North Star Input Tree

---
name: north-star-input-tree
description: "Build a North Star Metric and decompose it into 3-5 controllable input metrics that sum/multiply to it — the Amplitude framework as a standalone build. Covers NSM selection criteria, the input-metric tree, finding the weakest input to focus on, and the anti-patterns (revenue/vanity NSMs). Makes a North Star *actionable* instead of a scoreboard. Used by metrics-architect."
---

## Codex host binding

- Treat references to Claude slash workflows as the equivalently named Codex skill.
- Before delegating to any specialist, read the `great-pm-runtime` skill and the selected packaged role file.
- Treat "invoke", "assign", "delegate", "spawn", and source Agent-tool instructions as a required Codex `spawn_agent` call with that role and a bounded assignment.
- For the spawn tool's `task_name` or `agent_name` identifier, replace canonical role-name hyphens with underscores (`query-refiner-pm` → `query_refiner_pm`). Keep canonical names unchanged in prompts, filenames, verdict logs, and user-facing output; this is an internal identifier mapping, not a role rename.
- Store every returned agent identifier. Never call a wait tool until a spawn has returned an identifier, and wait only on identifiers returned by successful spawns.
- If `spawn_agent` is unavailable or a spawn fails, report BLOCKED; do not impersonate the specialist or wait on an empty agent set.
- Resolve bundled paths from the installed GreatPM plugin root.
- Ignore Claude-only model aliases, colors, turn limits, and tool allowlists.
- Preserve GreatPM human gates, governance, state, and reporting contracts.


> **Provenance.** great-pm-original, authored 2026-05-29, grounded in the
> cited sources (Amplitude North Star Playbook, Open Practice Library,
> Alex Morcillo). Host agent: metrics-architect.

# North Star + Input-Metric Tree — making the one number actionable

A North Star Metric (NSM) on its own is a **scoreboard**: it tells you the
score but not which move to make. It becomes *actionable* only when
decomposed into **3–5 input metrics** — the levers teams directly
influence that *combine* (sum or multiply) to produce the North Star.
Sean Ellis coined the NSM concept; **Amplitude formalized the
input-metrics tree** that turns it from a vanity dashboard into a
decision-making instrument. This skill builds both halves: select an NSM
that survives the criteria, then decompose it into a tree the team can
move.

The defining property: a North Star expresses the **value customers get
from the product**, framed as a *leading* indicator of business success —
not the business success itself (that's revenue, a lagging output). The
input tree is what lets a decentralized team find the *weakest input*,
run experiments there, and check whether the North Star moved.

## 1. Selecting the North Star — the criteria

A candidate NSM must satisfy *all* of these. Score each candidate against
the checklist; the one that passes cleanly wins.

| Criterion | What it means | Failure mode it catches |
|-----------|---------------|--------------------------|
| **Expresses customer value** | Captures the value the *user* gets, not the value *you* extract | Revenue/bookings (value to you, lagging) |
| **Reflects vision & strategy** | A proxy for the strategy working; if it moves, the bet is paying off | A generic metric unconnected to the actual bet |
| **Leading, not lagging** | Predicts future business results; moves *before* revenue does | Revenue, churn (rear-view) |
| **Actionable** | Teams can take actions that measurably move it | A macro metric no single team can influence |
| **Understandable** | Stateable in plain language to anyone in the company | A composite "engagement score" nobody can explain |
| **Measurable** | A clear unit + period, instrumentable without ambiguity | "Customer happiness" with no definition |
| **Can credibly drop** | Has a real down direction (the honesty test) | Cumulative totals that only ever rise (vanity) |

**The single best heuristic:** the NSM sits at the intersection of
*customer value* and *business value*, expressed as the **moment/quantity
of value delivered**. Examples that pass:
- Spotify — *time spent listening* (value experienced, leading on
  subscription retention).
- Airbnb — *nights booked* (the core value exchange).
- Slack — *teams actively messaging* (value-in-use, leading on expansion).
- WhatsApp — *messages sent* (the product *is* messaging).

Examples that **fail** (anti-patterns): revenue / ARR (lagging, business-
not-customer value), "total registered users" (cumulative, can't drop,
value not delivered), a vague "engagement" composite (un-understandable,
ungameable into meaninglessness), DAU *with no value qualifier* (opened ≠
got value).

## 2. The input-metric tree — decomposing the NSM

The NSM is the trunk. The branches are **3–5 input metrics** that satisfy
two joint conditions:

1. **They collectively produce the NSM.** Stated as a rough equation —
   they should *sum or multiply* to the North Star. The tree is wrong if
   moving every input wouldn't move the NSM.
2. **Each is directly influenceable** by the daily work of a team. An
   input metric a team can't actually move is decoration.

