Intervention
An intervention is a deliberate change a team ships to move one specific user behavior: a reworded prompt, a reordered signup step, a landing page that matches the ad that brought someone in, an email nudging a lapsed user back.
Definition
Intervention
A change a team ships to an offering — copy, sequencing, a new affordance, a deepened path — to move more users across a behavioral proxy and lift its connected commercial outcome.
What an intervention can be
An intervention has no canonical form. It can be a one-line copy edit, a reordered onboarding step, a lifecycle email sequence, a painted-door CTA, or a fully guided flow. What governs the choice is the cost-feedback ratio in service of the specific thing the team is trying to learn — how cheaply the intervention ships against how fast and how clearly it returns evidence — not membership in a standard playbook. Two interventions aimed at the same threshold can differ by an order of magnitude in cost, and the cheaper one wins until evidence justifies more.
Testing the candidate
Shipping an intervention does two things at once. It attempts the lift, and it tests whether the targeted behavior is a lever the team can pull or a signal of intent that was already there. A candidate behavior surfaced by multiplier analysis is only an observed correlation — users who did it converted more often. Whether amplifying the behavior produces conversion, or merely tracks users who would have converted anyway, stays unknown until an intervention designed to amplify it either moves the connected outcome or doesn't.
A behavior earns value-milestone status by surviving this test across repeated interventions, not on one strong reading. When amplification produces no lift, the behavior was a signal, and continuing to engineer it is proxy failure — optimizing a downstream marker of success rather than a cause of it.
Cheapest first, deepen on signal
Because the result is uncertain until the intervention runs, the discipline is to start with the cheapest, most reversible version available and deepen only on evidence — the path-investment principle applied to experiments. A copy change that produces intent-matched lift earns a tailored email sequence; the sequence that holds up earns a full guided experience. Lift that holds is the license to invest further. Lift that fails to appear sends the team back a step — either the segment the intervention assumed isn't there, or what the intervention varied isn't what that segment responds to — to find a different cheap test before deepening.
Example
A team's multiplier analysis on a signup cohort shows that users who import a contact list reach first paid conversion at four times the baseline rate. That makes importing a candidate behavior, not a lever — the correlation could simply mean that users already committed to paying are the ones who bother to import. To test it, the team ships the cheapest intervention available: a one-line prompt during onboarding suggesting the import. If the prompted cohort's conversion lifts and the lift holds when the analysis is rerun on fresh cohorts, importing is behaving like a lever, and the team deepens — a guided importer, a pre-filled template. If conversion doesn't move, importing signaled intent that already existed; the team can't manufacture conversions by manufacturing imports, and the candidate gets demoted.