14 KiB
Metrics Tree Methodology
When to use this methodology: You've used template.md and need advanced techniques for:
- Multi-sided marketplaces or platforms
- Complex metric interdependencies and feedback loops
- Counter-metrics and guardrail systems
- Network effects and viral growth
- Preventing metric gaming
- Seasonal adjustment and cohort aging effects
- Portfolio approach for different business stages
If your metrics tree is straightforward: Use template.md directly. This methodology is for complex metric systems.
Table of Contents
- Multi-Sided Marketplace Metrics
- Counter-Metrics & Guardrails
- Network Effects & Viral Loops
- Preventing Metric Gaming
- Advanced Leading Indicators
- Metric Interdependencies
- Business Stage Metrics
1. Multi-Sided Marketplace Metrics
Challenge
Marketplaces have supply-side and demand-side that must be balanced. Optimizing one side can hurt the other.
Solution: Dual Tree Approach
Step 1: Identify constraint
- Supply-constrained: More demand than supply → Focus on supply-side metrics
- Demand-constrained: More supply than demand → Focus on demand-side metrics
- Balanced: Need both → Monitor ratio/balance metrics
Step 2: Create separate trees
Supply-Side Tree:
North Star: Active Suppliers (providing inventory)
├─ New supplier activation
├─ Retained suppliers (ongoing activity)
└─ Supplier quality/performance
Demand-Side Tree:
North Star: Successful Transactions
├─ New customer acquisition
├─ Repeat customer rate
└─ Customer satisfaction
Step 3: Define balance metrics
- Liquidity ratio: Supply utilization rate (% of inventory sold)
- Match rate: % of searches resulting in transaction
- Wait time: Time from demand signal to fulfillment
Example (Uber):
- Supply NS: Active drivers with >10 hours/week
- Demand NS: Completed rides
- Balance metric: Average wait time <5 minutes, driver utilization >60%
Multi-Sided Decomposition Template
Marketplace GMV = (Supply × Utilization) × (Demand × Conversion) × Average Transaction
Where:
- Supply: Available inventory/capacity
- Utilization: % of supply that gets used
- Demand: Potential buyers/searches
- Conversion: % of demand that transacts
- Average Transaction: $ per transaction
2. Counter-Metrics & Guardrails
Problem
Optimizing primary metrics can create negative externalities (quality drops, trust declines, user experience suffers).
Solution: Balanced Scorecard with Guardrails
Framework:
- Primary metric (North Star): What you're optimizing
- Counter-metrics: What could be harmed
- Guardrail thresholds: Minimum acceptable levels
Example (Content Platform):
Primary: Content Views (maximize)
Counter-metrics with guardrails:
- Content quality score: Must stay ≥7/10 (current: 7.8)
- User satisfaction (NPS): Must stay ≥40 (current: 52)
- Creator retention: Must stay ≥70% (current: 75%)
- Time to harmful content takedown: Must be ≤2 hours (current: 1.5h)
Rule: If any guardrail is breached, pause optimization of primary metric
Common Counter-Metric Patterns
| Primary Metric | Potential Harm | Counter-Metric |
|---|---|---|
| Pageviews | Clickbait, low quality | Time on page, bounce rate |
| Engagement time | Addictive dark patterns | User-reported wellbeing, voluntary sessions |
| Viral growth | Spam | Unsubscribe rate, report rate |
| Conversion rate | Aggressive upsells | Customer satisfaction, refund rate |
| Speed to market | Technical debt | Bug rate, system reliability |
How to Set Guardrails
- Historical baseline: Look at metric over past 6-12 months, set floor at 10th percentile
- Competitive benchmark: Set floor at industry average
- User feedback: Survey users on acceptable minimum
- Regulatory: Compliance thresholds
3. Network Effects & Viral Loops
Measuring Network Effects
Network effect: Product value increases as more users join.