### How to find the inputs (the decomposition)

Ask: *"What has to be true for the North Star to go up?"* The answers,
deduplicated to the few that matter, are the inputs. Use these common
decomposition lenses:
- **Breadth × Depth × Frequency × Efficiency.** Many NSMs factor as
  (how many users) × (how deeply each engages) × (how often) × (how
  reliably the product delivers value when they try).
- **Acquisition → Activation → Retention** of the value-event (each a
  controllable rate feeding the value the NSM counts).
- **Quality × Quantity.** A volume metric times a quality/success rate
  (this catches the trap of pumping volume while value-per-event falls).

Keep it to **3–5**. Two inputs is usually under-decomposed (you've hidden
sub-levers); more than five and no team can focus — the whole point is to
let a team pick *the weakest input* and attack it.

### Operating the tree

1. **Measure all inputs.** Each is itself a metric with a unit, period,
   and instrumentation spec (see `metrics-design`).
2. **Find the weakest input** — the one most below its potential / most
   constraining the NSM. That is where the next experiments go.
3. **Run experiments on the weakest input** (hand to `experiment-design`).
4. **Check the NSM actually moved.** If you improved an input and the NSM
   didn't budge, your tree is wrong — the input either doesn't really
   feed the NSM, or another input fell as this one rose. Re-derive.
5. **Re-balance over time.** As the weakest input is fixed, a different
   one becomes the constraint. The tree is a living instrument.

### The three jobs the tree does (Amplitude)

- **Prioritize & accelerate decentralized decisions** — a team can pick
  an input and act without escalating; the tree is the shared map.
- **Align & communicate** — everyone sees how their work ladders to the
  one number.
- **Focus on sustainable product-led growth** — inputs are value-and-
  product levers, steering away from vanity and short-term hacks.

## 3. When to use — and when NOT to

| Use it when… | Because… |
|---------------|----------|
| You have a North Star nobody can act on | The tree converts a scoreboard into levers |
| Setting up measurement for a new product/initiative | NSM + inputs is the backbone the KPI plan hangs on |
| Teams are optimizing local metrics that don't ladder up | The tree realigns everyone to one number |
| "Revenue is our North Star" is being proposed | Apply §1 criteria — revenue is a lagging output, not an NSM |

**When NOT to use:**
- **Pre-product-market-fit / pure exploration.** Before you know what
  value looks like, a fixed NSM tree can ossify the wrong bet. Use
  discovery + a provisional metric instead; build the tree once the value
  is clear.
- **A single team / single feature.** A full company NSM tree is overkill
  for one feature's success metric — use `metrics-design` for the local
  KPI and slot it under the existing tree.
- **As a substitute for strategy.** The NSM *reflects* strategy; it
  doesn't *create* it. If there's no strategy bet, fix that first
  (`brainstorm-okrs` / strategy work), then derive the NSM from it.
- **Don't let the tree become a giant scorecard.** More than ~5 inputs and
  it stops focusing decisions — that's the anti-use.

## 4. Worked example — Acme

**Strategy bet** (per MEMORY / instructions.md §1): Acme wins by giving
Indian consumers *effortless, accurate, recurring* visibility into their
cash flow — a money-awareness *habit*. The North Star must express that
value as a leading indicator.

**NSM candidate (passes §1 criteria):**
> **Weekly active users who viewed an accurate categorized cash-flow
> view** ("weekly money-aware sessions").

Check it:
- Customer value? Yes — the user *gets* visibility (the §1 promise). ✓
- Reflects strategy? Yes — the recurring-awareness-habit bet. ✓
- Leading? Yes — predicts retention/subscription, ahead of revenue. ✓
- Actionable? Yes — multiple teams can move it. ✓
- Understandable? Yes — plain language. ✓
- Measurable? Yes — weekly count, defined "accurate view" event. ✓
- Can drop? Yes — if accuracy or habit slips, it falls. ✓

(Rejected alternatives: *transactions categorized* — volume without value
felt; *revenue* — lagging, business-not-user; *total users* — cumulative,
can't drop.)

**The input tree (rough product):**

```
North Star: Weekly money-aware sessions
  ≈ (active users)
    × (share who connected ≥1 data source)        ← INPUT 1: data breadth
    × (categorization accuracy rate)               ← INPUT 2: quality
    × (share returning within their weekly cadence) ← INPUT 3: habit/freq
  with TTV gating who ever reaches the first session ← INPUT 4: time-to-first-insight
```

| # | Input metric | Who moves it | Lens |
|---|--------------|--------------|------|
| 1 | % of users with ≥1 source connected (breadth) | onboarding / integrations | breadth |
| 2 | Categorization accuracy rate (% not overridden) | the AI gate pipeline / evals | quality |
| 3 | % returning within their weekly cadence (habit) | retention / notifications | frequency |
| 4 | Time-to-first-insight (TTV) | onboarding | activation/efficiency |

**Operating it.** Suppose accuracy (Input 2) is at 78% and overrides are
high — that's the **weakest input**, and it's *causal*: an inaccurate view
isn't a money-aware session, it's a chore. Focus experiments there (better
gate routing, more merchant rules, eval-driven prompt fixes). Improve
accuracy → fewer overrides → more sessions *count* as money-aware → NSM
rises. Crucially, **verify the NSM moved**: if accuracy rose but weekly
money-aware sessions didn't, the tree is mis-specified (maybe habit, Input
3, is the real constraint and users churn before accuracy matters).

