Metric Tree
A metric tree maps a high-level product or business outcome into component measures and hypothesized drivers that teams can diagnose and influence.
Download Markdown (.md)Soroush’s Product Wiki · https://wiki.soroush.uk/topic/metric-tree/
01What it means
A metric tree is a hierarchical model connecting a top-level outcome to lower-level metrics. Some branches are mathematical components, while others represent hypothesized influence. The tree makes definitions, relationships, and ownership visible so teams can move from a changing headline number to testable questions.[1][2]
Label mathematical components and hypothesized influences differently.
02Why it matters
Dashboards often present unrelated numbers without explaining how they fit together. A metric tree gives teams a shared model of product performance, links local work to broader outcomes, and highlights where diagnostic evidence or instrumentation is missing. It can also expose double counting and incompatible definitions.
03When to use it
Use a metric tree when choosing KPIs, diagnosing movement in a north-star or financial outcome, aligning teams, or planning analytics. Start small around one decision domain. Distinguish identities from assumptions, because “revenue equals customers times revenue per customer” is different from “invites may improve retention.”
04How it works
- Choose a clearly defined top-level outcome with a population and time window.
- Decompose mathematical components, then add behavioral or operational drivers as explicitly labeled hypotheses.
- Define every node, owner, data source, refresh cadence, and permitted segmentation.
- Check for overlap, missing branches, circular reasoning, and measures the team cannot influence.
- Use data, experiments, and research to test relationships, then update the tree as the product model changes.
05A practical example
A learning product places weekly successful learners at the top. It decomposes this into eligible learners multiplied by success rate, then maps possible drivers such as first-session setup, practice frequency, content availability, and assessment reliability. The team investigates a falling branch before proposing features.
06Where it breaks
A tree can imply causality that has not been established and may oversimplify feedback loops or network effects. Large trees become decorative and hard to govern. Mathematical decomposition identifies where change occurred, but not necessarily why or what intervention will improve it.
Common mistakes
- Mixing arithmetic components and causal hypotheses without labeling them.
- Building a comprehensive diagram that no recurring decision uses.
- Leaving nodes without stable definitions, owners, or source data.
07Sources
- What Is a Metric Tree? The Complete Guide with ExamplesMixpanel · Product analytics practice guideWhat Is a Metric Tree? The Complete Guide with Examples
- Metric Trees for Product AnalyticsMixpanel · Product analytics framework overviewMetric Trees for Product Analytics