Journal
Metric trees for subscription apps
A tree is a decision aid, not a map of every event in Segment.
Subscription apps collect events the way kitchens collect utensils. The taxonomy grows. Someone then draws a “north star” that is really three north stars taped together: trial start, paid conversion, and weekly active. An experiment that moves one will be sold as moving all three.
A usable metric tree for A/B Test Analytics for Apps has a single primary at the top for a given test. Under it, two or three diagnostic branches that explain movement, and a guardrail branch that is allowed to veto. For a typical consumer subscription that means: primary = first paid conversion in a 14-day window; diagnostics = trial start and paywall view; guardrail = refunds plus crash-free sessions. Weekly active can live on a different tree for a different test.
The mistake we see in GB product orgs is duplicating the primary as a diagnostic. “Trial start” and “paywall conversion” are not independent if the copy change sits on the paywall. You will spend the meeting arguing about which number is the real one. Name the conflict in the plan: this test is allowed to hurt trial start if paid conversion holds, or it is not. Write the sentence.
Prune nodes that exist because an analyst once built a dashboard tile. If nobody can say what decision the node supports, it is decoration. Decoration is how vanity lift sneaks back in.
We practise this drawing in the first module of Readout Craft because later modules are wasted if the tree is a wish list. Bring a printed tree to the first seminar. Cross things out in pen. It photographs well and it is more honest than a slide.