florianbonnet14 Cohort Analysis Specialist
Track customer behavior over time by grouping customers into cohorts for temporal analysis. Essential for understanding retention, lifecycle patterns, and product evolution.
Core Framework: Understanding Cohorts
What is a Cohort?
A group of users who share a common characteristic or experience within a defined time period.
Most common: Acquisition cohort (when they signed up)
Why Cohorts Matter: Overall metrics can hide important trends. Example:
- Overall retention: 60%
- But actually: 2023 cohorts 75%, 2024 cohorts 45%
- Without cohorts, you wouldn't know new users perform worse!
Cohort Types
Type 1: Acquisition Cohorts
Definition: Grouped by signup/first purchase date
Time granularity:
- Daily (high volume, short analysis)
- Weekly (most common, balances detail/manageability)
- Monthly (strategic analysis, lower volume)
- Quarterly (very long-term trends)
Best for: Retention analysis, lifecycle understanding, LTV calculation
Type 2: Behavioral Cohorts
Definition: Grouped by specific action
Examples: Activated cohort, Converters, Feature adopters, Engagement level
Best for: Feature impact analysis, engagement optimization
Type 3: Attribute Cohorts
Definition: Grouped by characteristic at acquisition
Examples: Acquisition channel, Plan tier, Geographic, Segment
Best for: Channel quality, segment performance, market analysis
Key Analysis: The Retention Table
Classic Cohort Retention Table
Structure:
M0 M1 M2 M3 M4 M5 M6
Jan 24 100% 65% 52% 45% 41% 38% 36%
Feb 24 100% 68% 55% 48% 44% 40% 38%
Mar 24 100% 70% 58% 51% 46% 42% --
Apr 24 100% 72% 60% 53% 48% -- --
May 24 100% 74% 62% 55% -- -- --
Jun 24 100% 75% 63% -- -- -- --
Jul 24 100% 76% -- -- -- -- --
How to read:
- Rows: Each cohort (signup month)
- Columns: Time since signup
- Cells: % of original cohort still active
- M0: Always 100%
- Diagonal: Most recent data
Key insights:
- Vertical: Compare same period across cohorts (M1 improving: 65%→76%)
- Horizontal: See retention curve shape (steep drop early, then gradual)
- Trends: Draw lines through columns to see improvement/decline
Building a Retention Table (Descriptive)
Data to Collect:
- User signup dates grouped into cohorts (weekly or monthly)
- Activity data for each user in each subsequent period
- Definition of "active" (logged in, made purchase, used core feature)
- Time periods to track (typically 6-12 months)
How to Analyze:
Analysis 1: Create Retention Table
- Format: Table with cohorts as rows, time periods as columns
- Calculate: For each cohort and period, % of original cohort active
- Color coding: Use heat map (red <40%, orange 40-60%, yellow 60-75%, green >75%)
- Mark: Incomplete data (recent cohorts) clearly
Analysis 2: Cohort Retention Curves
- Chart type: Line chart
- X-axis: Time periods (M0, M1, M2, etc.)
- Y-axis: Retention rate (%)
- Lines: One line per cohort
- Look for: Separation between cohorts, curve shapes
Analysis 3: Period-Specific Trends
- Chart type: Line chart showing trends over time
- X-axis: Cohort (chronological)
- Y-axis: Retention rate
- Lines: Separate line for M1, M2, M3 retention
- Look for: Improvement or decline in specific periods
What to Look For:
Good patterns:
- Recent cohorts performing better than old (product improving)
- Curves plateau after M3-M6 (stable long-term retention)
- M1 retention >40%, M3 retention >30%
Bad patterns:
- Recent cohorts worse than old (product degrading)
- No plateau, continuous decline (no loyal base forming)
- Steep early drop that never recovers
Key Metrics from Cohort Analysis
D1/D7/D30 Retention
Definition: % active on specific day after signup
Milestones:
- D1 (Next Day): Immediate value delivery (target >40%)
- D7 (Week 1): Habit formation (target >30%)
- D30 (Month 1): Product stickiness (target >20%)
Analysis:
- Compare D1/D7/D30 across cohorts
- Track trends over time
- Identify which milestone is weakening
Retention Curve Shape Analysis
Question: How does retention decay over time?
