# Build Metric Tree > Decompose a top-line metric into a quantified metric tree with mathematical relationships, size each node, and identify where the real leverage is. Use when a PM needs to understand what drives a metric, where to focus, or where NOT to focus. - Skill: `amplitude/build-metric-tree` (Agent Skill) - Install (CLI): `npx skillmds@latest add amplitude/build-metric-tree` - Raw SKILL.md: https://api.skillmd.com/api/skills/amplitude/build-metric-tree/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Marketing & Growth - Author: amplitude (https://skillmd.com/u/amplitude) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/amplitude/build-metric-tree --- # Build a Metric Tree **Decompose a top-line metric into a tree of sub-metrics with real math, size every node, and identify where the leverage actually is — and where it isn't.** You want to move a number but you're not sure where to focus. A metric tree forces you to break it down into its component parts, understand the mathematical relationships between them, and confront where the actual volume is. The output isn't just a diagram — it's a decision about where to invest and what to ignore. --- ## Prompt Template ``` You are a growth-minded product strategist who is rigorous about quantitative decomposition. You are allergic to wishful thinking and you always follow the volume. Here is what I'm working with: $ARGUMENTS > If the above is blank, ask the user: "{{DESCRIBE YOUR TOP-LINE METRIC, YOUR PRODUCT, AND ANY DATA YOU HAVE — e.g. current values, segments, known breakdowns, or a goal you're trying to hit}}" Help me build a metric tree. Follow these steps precisely: ### Step 1: Establish the Metric Hierarchy Start by identifying three complementary metrics that should be in tension: - **North Star Metric (NSM):** The metric that captures aggregate value delivered to customers. This is not revenue — it's the thing that, if it goes up, means customers are getting more value (e.g. "weekly active analysts," "dashboards created," "queries run"). - **Top Business KPI:** The metric that captures value back to the business (e.g. revenue, paid seats, expansion ARR). This must be structurally connected to the NSM — if NSM goes up but the business KPI doesn't, there's a monetization problem. - **One Metric That Matters (OMTM):** The single tactical metric the team should obsess over right now. This will emerge from the tree — it's the node with the most leverage. If the user has already defined these, confirm them. If not, propose candidates and explain the tradeoffs. ### Step 2: Decompose into a Tree Break the NSM (or whatever top-line metric the user specifies) into its component parts. For each branch: - **State the mathematical relationship explicitly.** Use addition (+) for components that sum to the parent (e.g. MAU = New Users + Returning Users + Reactivated Users). Use multiplication (x) for rate-based relationships (e.g. New Users = App Store Views x Conversion Rate). - **Go 2-3 levels deep.** Stop when you reach metrics that a single team could own and directly influence. - **Mark hypothesis branches.** If a relationship is assumed but not proven (e.g. "we believe retention is driven by users who complete onboarding in the first session"), flag it as a hypothesis that needs validation. Distinguish these clearly from definitional decompositions. Present the tree in a structured text format like: ``` NSM: Weekly Active Users (WAU) = 50,000 ├── [+] New Users: 8,000 │ ├── [x] Signups: 20,000 │ │ ├── [+] Organic: 14,000 │ │ └── [+] Paid: 6,000 │ └── [x] Activation Rate: 40% ├── [+] Returning Users: 38,000 │ └── [x] Prior WAU x Retention Rate: 45,000 x 84% └── [+] Reactivated Users: 4,000 └── [x] Dormant Pool x Reactivation Rate: 80,000 x 5% ``` ### Step 3: Size Every Node For each node in the tree, include the current value (or best estimate). This is non-negotiable — an unsized tree is just a diagram. If the user doesn't have exact numbers, work with them to estimate using whatever data is available. The sizing reveals the leverage. A 50% improvement on a small node might be less impactful than a 5% improvement on a massive one. Do the math explicitly. ### Step 4: Validate Metric Quality For each metric in the tree, quickly assess against these criteria. Flag any that fail: - **Tangible:** Does it represent a real, understandable goal? - **Controllable:** Can the team actually influence it? - **Fast feedback:** Does it move quickly enough to learn from? - **Testable:** Can you A/B test changes to it? - **Explainable:** When it moves, can you understand why? - **Hard to game:** Could the team hit the number through perverse means? If so, define a counter-metric. - **Prefer absolute counts over averages/percentages.** Averages hide volume. "Average session length increased 20%" means nothing if total sessions dropped 50%. Always pair rates with their denominators. ### Step 5: Identify the Lever (and the Anti-Levers) This is the most important step. Based on the sized tree: 1. **Where is the leverage?** Which node, if improved, would produce the largest absolute impact on the top-line metric? Show the math. "Improving activation rate from 40% to 50% on 20,000 signups = 2,000 additional WAU." 2. **Where is it NOT?** Explicitly call out nodes that are tempting but low-leverage. Be direct: "Reactivation looks appealing but the dormant pool is small — even doubling the reactivation rate only adds 4,000 WAU." This is the "don't be wishful" check. 3. **Where are the eyeballs?** For product-led growth, you have to go where users already are. Don't build strategies around surfaces nobody visits or flows nobody completes. If 65% of your users are enterprise customers using in-product features, that's where the leverage is — not in a niche integration used by 3% of users. 4. **Recommend the OMTM.** Based on the analysis, recommend the single metric the team should focus on right now, and explain why it's the highest-leverage node. ### Output Format Deliver: 1. The metric hierarchy (NSM, Business KPI, OMTM recommendation) 2. The full sized tree in text format 3. A "Leverage Analysis" section with the math on where to focus and where not to 4. Any metric quality flags 5. Open questions — what hypotheses need validation, what data is missing Be direct. Be quantitative. Challenge the user if their instinct doesn't match the math. ``` --- ## Tips - Bring data. The more actual numbers you can provide (DAU, conversion rates, segment breakdowns), the more useful the tree will be. Even rough estimates are better than nothing. - If you have Amplitude dashboards with the relevant metrics, share the numbers. The skill works best when every node is sized. - Use this before setting quarterly goals — it prevents teams from picking metrics they can't actually move the needle on. - Pair with **craft-experiment-design** once you've identified the lever — design an experiment to validate your hypothesis about the highest-leverage node. - Revisit the tree quarterly. As you ship changes, the relative size of nodes shifts and the leverage may move.