**Cross-tie:** Input 4 (TTV) is exactly the lever
`activation-aha-moment` optimizes; Input 2 (accuracy) is what `ai-evals`
measures; the drops *between* connecting a source and the first insight
are what `funnel-diagnostics` repairs. The tree is the spine that
organizes all three.

## 5. Common pitfalls

| Pitfall | Why it fails | Fix |
|---------|--------------|-----|
| NSM with no input tree | Un-actionable scoreboard; teams optimize local metrics that don't ladder up | Decompose into 3–5 controllable inputs that combine to the NSM |
| Revenue (or any lagging output) as the NSM | Rear-view, expresses *your* value not the *customer's*; can't be steered directly | Pick a leading customer-value metric; revenue is the *outcome* of the NSM |
| Vague "engagement" composite | Un-understandable, un-actionable, gameable into meaninglessness | Use a concrete value event with a unit and period |
| A metric that can't credibly drop | Cumulative totals (total users/transactions) are vanity | Apply the honesty test: does it have a real down direction? |
| Too many inputs (> 5) | No team can focus; defeats the "attack the weakest input" purpose | Deduplicate to the few that genuinely drive the NSM |
| Inputs the team can't move | Decoration, not levers | Every input must be directly influenceable by daily work |
| Never re-checking that the input moved the NSM | You "win" on an input while the NSM is flat — the tree is wrong | Always verify the NSM responded; re-derive the tree if not |
| Volume input with no quality counterpart | You pump events while value-per-event falls (more sessions, less awareness) | Pair volume with a quality/success rate input |

## 6. Ethical boundary

The North Star must measure value the **user actually receives**, not
value extracted *from* them. An NSM secretly optimized for the company's
take (sessions inflated by manipulative notifications, "active" counted on
hollow opens) corrupts every decision the tree drives downstream — teams
will dutifully grow a number that no longer represents user benefit. The
honesty test ("can it drop?") is also an ethics test: if your metric only
rises because it counts something the user didn't value, you've built a
vanity engine with a strategy doc attached. Anchor the NSM and its inputs
on genuine value delivered.

## 7. Quick diagnostic

| Question | If "no" → |
|----------|-----------|
| Does the NSM express **customer value**, leading not lagging? | Re-select against §1 criteria; reject revenue/totals |
| Can you state it in **plain language** with a unit + period? | Sharpen the definition |
| Can it **credibly drop**? | It's vanity — pick an honest metric |
| Is it decomposed into **3–5 controllable inputs** that combine to it? | Build the tree (§2) |
| Can each team **directly move** its input? | Replace decoration inputs with real levers |
| Have you identified the **weakest input** to focus on? | Measure all inputs; find the constraint |
| When you moved an input, did the **NSM actually respond**? | Re-derive the tree — it's mis-specified |

## 8. Cross-links & boundary

- **`metrics-design`** — names the North Star + AARRR at a *definition*
  level (one paragraph each); this skill is the full Amplitude *tree*
  methodology. metrics-design specs the events/thresholds for each node
  *after* the tree is built.
- **`activation-aha-moment`** — activation rate / TTV is almost always one
  of the input metrics; build the aha there, slot it in here.
- **`funnel-diagnostics`** — funnel conversion rates are common input
  metrics; a leaking funnel step is often the *weakest input*.
- **`improve-retention`** — a retention/habit input lives here as a
  branch; retention *tactics* live in that skill.
- **`cohort-analysis`** — supplies the retention-curve data behind
  retention-flavored inputs.
- **`experiment-design`** — how you act on the weakest input: experiments
  on that branch, then verify the NSM moved.
- **`brainstorm-okrs`** — input metrics make excellent OKR key results;
  the NSM is the objective they ladder to. (The NSM *reflects* strategy —
  derive it from the strategy bet, don't invent it free-standing.)

**Differs from `metrics-design` because** that defines a KPI set
conceptually (one line on the NSM); this *builds the decomposition tree* —
the actionable mechanism. **Differs from `brainstorm-okrs` because** that
sets objectives/key-results for a period; this builds the *enduring
measurement spine* objectives ladder to. **Differs from `cohort-analysis`
/ `funnel-diagnostics` because** those are diagnostic techniques that
*feed* input branches; this is the *framework that organizes* them under
one number.

## Sources

- Amplitude, "The North Star Playbook" — https://amplitude.com/books/north-star/about-the-north-star-framework
- Open Practice Library, "North Star Framework" — https://openpracticelibrary.com/practice/north-star-framework/
- Alex Morcillo, "Your North Star Metric Is Wrong: Spotify, Airbnb, Slack" — https://alexmorcillo.com/blog/strategy/what-is-a-north-star-and-what-is-not/
- Sean Ellis (originator of the North Star Metric concept; GrowthHackers) — widely documented in the Amplitude playbook above