that a channel created value. It is useful for checking acquisition volume, but poor at answering whether the people acquired reached the moments that matter to the business.
Choose a quality event first
Start with an event that indicates genuine progress: completing onboarding, making a first booking, funding an account or returning to use a core feature. It should happen often enough to compare cohorts and late enough to mean something.
Compare cohorts on the same clock
A user acquired yesterday has had less opportunity to retain than one acquired last month. Compare channel cohorts over consistent windows—day 1 activation, day 7 retention or day 30 revenue—before drawing conclusions.
Keep cost and uncertainty visible
Pair quality rates with spend and sample size. A channel with strong activation among twelve users is promising evidence, not a settled winner. Decisions improve when the limits travel with the result.