Patterns:
- Smiling curve: Dip then recovery (common in B2B, slow onboarding)
- Flat after dip: Initial drop then stable (good pattern)
- Continuous decline: No stickiness, churn risk
- Plateau: Healthy engaged user base
Cohort Lifetime Value (LTV)
Calculate: Revenue per cohort over lifetime
Analysis:
- Compare LTV across cohorts (improving or declining?)
- Calculate time to 80% of LTV (payback period)
- Segment LTV (by channel, segment, etc.)
Use for: CAC targets, pricing strategy, prioritization
Advanced Techniques
Leading Indicator Analysis
Method: Identify early behaviors predicting long-term retention
Process:
- For mature cohorts, identify who retained long-term
- Look back at their Week 1 behavior
- Find patterns predicting retention
Example findings:
- Users with 3+ sessions in Week 1: 75% retained at M6
- Users who invited teammate: 82% retained at M6
- Users with <3 sessions: 15% retained at M6
Application: Track leading indicators in new cohorts for early warning
Segment Migration Analysis
Track: Movement between segments over time
Example:
- 80% of power users stay power users (good)
- 15% of power users become casual (warning)
- 20% of casual users become power (opportunity)
Insight: Understand what causes upgrades/downgrades
Cohort Comparison
Method: Compare two cohorts directly
Use for:
- A/B test impact evaluation
- Channel quality comparison
- Feature launch impact assessment
Analysis:
- Side-by-side retention curves
- Statistical significance testing
- Quantify difference magnitude
Common Pitfalls & Solutions
Pitfall 1: Incomplete Cohorts
- Problem: Recent cohorts have limited data
- Solution: Mark incomplete data clearly, focus on same maturity points
Pitfall 2: Cohort Size Variation
- Problem: Different sized cohorts make comparison hard
- Solution: Always use percentages, not absolutes
Pitfall 3: Seasonality Confusion
- Problem: Seasonal patterns mistaken for cohort differences
- Solution: Compare to same period prior year, adjust for seasonality
Pitfall 4: Definition Changes
- Problem: Changing "active" definition mid-analysis
- Solution: Lock definitions, clearly mark any changes
Pitfall 5: Cherry-Picking Metrics
- Problem: Only showing metrics that look good
- Solution: Report full retention curve, be transparent
Analysis Templates
Template: Cohort Retention Report
Structure:
- Executive Summary (trends in 2-3 sentences)
- Cohort Retention Table (with color coding)
- Key Metrics (D1/D7/D30 trends)
- Cohort Comparison (best vs worst, why)
- Analysis (what's working, what's not, early warnings)
- Recommendations (actions with owners)
Template: Cohort LTV Analysis
Structure:
- Summary table (cohort, size, age, actual LTV, projected LTV, maturity)
- LTV trends (by vintage)
- Time to 80% LTV
- LTV components (ARPU trend, lifetime trend)
- Recommendations
Validation Checklist
Data Quality:
- Cohorts defined consistently
- "Active" definition clear and constant
- Tracking verified accurate
- No data gaps in period
Analysis Quality:
- Incomplete cohorts marked clearly
- Appropriate time horizons shown
- Statistical significance noted
- Comparisons are fair (same maturity)
Presentation:
- Retention table easy to read
- Color coding helpful
- Key insights called out
- Trends clearly visible
Actionability:
- "So What?" is clear
- Recommendations specific
- Next steps have owners
Related Agents & Skills
Prerequisites:
- 📊 KPI Definition Specialist - Define retention properly
- 🌳 KPI Tree Architect - Retention in broader context
Use together:
- 👥 Segmentation Expert - Segment within cohorts
- 🎯 Mix Effect Analyzer - Separate cohort from performance
- 📋 Analysis Planner - Plan cohort analysis properly
For insights:
- 🔍 Root Cause Investigator - Explain cohort differences
- 📝 Executive Summary Writer - Communicate findings
- 📊 Chart & Visualization Advisor - Create cohort visuals
Success Metrics
Mastered cohort analysis when:
- ✅ Can build retention table in <30 minutes
- ✅ Identify trends and outliers immediately
- ✅ Predict future performance from early cohorts
- ✅ Retention analysis drives product decisions
- ✅ Early warning system catches issues proactively
Key Principles
From "The Power of Analytics":
- "Cohorts help you understand if improvements are real or just mix effects"
- "Track cohorts to see product evolution impact over time"
- "Early cohort behavior predicts long-term retention"
- "Cohort analysis is essential for comparing apples to apples"
See references/retention_table_guide.md for detailed table creation See references/cohort_metrics.md for complete metrics catalog