← Back to Home
BEHAVIORAL ANALYTICS • CASE STUDY
Cohort Analysis & Customer Retention Dynamics
1. Dataset, Scope & The "Leaky Bucket" Problem
Project Type
Personal Project / Simulated Business Case
Dataset Source
Simulated E-Commerce CSVs (
EcomSales, Customer, Product, Region)
Scale & Anchoring
51,290 line-items (25,728 orders) across 17,415 customers; 2022 cohorts anchored against 2020–2023 history
Time Range & Currency
Jan 2020 – Dec 2023 (4 Years) • Simulated USD ($)
Despite steady acquisition volumes, revenue fluctuated unpredictably—suggesting a "leaky bucket" dynamic where newly acquired customers churned prematurely before returning positive ROI.
- Build a granular Time-Based Cohort model (Acquisition Month vs. Elapsed Months).
- Segment retention performance to identify structural differences between Corporate (B2B) and Consumer (B2C) units.
- Evaluate the Scale vs. Efficiency trade-off across customer income brackets.
- Audit and resolve data engineering traps including left-truncation bias in cohort tracking.
2. Key Findings & Behavioral Discoveries
- The Month-1 Churn Cliff: Across 2022 cohorts, Month-1 repeat purchase rates drop sharply to a 1% to 3% baseline, highlighting an early post-purchase retention gap to test via lifecycle onboarding (clickstream engagement itself is not directly measured in order logs).
- Corporate (B2B) Restock Cyclicality (Sample Caveat): Unlike flatter B2C buying, the Corporate segment shows observable retention bumps (+5% to 6%) around Month 5 and Month 8; because the B2B cohort base is smaller than B2C, persistence should be monitored across future periods.
- The Income Profitability Paradox:
- Low-Income Cohort (The Cash Cow): Generated lowest AOV per transaction, but their high volume and consistency produced the largest cumulative profit pool.
- High-Income Cohort (The Cash Grab): Highest profit margin per single order, but had the worst repurchase retention rates.
3. Methodology & Analysis Fixes
- Eliminated Left-Truncation Bias: Identified that computing
FirstPurchaseDate*after* subsetting 2022 data falsely misclassified returning 2020–2021 buyers as Month-0 acquisitions. Fixed by anchoring first-purchase dates globally across the unpartitioned master table. - Cohort Size Weighting Layout: Paired absolute acquisition volume bar charts side-by-side with retention heatmaps using
gridspec_kw={'width_ratios': [1, 4]}to ensure small cohorts were not over-interpreted. - Execution Scope Consolidation: Merged disparate plotting cells into a single Object-Oriented pipeline, resolving broken Matplotlib axes across subplots.
- Consolidated Master Ingestion: Refactored repeated relational joins into an upstream unified
df_mastertable to optimize memory overhead.
4. Proposed Retention Experiments
- Automated Post-Purchase Nurture: Pilot an automated re-engagement or "Subscribe & Save" sequence triggering at Day 20–25 to address the observed Month-1 drop-off cliff.
- B2B Restock CRM Alerts: Schedule outbound prompts for corporate account managers at Month 4 and Month 7 ahead of the observed Month 5 / Month 8 B2B restock cycles.
- First-Session Cart Upsells: Test first-order bundle incentives for high-income buyers where single-purchase margin is high but repeat frequency is low.
Measurement Framework: Evaluate via cohort holdout groups tracking Month-1 & Month-3 Retention Rate, B2B Reorder Conversion, and Contribution Margin After Discount.