RFM Segmentation for Subscriber Re-Engagement

Project Summary

This project applies RFM (Recency, Frequency, Monetary) clustering to segment over 50,000 Quistic email subscribers based on engagement behavior. The goal was to identify dormant but high-value users for targeted re-engagement and to improve overall list health through strategic pruning. The resulting campaign changes led to a measurable increase in clickthrough rates.

Objectives

• Segment subscribers by behavioral patterns using RFM methodology
• Identify high-value users with declining engagement
• Enable targeted re-engagement campaigns and list hygiene strategies
• Measure the impact of these actions on clickthrough rate (CTR)

Data

• Source: Internal email engagement data from Quistic campaigns
• Scope: 50,000+ subscribers
• Features included:
  • Last engagement timestamp (Recency)
  • Number of opens/clicks over past 180 days (Frequency)
  • Revenue (Monetary)

Methods

• Data cleaning and feature engineering with pandas
• RFM score generation (all dimensions normalized)
• K-Means clustering with scikit-learn
• Cluster evaluation using silhouette score
• Cluster visualization (2D and 3D plots with matplotlib / seaborn)
• Interpretation of segments for business actionability

Key Segments Identified

• Loyal Engagers: High frequency + recent + purchases
• Dormant Buyers: Low recency, but past purchase activity
• Cold Leads: Low across all RFM dimensions

Actions Taken

• Custom re-engagement campaigns for “Dormant Buyers”
• Removal or suppression of low-engagement, non-paying users
• Adjustments to email frequency and content personalization

Results

• CTR increase: +15%
• Improved email deliverability and engagement consistency
• Created an internal playbook for ongoing segmentation strategy

Tools Used

• Python (pandas, scikit-learn, matplotlib, seaborn)
• Jupyter Notebooks

Files Included

• rfm_segmentation.ipynb – data cleaning, scoring, clustering
• cluster_analysis.png – visuals of key segment distributions colored by cluster
• 2D_plots.png – pairwise plots of RFM features colored by cluster
• rfm_segmentation_readout.pdf – full visual of processing and analysis viewable by CEO
• README.md – this document

Michael Gadhia
Data Analyst | Boston, MA
michaelmgadhia@proton.me