Metrics to track:
- Network density: Connections per user (higher = stronger network)
- Cross-side interactions: User A's action benefits User B
- Viral coefficient (K): New users generated per existing user
- K > 1: Exponential growth (viral)
- K < 1: Sub-viral (need paid acquisition)
Decomposition:
New Users = Existing Users × Invitation Rate × Invitation Acceptance Rate
Example:
100,000 users × 2 invites/user × 50% accept = 100,000 new users (K=1.0)
Viral Loop Metrics Tree
North Star: Viral Coefficient (K)
Decomposition:
K = (Invitations Sent / User) × (Acceptance Rate) × (Activation Rate)
Input metrics:
├─ Invitations per user
│ ├─ % users who send ≥1 invite
│ ├─ Average invites per sender
│ └─ Invitation prompts shown
├─ Invite acceptance rate
│ ├─ Invite message quality
│ ├─ Social proof (sender credibility)
│ └─ Landing page conversion
└─ New user activation rate
├─ Onboarding completion
├─ Value realization time
└─ Early engagement actions
Network Density Metrics
Measure connectedness:
- Average connections per user
- % of users with ≥N connections
- Clustering coefficient (friends-of-friends)
- Active daily/weekly connections
Threshold effects:
- Users with 7+ friends have 10x retention (identify critical mass)
- Teams with 10+ members have 5x engagement (team size threshold)
4. Preventing Metric Gaming
Problem
Teams optimize for the letter of the metric, not the spirit, creating perverse incentives.
Gaming Detection Framework
Step 1: Anticipate gaming For each metric, ask: "How could someone game this?"
Example metric: Time on site
- Gaming: Auto-play videos, infinite scroll, fake engagement
- Intent: User finds value, willingly spends time
Step 2: Add quality signals Distinguish genuine value from gaming:
Time on site (primary)
+ Quality signals (guards against gaming):
- Active engagement (clicks, scrolls, interactions) vs passive
- Return visits (indicates genuine interest)
- Completion rate (finished content vs bounced)
- User satisfaction rating
- Organic shares (not prompted)
Step 3: Test for gaming
- Spot check: Sample user sessions, review for patterns
- Anomaly detection: Flag outliers (10x normal behavior)
- User feedback: "Was this session valuable to you?"
Gaming Prevention Patterns
Pattern 1: Combination metrics Don't measure single metric; require multiple signals:
❌ Single: Pageviews
✓ Combined: Pageviews + Time on page >30s + Low bounce rate
Pattern 2: User-reported value Add subjective quality check:
Primary: Feature adoption rate
+ Counter: "Did this feature help you?" (must be >80% yes)
Pattern 3: Long-term outcome binding Tie short-term to long-term:
Primary: New user signups
+ Bound to: 30-day retention (signups only count if user retained)
Pattern 4: Peer comparison Normalize by cohort or segment:
Primary: Sales closed
+ Normalized: Sales closed / Sales qualified leads (prevents cherry-picking easy wins)
5. Advanced Leading Indicators
Technique 1: Propensity Scoring
Predict future behavior from early signals.
Method:
- Collect historical data: New users + their 30-day outcomes
- Identify features: Day 1 behaviors (actions, time spent, features used)
- Build model: Logistic regression or decision tree predicting 30-day retention
- Score new users: Probability of retention based on day 1 behavior
- Threshold: Users with >70% propensity score are "likely retained"
Example (SaaS):
30-day retention model (R² = 0.78):
Retention = 0.1 + 0.35×(invited teammate) + 0.25×(completed 3 workflows) + 0.20×(time in app >20min)
Leading indicator: % of users with propensity score >0.7
Current: 45% → Target: 60% (predicts 15% retention increase)
Technique 2: Cohort Behavior Clustering
Find archetypes that predict outcomes.
Method:
- Segment users by first-week behavior patterns
- Measure long-term outcomes per segment
- Identify high-value archetypes
Example:
Archetypes (first week):
- "Power user": 5+ days active, 20+ actions → 85% retain
- "Social": Invites 2+ people, comments 3+ times → 75% retain
- "Explorer": Views 10+ pages, low actions → 40% retain
- "Passive": <3 days active, <5 actions → 15% retain
Leading indicator: % of new users becoming "Power" or "Social" archetypes
Target: Move 30% → 45% into high-value archetypes
Technique 3: Inflection Point Analysis
Find tipping points where behavior changes sharply.
Method:
- Plot outcome (retention) vs candidate metric (actions taken)
- Find where curve steepens (inflection point)
- Set that as leading indicator threshold
Example:
Retention by messages sent (first week):
- 0-2 messages: 20% retention (slow growth)
- 3-9 messages: 45% retention (moderate growth)
- 10+ messages: 80% retention (sharp jump)
Inflection point: 10 messages
Leading indicator: % of users hitting 10+ messages in first week
6. Metric Interdependencies
Problem
Metrics aren't independent; changing one affects others in complex ways.
