{{PREAMBLE}}

Metrics Definition Mode

Philosophy

The purpose of metrics is to help you make decisions. If a metric doesn't change a decision you'd make, it's decoration.

Metric theater: Tracking numbers that go up and to the right but don't tell you anything actionable. DAU, page views, "engagement" — these feel productive but rarely drive decisions. "We had 10,000 page views" → so what? What did you learn? What will you do differently?

The test for every metric: "If this metric moves 20% in either direction, what will we do differently?" If the answer is "nothing," it's not a metric — it's a dashboard ornament.

Input vs. Output metrics: Output metrics (revenue, retention, NPS) tell you if you won. Input metrics (feature adoption, task completion rate, time-to-value) tell you WHY you won or lost — and what to do about it. You need both, but most teams only track outputs.

Input

Ask the user (via AskUserQuestion):

What are we defining metrics for?

RECOMMENDATION: Choose A if you're building a new feature. Choose C if you have existing metrics that need review.

A) New feature or product — define metrics from scratch B) Company or team OKRs — define goal metrics with targets C) Existing metrics audit — review what we're tracking and fix gaps D) Dashboard design — I know what to measure, help me structure it

Step 1: North Star Identification

Before individual metrics, define what "winning" looks like.

NORTH STAR METRIC
═════════════════
Metric: [Name — one metric that best captures the value you deliver to users]
Definition: [Exactly what is measured — no ambiguity]
Why this: [Why this metric, and not 3 others, is the North Star]
Current: [Baseline value, or "Unknown — needs instrumentation"]
Target: [Specific number + timeframe]
Cadence: [How often reviewed — daily/weekly/monthly]

Example: "Weekly active searchers who find an answer (not just search, but succeed)"
Not: "Total searches" (activity, not outcome)
Not: "DAU" (too broad — doesn't capture the specific value)

North star rules:

• ONE metric, not five. If everything is the North Star, nothing is.
• Measures value delivered TO THE USER, not activity or revenue directly.
• Revenue is an OUTPUT of delivering user value, not the goal itself (for product metrics).
• If the North Star improves but the business gets worse, you chose the wrong North Star.

Step 2: Metrics Framework

Build the full metrics tree:

METRICS FRAMEWORK: [Product/Feature]
═════════════════════════════════════

NORTH STAR: [Metric name]
│
├── INPUT METRICS (things you can directly influence)
│   ├── [Metric 1]: [Definition]
│   │   Current: [Value] → Target: [Value] by [Date]
│   │   Lever: [What product action moves this metric?]
│   │
│   ├── [Metric 2]: [Definition]
│   │   Current: [Value] → Target: [Value] by [Date]
│   │   Lever: [What product action moves this metric?]
│   │
│   └── [Metric 3]: [Definition]
│       Current: [Value] → Target: [Value] by [Date]
│       Lever: [What product action moves this metric?]
│
├── OUTPUT METRICS (results you're trying to achieve)
│   ├── [Metric 4]: [Definition]
│   │   Current: [Value] → Target: [Value] by [Date]
│   │   Lag time: [How long after input change before this moves?]
│   │
│   └── [Metric 5]: [Definition]
│       Current: [Value] → Target: [Value] by [Date]
│       Lag time: [How long after input change before this moves?]
│
├── GUARDRAIL METRICS (must NOT get worse)
│   ├── [Metric 6]: [Definition]
│   │   Threshold: [Value that triggers investigation]
│   │   Alert: [Who gets paged and how]
│   │
│   └── [Metric 7]: [Definition]
│       Threshold: [Value that triggers investigation]
│       Alert: [Who gets paged and how]
│
└── LEADING INDICATORS (early signals before lagging metrics move)
    ├── [Signal 1]: [What to watch in first week/month]
    │   Healthy range: [Expected values if things are working]
    │
    └── [Signal 2]: [What to watch in first week/month]
        Healthy range: [Expected values if things are working]

Metric Types Explained

Input metrics = things you can influence through product changes. Feature adoption rate, onboarding completion %, search success rate. These are YOUR levers.

Output metrics = results that follow from good inputs. Revenue, retention, NPS, customer lifetime value. These are lagging — they move weeks/months after inputs change.

Guardrail metrics = things that should NOT get worse when you optimize inputs. If you improve search speed but degrade search quality, the guardrail catches it. Common guardrails: error rate, latency p95, support ticket volume, churn rate.

Leading indicators = early signals that predict whether output metrics will move. If you launch a feature today, you can't measure 90-day retention for 90 days. But you CAN measure Day 1 activation, which correlates with retention. Leading indicators give you signal in days, not months.

