A KPI tree should explain how a business outcome connects to actions a team can influence. A dashboard with many metrics but no causal hypothesis is a collection, not a decision tool.
Draw the path with explicit assumptions
Start at one outcome: delivered contribution, completed appointments, retained customers, or another business-defined result. Break it into the measurable steps that precede it. For a service inquiry, a possible path is:
Completed service = qualified requests × booking rate × attendance rate × completion rate.
This identity is useful only if each term has a stable definition and the data can be joined without double-counting. Add acquisition measures upstream—relevant sessions or inquiry cost—but do not add unrelated metrics to make the tree look complete.
For each node, record: definition, source, owner, reporting window, segment, and what decision changes if it moves. Label causal assumptions. A higher click-through rate may change inquiry volume, but it does not necessarily improve qualified demand or contribution.
Find the controllable branch
If qualified requests are steady but attendance falls, inspect scheduling, reminders, and service coverage before changing ads. If page visits rise but valid requests do not, inspect the offer, audience, and form. If only one market shifts, separate its path rather than averaging it into a regional total.
Choose one leading indicator per active decision and retain the final outcome as a guardrail. Review the tree when event definitions or sales processes change; keep version notes so historical comparisons remain interpretable.
Optional AI-assisted work brief — Outcome-to-metric tree
Start with one business outcome and decompose it into controllable stages, observable events, data source, owner, and quality caveat. Label leading indicators separately from realized outcomes. Identify where two metrics can disagree and the reconciliation source. Remove metrics that have no owner or decision attached. Do not imply causal contribution from a correlation alone.
Worked scenario — Separate a leading signal from the outcome
A useful tree might connect campaign spend → qualified landing visits → requests received → requests eligible → appointments attended. Each arrow is a relationship to investigate, not an automatic causal proof. Name the system and owner for every stage, then reconcile sample records. If attendance data is missing, report requests as requests; do not rename them revenue.
AI disclosure: AI tools assisted with drafting. The equation is an illustrative model, not a business result.