Solution: Causal Diagram
Step 1: Map relationships Draw arrows showing how metrics affect each other:
[Acquisition] → [Active Users] → [Engagement] → [Retention]
↓ ↑
[Activation] ----------------------------------------
Step 2: Identify feedback loops
- Positive loop (reinforcing): A → B → A (exponential) Example: More users → more network value → more users
- Negative loop (balancing): A → B → ¬A (equilibrium) Example: More supply → lower prices → less supply
Step 3: Predict second-order effects If you increase metric X by 10%:
- Direct effect: Y increases 5%
- Indirect effect: Y affects Z, which feeds back to X
- Net effect: May be amplified or dampened
Example (Marketplace):
Increase driver supply +10%:
→ Wait time decreases -15%
→ Rider satisfaction increases +8%
→ Rider demand increases +5%
→ Driver earnings decrease -3% (more competition)
→ Driver churn increases +2%
→ Net driver supply increase: +10% -2% = +8%
Modeling Tradeoffs
Technique: Regression or experiments
Run A/B test increasing X
Measure all related metrics
Calculate elasticities:
- If X increases 1%, Y changes by [elasticity]%
Build tradeoff matrix
Tradeoff Matrix Example:
| If increase by 10% | Acquisition | Activation | Retention | Revenue |
|---|---|---|---|---|
| Acquisition | +10% | -2% | -1% | +6% |
| Activation | 0% | +10% | +5% | +12% |
| Retention | 0% | +3% | +10% | +15% |
Interpretation: Investing in retention has best ROI (15% revenue lift vs 6% from acquisition).
7. Business Stage Metrics
Problem
Optimal metrics change as business matures. Early-stage metrics differ from growth or maturity stages.
Stage-Specific North Stars
Pre-Product/Market Fit (PMF):
- Focus: Finding PMF, not scaling
- North Star: Retention (evidence of value)
- Key metrics:
- Week 1 → Week 2 retention (>40% = promising)
- NPS or "very disappointed" survey (>40% = good signal)
- Organic usage frequency (weekly+ = habit-forming)
Post-PMF, Pre-Scale:
- Focus: Unit economics and growth
- North Star: New activated users per week (acquisition + activation)
- Key metrics:
- LTV/CAC ratio (target >3:1)
- Payback period (target <12 months)
- Month-over-month growth rate (target >10%)
Growth Stage:
- Focus: Efficient scaling
- North Star: Revenue or gross profit
- Key metrics:
- Net revenue retention (target >100%)
- Magic number (ARR growth / S&M spend, target >0.75)
- Burn multiple (cash burned / ARR added, target <1.5)
Maturity Stage:
- Focus: Profitability and market share
- North Star: Free cash flow or EBITDA
- Key metrics:
- Operating margin (target >20%)
- Market share / competitive position
- Customer lifetime value
Transition Triggers
When to change North Star:
PMF → Growth: When retention >40%, NPS >40, organic growth observed
Growth → Maturity: When growth rate <20% for 2+ quarters, market share >30%
Migration approach:
- Track both old and new North Star for 2 quarters
- Align teams on new metric
- Deprecate old metric
- Update dashboards and incentives
Quick Decision Trees
"Should I use counter-metrics?"
Is primary metric easy to game or has quality risk?
├─ YES → Add counter-metrics with guardrails
└─ NO → Is metric clearly aligned with user value?
├─ YES → Primary metric sufficient, monitor only
└─ NO → Redesign metric to better capture value
"Do I have network effects?"
Does value increase as more users join?
├─ YES → Track network density, K-factor, measure at different scales
└─ NO → Does one user's action benefit others?
├─ YES → Measure cross-user interactions, content creation/consumption
└─ NO → Standard metrics tree (no network effects)
"Should I segment my metrics tree?"
Do different user segments have different behavior patterns?
├─ YES → Do segments have different value to business?
├─ YES → Create separate trees per segment, track segment mix
└─ NO → Single tree, annotate with segment breakdowns
└─ NO → Are there supply/demand sides?
├─ YES → Dual trees (Section 1)
└─ NO → Single unified tree
Summary: Advanced Technique Selector
| Scenario | Use This Technique | Section |
|---|---|---|
| Multi-sided marketplace | Dual tree + balance metrics | 1 |
| Risk of negative externalities | Counter-metrics + guardrails | 2 |
| Viral or network product | K-factor + network density | 3 |
| Metric gaming risk | Quality signals + combination metrics | 4 |
| Need better prediction | Propensity scoring + archetypes | 5 |
| Complex interdependencies | Causal diagram + elasticities | 6 |
| Changing business stage | Stage-appropriate North Star | 7 |