/pulse

North Star Metric Deep-Dive Reference

Purpose: NSM selection rubric, classification, input-metric decomposition, counter/guardrail pairing, and stability contract. For teams that need a single, actionable North Star anchored to value delivery instead of vanity activity.

Scope Boundary

• pulse northstar: NSM selection, type classification, decomposition tree, counter/guardrail, stability contract (this document).
• pulse kpi (elsewhere): Metric tree construction from an already-chosen NSM. Default entry point.
• pulse retention / activation (elsewhere): Curve-shape diagnostics and aha-moment discovery. NSM decomposition often references these outputs.
• Experiment (elsewhere): Hypothesis validation. NSM is the primary metric a hypothesis should move, not the experiment surface itself.
• Magi (elsewhere): Business strategy and forecast. NSM is a measurement commitment, not a strategic forecast.

Workflow

DISCOVER  →  interview product/business; surface 2-4 NSM candidates
          →  map to Value Exchange vs Engagement vs Experience

CLASSIFY  →  pick NSM type; confirm "drives sustainable revenue"
          →  reject vanity (total signups, pageviews)

DECOMPOSE →  NSM = f(breadth × depth × frequency × efficiency)
          →  derive 3-5 input metrics, each team-controllable

COUNTER   →  pair with counter metric (prevent gaming)
          →  add guardrail (must-not-degrade)

CONTRACT  →  commit to ≥6mo stability (12mo preferred)
          →  document change-triggers (structural shift only)

PUBLISH   →  NSM one-liner + tree diagram + owner + cadence

NSM Type Classification (Amplitude Playbook)

Rejection rule: if the NSM grows while customer value does not (e.g., pageviews after re-theme), it is a vanity metric. Re-classify.

AI-native product NSM (2026 caveat)

For LLM/agent products, do not use raw token consumption as the NSM. Token volume grows because the model is verbose, not because the product creates value. Common 2026 vanity traps to reject: tokens generated, completions returned, average prompt length, model calls per session.

Better candidates (one or two, never all):

• Task-completion NSM ("successful outcomes / week"): the share of agent runs where the user's intent is satisfied per a downstream eval (LLM-judge or rule-based). Pair with explicit success criteria.
• Decision Velocity NSM: time from user intent → trusted decision/action delivered. Useful for analytics copilots and dev tools.
• Eval-aligned NSM: an offline-eval pass rate (e.g., "% queries with grounded, cited answer") promoted as the user-facing NSM, so quality regression is visible to leadership.
• Retained-value NSM ("users completing ≥N satisfying agent runs in 7 days"): use when the product is consumption-oriented.

Always pair AI NSM with a CFO-translatable lagging indicator (revenue per task, retention lift, hours saved priced in $) — if you cannot translate the NSM to dollars within one quarter, it is still an internal dashboard, not a North Star. (Source: Eric Weber — North star metrics for AI data products)

Decomposition Formula

Generic form: NSM = Users_Active × Action_Depth × Frequency × Quality_Multiplier

Input-metric rule: every factor must be team-controllable and have an owner. If no team can move it, it is an output, not an input.

Counter and Guardrail Pairing

NSMs incentivize behavior; bad NSMs incentivize gaming. Always pair with:

Anti-pattern: NSM without counter → team optimizes for the number, not the value. Facebook's early "DAU" had no quality counter; led to notification spam.

Stability Contract

NSM changes should be rare and structural. Acceptable change triggers:

1. Product pivot (B2C → B2B, freemium → enterprise).
2. Market redefinition (new segment, new geography with different value).
3. NSM-to-value drift (metric moves up, revenue flat for 2+ quarters).

Never change NSM because:

• A campaign or feature launch did not move it (that's expected).
• Leadership wants a "cleaner number" (that's political).
• A new dashboard tool is easier with different metric.

Minimum commitment window: 6 months. Preferred: 12 months. Document the change in the NSM Registry with rationale, predecessor, and retention period for historical comparability.

NSM Registry Template

## NSM: [Name]
- **Definition**: [precise operational definition, including units]
- **Type**: Value Exchange | Engagement | Experience
- **Formula**: [decomposition]
- **Input Metrics (3-5)**: [team-owned drivers]
- **Counter Metric**: [prevents gaming]
- **Guardrails**: [must-not-degrade thresholds]
- **Owner**: [person / team accountable]
- **Measurement Window**: [daily | weekly | monthly]
- **Stability Commitment**: [start date] → [minimum end date]
- **Data Source**: [warehouse table / event name]
- **Known Gaming Vectors**: [how teams might gamble this]

Anti-Patterns

Deliverable Contract

When northstar completes, emit:

• NSM one-liner (under 15 words, no jargon).
• Decomposition tree (NSM → 3-5 inputs → leading indicators → events).
• Counter + guardrail pair with thresholds.
• Stability commitment (start date, minimum end date, change-trigger criteria).
• NSM Registry entry (template above).
• Gaming-vector audit (how the metric could be gamed; detection plan).
• Handoff targets: kpi for tree expansion, retention for leading-indicator overlay, dashboard for visualization.

References

• Amplitude — North Star Metric Playbook
• Reforge — Growth Loops and NSM framework (Brian Balfour)
• Sean Ellis — "Hacking Growth" NSM chapter
• Reid Hoffman / LinkedIn — Engagement NSM evolution
• Eric Weber (2026) — North star metrics for AI data products — argues against token-volume NSM for AI products
• LeanPivot (2025) — Finding Your Agent's "North Star" Metric