Metrics Tree

Purpose

A PM has been handed (or has set) a vague goal: "increase activation", "improve retention", "grow revenue". Before they can plan against it, they need to know what the goal actually measures, what inputs move it, and what they can observe sooner than the lagging outcome. This skill produces that structure.

It does NOT recommend targets, benchmarks, or "industry-standard" numbers. It will not say "good activation is 30%." Those numbers are situational; inventing them lets the PM plan against a fake target. The skill maps the structure of the goal and tells the PM what to measure. The PM (or the data team) supplies the numbers.

Procedure

1. Read the PM's stated goal. If it is two or more goals stuffed into one

sentence, split them and ask the PM to pick one before continuing.

1. Define the top metric. State it as a single measurable quantity and write

its defining equation. If the goal admits more than one plausible top metric (e.g. "activation" could be % of signups completing setup, or % reaching first value), name them, pick one as the primary, and note the others as competing definitions the PM should resolve.

1. Decompose the top metric into its input metrics — the multiplicands or

addends in its equation. Each input must be something instrumentable, not a vibe.

1. For each input metric, name 1–2 leading indicators — earlier signals

that predict movement in the input. Leading indicators are observable on a shorter time horizon than the input itself.

1. List instrumentation gaps: every metric in the tree the PM probably

cannot measure today without new tracking. Mark each "you must measure this before you can act on it."

1. Identify the single metric most worth moving first. Justify it in one

short paragraph: why it has the highest expected leverage on the top metric given what the tree shows. If the tree is too thin to support that pick honestly, say so and name what would need to be known to choose.

1. Emit the output contract. Do not emit target numbers, benchmarks, or

comparisons to other companies.

Output contract

Emit exactly these five sections, in this order.

Top metric

The single primary metric and its defining equation. If competing definitions exist, list them and note which the rest of the tree assumes.

Input metrics

The tree, one level deep. Each input metric shown with the operator that combines it into the top metric (multiplied, summed, conditioned on, etc.).

Leading indicators

For each input metric, 1–2 earlier-observable signals that predict its movement.

Instrumentation gaps

The metrics in the tree the PM likely cannot measure today. Each one labeled "you must measure this." No assumed values.

Move this first

The single metric to pursue first, with one short paragraph of reasoning grounded in the tree above. If the tree is too thin to support a confident pick, say so explicitly and name the missing structural information.

Guardrails

• NEVER state a numeric benchmark, target, or "industry standard." No "good

activation is X%", no "best-in-class retention is Y." If the PM asks for numbers, restate that the tree is structural and the numbers come from their own measurement.

• NEVER recommend a metric the PM cannot plausibly instrument. If a metric

belongs in the tree but is uninstrumentable today, put it in "Instrumentation gaps", not in the active tree as if it were live.

• Treat Goodhart explicitly: if the chosen "move this first" metric is highly

gameable (e.g. "logins per week" as a proxy for engagement), call out the failure mode in the reasoning paragraph. Don't recommend a target the team could juke without moving the top metric.

• Do not invent input metrics that are not derivable from the stated goal. If

the goal is genuinely ambiguous about its decomposition, say so under "Top metric" rather than committing to a guess.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: Waddling-Penguin
• Source: Waddling-Penguin/mogkit
• License: MIT
• Homepage: https://mogkit.com

Install and usage instructions live in the source repository linked above.