Metrics Definitions
A metric without a definition is a slogan. Write metrics so two analysts get the same number from the same warehouse.
Workflow
- Clarify decision the metric informs (grow activation? reduce churn?).
- Propose a small set: North Star (optional), input metrics, guardrails.
- For each metric: formula, grain, time window, filters, owners.
- Map required events/properties (tracking plan lite).
- List non-goals and common misreads.
- Suggest validation checks (reconciliation, anomaly thresholds).
Output format
## Metrics spec: <product/area>
### North Star (if any)
**Name:** …
**Formula:** …
**Why it matters:** …
### Input metrics
| Name | Formula | Grain | Window | Owner |
### Guardrails
| Name | Formula | Alert idea |
### Tracking requirements
- event.property …
### Edge cases
- …
### Anti-metrics / misreads
- …
Rules
- Always define numerator, denominator, and who is in the population.
- Prefer ratios with explicit windows (e.g. D7 retention) over vague "engagement".
- Separate leading indicators from lagging outcomes.
- Do not invent data you cannot collect; flag instrumentation gaps.
- One source of truth name per metric; note deprecated aliases.
- Call out gaming risk (metrics people will optimize naively).
Edge cases
- B2B seats vs accounts: state which entity is the unit.
- Marketplace: define both sides; avoid averaging away imbalance.
- Early stage: fewer metrics, higher instrumentation quality.