Subscriber Cohort Analysis — Autostay
Purpose
Analyze Autostay subscriber engagement and retention patterns by cohort to identify trends in wash usage, plan behavior, and long-term subscription health. Specialized for subscription-specific metrics including wash frequency decay, upgrade/downgrade patterns, and retention by plan type and acquisition channel.
Domain Context
Autostay — O2O 세차 구독 서비스
- 비즈니스 모델: 월간/연간 구독으로 세차 서비스 제공
- 고객: 차량 소유자 (구독자)
- 공급: 파트너 세차장 네트워크
- 핵심 지표: MRR, 구독자 수, Churn Rate, LTV, NPS
- O2O 루프: 온라인 예약 → 오프라인 세차 → 디지털 피드백 루프
How It Works
Step 1: Read and Validate Your Data
- Accept CSV, Excel, or JSON data files with subscriber cohort information
- Verify data structure: cohort identifier, time periods, subscription metrics
- Check for missing values and data quality issues
- Summarize key statistics (cohort sizes, date ranges, metrics available)
Step 2: Define Cohort Dimensions
For Autostay, cohorts should be analyzed across these dimensions:
Primary Cohort Groupings:
- Sign-up Month: When the subscriber first activated their paid plan
- Plan Type: 월간 (monthly) vs 연간 (annual) subscription
- Acquisition Channel: Organic, referral, 아파트 단지 제휴, SNS 광고, 차량 커뮤니티, etc.
Secondary Cohort Groupings:
- Region/City: Subscriber location or nearest partner wash center
- Vehicle Type: Sedan, SUV, premium vehicle
- First Wash Timing: Subscribers who completed first wash within 7 days vs later
Step 3: Generate Quantitative Analysis
Retention Analysis:
- Calculate cohort retention rates month-over-month
- Build retention curves by sign-up month cohort
- Compare retention by plan type (월간 vs 연간)
- Compare retention by acquisition channel
- Identify the "critical retention window" — the month where most churn occurs
Subscription-Specific Metrics:
- Wash Frequency Decay: Track average washes per subscriber over time since signup. Identify when usage starts declining (a leading churn indicator).
- Upgrade/Downgrade Patterns: Track plan changes by cohort — who upgrades from 월간 to 연간? Who downgrades? At what point in the lifecycle?
- Revenue Cohort Analysis: Track MRR contribution of each cohort over time
- Reactivation Rate: Among churned subscribers per cohort, what % reactivate?
Generate Python analysis scripts using pandas and numpy if requested.
Step 4: Create Visualizations
- Retention Heatmap: Cohorts (rows) vs months since signup (columns), colored by retention %
- Wash Frequency Decay Curves: Line charts showing avg washes/month by cohort over time
- Upgrade/Downgrade Flow: Sankey diagram or stacked bar showing plan transitions
- Channel Comparison: Retention curves overlaid by acquisition channel
- Plan Type Comparison: Side-by-side retention for 월간 vs 연간 subscribers
- Output as interactive charts or static images
Step 5: Identify Insights & Patterns
- Spot significant patterns:
- Which cohorts retain best? What was different about that period?
- Is there a "Month X" cliff where most churn happens?
- Do 연간 plan subscribers show different wash frequency patterns?
- Which acquisition channels produce the highest-LTV subscribers?
- Does wash frequency decay predict churn? What is the threshold?
- Are upgrade patterns correlated with specific engagement triggers?
- Highlight surprising findings and deviations
- Compare cohort performance to establish baselines
Step 6: Suggest Follow-Up Research
- Recommend qualitative research methods:
- Targeted interviews with churning subscribers from worst-performing cohorts
- Post-wash surveys with engaged vs declining-frequency subscribers
- Session replays of booking flow for low-engagement cohorts
- Win/loss analysis: Why do some subscribers upgrade while others downgrade?
- Design follow-up quantitative studies
- Suggest A/B tests or retention experiments
Usage Examples
Example 1: Upload Subscriber CSV Data
Upload subscriber_data.csv with columns: subscriber_id, signup_date,
plan_type, acquisition_channel, monthly_washes, current_status
Request: "Analyze retention patterns by signup month and plan type.
Which cohorts are churning fastest?"
Example 2: Wash Frequency Analysis
"I have monthly wash counts per subscriber from Jan-Dec 2025.
Analyze wash frequency decay curves. At what point do subscribers
start washing less, and does that predict churn?"
Example 3: Channel Performance
Upload channel_cohorts.xlsx with acquisition channel data.
Request: "Compare subscriber retention by acquisition channel.
Which channels produce the most durable subscribers?"
Key Capabilities
- Data Reading: Import CSV, Excel, JSON, SQL query results
- Retention Analysis: Calculate and visualize retention rates over time by plan type and channel
- Wash Frequency Decay: Track usage decline patterns as a leading churn indicator
- Plan Transition Tracking: Monitor upgrade, downgrade, and reactivation flows
- Cohort Comparison: Compare metrics across subscriber cohort groups
- Anomaly Detection: Flag unusual patterns or drop-offs
- Python Scripts: Generate reusable analysis code for ongoing analysis
- Visualizations: Create heatmaps, charts, and interactive dashboards
- Research Design: Suggest targeted follow-up studies and interview approaches
Tips for Best Results
- Include time dimension: Provide data across multiple months of subscriber history
- Define cohort clearly: Specify cohort grouping (signup month, plan type, channel)
- Provide context: Explain product changes, pricing changes, partner network expansions
- Multiple metrics: Include retention, wash frequency, plan type, revenue, satisfaction
- Sufficient data: At least 3-4 monthly cohorts for meaningful pattern identification
- Request specific output: Ask for visualizations, Python scripts, or research recommendations
Output Format
You'll receive:
- Data Summary: Cohort overview and data quality assessment
- Retention Analysis: Retention heatmap + curves by plan type and channel
- Subscription Metrics: Wash frequency decay curves, upgrade/downgrade patterns
- Pattern Identification: 2-3 significant insights from the data
- Research Recommendations: Specific qualitative and quantitative follow-ups
- Analysis Scripts (if requested): Python code for reproducible analysis
- Next Steps: Prioritized actions based on findings
Further Reading