This page explains how to judge the library’s recommendations: what a comparison can show, when a pattern is strong enough to publish, and where each tactic still needs product-specific measurement.
The hardest part of growth evidence is not finding a lift. It is knowing whether the lift came from the play, from timing, or from the kind of subscriber who saw it. A subscriber who accepts an annual offer, opens the app during a payment failure, or returns after cancelling may already be different from the average subscriber.
When the data allows it, we compare subscribers against close matches: subscribers with similar plan value, tenure, engagement depth, and lifecycle state who did not receive the same play. The closer the comparison, the more confidence we have that the remaining difference belongs to the tactic.
Where a close comparison is not possible, the copy says so. Some tactics are still useful because the pattern repeats across many companies, but the evidence note should make clear whether you are reading an isolated effect, a repeated operational result, or a directional signal.
A finding does not become a tactic because it is interesting. It becomes a tactic when it clears a practical bar:
Scale. The pattern is broad enough to survive beyond a single deployment, cohort, or quarter.
Comparison. The evidence accounts for obvious subscriber differences before turning the result into a recommendation.
Consistency. The direction holds across enough contexts that it is not just one product’s quirk.
Durability. Predictive signals are checked against periods they were not tuned on, so the tactic is not just fitted to the past.
Actionability. The finding converts into something a team can deploy: a concrete trigger, a channel, and guardrails. A pattern you cannot act on stays in the research pipeline.
Plenty of patterns stay out of the library because they are too narrow, too hard to act on, or not stable enough across companies. The published set is intentionally smaller than the research backlog.
Evidence should stay attached to the claim it supports. When a number is broad enough to publish, the tactic shows the number. When a precise number would imply more certainty than the data supports, the tactic describes the direction and the conditions instead.
The same rule applies to implementation detail. The library gives each tactic a trigger, surface, guardrails, and measurement notes, but it avoids over-prescribing a workflow that should be adapted to the product, pricing model, and customer lifecycle in front of you.
Browse the register →