A KPI tree is not a dashboard. It is a contract about how the company defines success — and where the disagreements live. Three rules, four anti-patterns, one printable tree.
If you cannot defend why MRR sits where it sits — above retention, below growth, beside expansion — your KPI tree is decoration. The tree is the argument.
Rule 1: every node has an owner. Not a team. A person. The owner is the human a board member can ask 'why did this move' and get a non-evasive answer.
Rule 2: every node has a contract. Definition, denominator, vintage, freshness. Without a contract, the same tree on two dashboards drifts within a quarter.
Rule 3: every node names its anti-driver. Growth's anti-driver is contraction. Retention's anti-driver is voluntary churn. Without naming the anti-driver, the tree is a victory lap, not a model.
Anti-patterns we see most often. (1) The 'composite KPI' — something divided by something else divided by 'team mood'. Composite KPIs hide where the disagreement lives. (2) Two siblings that aren't actually mutually exclusive — most acquisition-channel trees have this. Fix: explicit overlap or rename to 'channel attribution buckets'. (3) Vintage drift — last-quarter's tree referenced a metric that has since been redefined. Fix: a versioned tree, with the redefinition annotated.
The printable tree. North-star → growth × retention × expansion. Growth = (acquired customers × ARPU first month). Retention = (1 − voluntary churn) × ARPU month-2 → ∞. Expansion = NRR − 1. We've been shipping variants of this for fintech, marketplace, gaming-analytics, and e-commerce. The shapes change. The discipline does not.