Activation experiments

What experimentation on activation is

Activation is the step where a new account reaches the thing it signed up for. In the pirate funnel it sits under acquisition, and it is usually written as new users who complete a specific action in their first week, engage with the core feature, or finish onboarding; a good activation metric captures the moment a user gets value from the product [1]. Onboarding experiments change the path to that moment. Activation experiments change what waits at the end of it: what the empty state holds, what the first screen offers to do, what is already filled in, and what the product does on its own for an account that has arrived and made nothing yet.

The metrics

The primary metric is the activation event itself: the share of new accounts that reach first value in a window of days. Which event that is carries the whole stage, and it is checked rather than picked, because the guidance is to confirm that the chosen activation metric leads to increased retention before trusting it [1]. Secondary metrics sit closer to the change: time to first value, the share reaching an intermediate step, the share that opened what was offered. They explain a result. They should not decide it.

The bottlenecks

An empty state that shows what the product could hold rather than what it does. Empty states are common during onboarding and initial use of an application that is not configured yet, and the guidance for one is to communicate system status, provide learning cues, and give a direct pathway to the key task [2]. A vague call to action repeats the failure inside a button: the phrase "Get started" is ambiguous and can apply to almost any goal a user might have, so it collects clicks without saying what happens next [3]. After those come a first action that needs data the user has not brought, and a product whose value needs other people in it before one person can see anything.

The guardrails

Retention past the activation window, because a shorter route to the event can produce accounts that reach it and never come back, and the metric keeps its place only while it still predicts retention [1]. Support volume, because a first session that does more on the user's behalf can leave them unsure what is theirs. The quality of what gets made, whenever the change prefills or suggests it. And the step a shortcut skipped, read later, because work that never happens is a leak moved rather than closed.

When a test here is feasible

An activation test counts new accounts, not visits, so its population is a fraction of a signup test's and the same relative effect takes proportionally longer to resolve. Whether it can resolve at all depends on the base rate of the chosen activation event, which the product knows and no benchmark supplies. Size it before it starts, with the calculator and the sample size guide; the feasibility guide and the feasibility checker say when not to run it, and the guardrail guide says what to hold while it runs.

The patterns

Sample data in the empty state instead of a blank workspace is the example written out in full below. A first action prefilled from what signup already collected, an import that produces something visible in one step, and a prompt in the first session that names the one thing worth doing are the same instinct on the stage's other surfaces, and they are written when they are.