Step 3: Metric Quality Audit

For each metric, run this checklist:

METRIC QUALITY AUDIT
════════════════════
| Check | [Metric Name] |
|──────────────────────────────|────────────────|
| Can be gamed? | [Yes/No — if yes, gaming vector: ___] |
| Actionable? (20% change → what do we do?) | [Yes/No — action: ___] |
| Measurable today? | [Yes/No — if no, instrumentation needed: ___] |
| Has a baseline? | [Yes/No — if no, measure for ___ before setting target] |
| Target has rationale? | [Yes/No — based on: benchmarks / growth rate / ___] |
| Leading indicator exists? | [Yes/No — if no, suggest one] |
| Guardrail exists? | [Yes/No — if no, what could break?] |
| Review cadence defined? | [Daily / Weekly / Monthly] |
| Owner defined? | [Name/Role] |

If a metric fails 3+ checks, flag it:

⚠️ [Metric name] — fails [N] quality checks.
Consider: [Replace with X / Add guardrail Y / Instrument Z first]

Step 4: Measurement Plan

How will each metric actually get measured?

MEASUREMENT PLAN
════════════════
| Metric | Source | Method | Frequency | Owner |
|--------|--------|--------|-----------|-------|
| [Metric 1] | [Analytics tool / DB query / Survey] | [How calculated] | [How often] | [Who] |
| [Metric 2] | [Source] | [Method] | [Frequency] | [Owner] |

INSTRUMENTATION NEEDED:
1. [Event/data point to add]: [Where — frontend/backend/infra] — Est. effort: [S/M/L]
2. [Event/data point to add]: [Where] — Est. effort: [S/M/L]

DATA GAPS:
- [What we can't measure today and what we'd need to fix it]
- [Data quality concern and remediation plan]

Step 5: Dashboard Wireframe

DASHBOARD: [Product/Feature] Metrics
═════════════════════════════════════

┌─────────────────────────────────────────────────────────────┐
│ NORTH STAR: [Metric name]                                    │
│ ████████████████████░░░░░  [Current] / [Target]   [Trend ↑] │
└─────────────────────────────────────────────────────────────┘

┌──────────────────────┐  ┌──────────────────────┐
│ INPUT: [Metric 1]     │  │ INPUT: [Metric 2]     │
│ [Value] → [Target]    │  │ [Value] → [Target]    │
│ Trend: [↑↓→]  [Spark] │  │ Trend: [↑↓→]  [Spark] │
└──────────────────────┘  └──────────────────────┘

┌──────────────────────┐  ┌──────────────────────┐
│ OUTPUT: [Metric 3]    │  │ GUARDRAIL: [Metric 4] │
│ [Value] → [Target]    │  │ [Value] / Threshold    │
│ Lag: [N] days         │  │ Status: 🟢🟡🔴        │
└──────────────────────┘  └──────────────────────┘

┌─────────────────────────────────────────────────────────────┐
│ LEADING INDICATORS                                           │
│ [Signal 1]: [Value] — Healthy: [Range]                       │
│ [Signal 2]: [Value] — Healthy: [Range]                       │
└─────────────────────────────────────────────────────────────┘

Dashboard rules:

• North Star is the biggest, most prominent element
• Guardrails show threshold + status (green/yellow/red), not just the number
• Every metric shows TREND, not just current value
• Leading indicators are separate from lagging outputs — different cadence, different audience

Step 6: Review Cadence

METRICS REVIEW CADENCE
═══════════════════════
| Cadence | What's reviewed | Who | Action on anomaly |
|---------|----------------|-----|------------------|
| Daily | Guardrails + leading indicators | PM + Eng lead | Investigate if breached |
| Weekly | Input metrics + North Star | Product team | Adjust priorities if off-track |
| Monthly | Output metrics + full framework | PM + leadership | Strategic pivot if needed |
| Quarterly | Framework audit — are we measuring the right things? | PM + leadership | Add/remove/redefine metrics |

Completion Summary

METRICS FRAMEWORK SUMMARY
═════════════════════════════════════
| Component | Status |
|──────────────────────|──────────────────────|
| North Star | ✓ Defined: [name] |
| Input metrics | ___ defined |
| Output metrics | ___ defined |
| Guardrails | ___ defined |
| Leading indicators | ___ defined |
| Quality audit | ___ pass, ___ flagged |
| Measurement plan | ✓ / Needs instrumentation |
| Dashboard wireframe | ✓ Designed |
| Review cadence | ✓ Defined |
| Instrumentation gaps | ___ items |
═════════════════════════════════════

Ready for: /launch (metrics feed into launch success criteria) or /brief (to communicate metrics strategy)

Common Anti-Patterns

Flag these if you see them in existing metrics:

1. Vanity metrics — Numbers that only go up (total users, total revenue) without rate or cohort analysis
2. Activity metrics as outcomes — "Number of searches" instead of "search success rate"
3. Missing guardrails — Optimizing one metric at the expense of another without noticing
4. No baseline — Targets set without knowing where you're starting from
5. Unmeasurable metrics — "User satisfaction" with no survey, no NPS, no proxy
6. Output-only dashboard — Shows if you won but not why (missing input metrics)
7. Lagging-only indicators — Won't know if it's working for 90 days (add leading indicators)

Formatting Rules

• Metrics tree uses the branch diagram format (├── └──) for hierarchy
• Every metric has: Definition, Current, Target, Cadence, Owner
• Guardrails show thresholds and alert mechanisms
• Dashboard wireframes use ASCII box art
• After the metrics framework, pause: "Does this framework capture what matters? Any metrics missing?"