Segmenting Customers of an Online Fashion Retailer

Executive Summary

Background: As part of its customer-centric approach, our online fashion retailer should understand and segment its diverse customer base to effectively channel marketing resources and increase customer retention.

Objectives: Identify distinct customer segments, develop targeted marketing strategies, A/B test the strategies (not part of this project).

Methodology: Generate random (but tilted) data, build a PostgreSQL database, plot entity-relationship diagram (ERD). Write complex queries with CTEs, segment customers according to demographics, purchasing data, membership statuses (plus, loyalty programs, newsletter) using Recency-Frequency-Monetary (RFM) and K-Means Clustering models.

Recommendations: Following customer segments were identified with actionable insights for boosting customer value and retention.

Customer Segments (Recency-Frequency-Monetary)

Segment	Description	Action
Champion	Best customers. Buys often and big. Has purchased recently.	Analyze them to better understand our product-market fit and main buying personas.
Potential Loyalist	Similar to a champion but with smaller order size and frequency.	Target them with a loyalty program.
New Customer	Recent purchase, but little to no order history.	Introduce them to novel features on our website, since they probably do not know (e.g., style advice, augmented reality).
At Risk	Low recency, occasional buying.	Create sense of urgency with a flash deal.
Other	Mid-scores, no clear insight.	Try a different model or target with low-cost, non-personalized, fail-safe methods.

Customer Segments (K-Means Clustering):

Segment	Description	Action
Youthful & Sporty	Young, Nike Loyalty, frequent but small orders.	Loyalty program increases sales, keep it up!
Quality over Quantity	Middle-aged, infrequent yet large orders, affinity to Polo Ralph Lauren.	Stick with a traditional campaign calendar (e.g., Christmas, Black Friday).
Others	No clear insight.	Try a different model or target with low-cost, non-personalized, fail-safe methods.

Folder Structure

└── 📁customer-segmentation
    └── README.md
    └── 📁code
        └── 01_create_data.ipynb
        └── 02_create_database.sql
        └── 03_segment_customers.ipynb
    └── 📁data
        └── brands.csv
        └── customer_loyalty.csv
        └── customers.csv
        └── loyalty_programs.csv
        └── order_brands.csv
        └── orders.csv
    └── 📁erd
        └── erd.json
        └── erd.png
        └── erd.sql

Code

01_create_data.ipynb

This Jupyter notebook is used for generating all the CSV files in the 📁data folder. The data are random, yet biased, such that it yields some fruitful results for customer segmentation. The data are created by the author and are not downloaded from anywhere.

02_create_database.sql

This SQL script is responsible for creating the database schema. It creates a local PostgreSQL database on user's computer. It defines the tables, relationships, and constraints necessary to store the customer, order, brand, and loyalty program data.

03_segment_customers.ipynb

This Jupyter notebook performs customer segmentation analysis. Recency-Frequency-Monetary (RFM) and K-Means Clustering were used to segment customers.

Entity-Relationship Diagram

Contact

For any questions or inquiries, please contact ekin.derdiyok@icloud.com

Reference

Zalando SE. (2017). The zTypes Mission: Breaking Down the Marketing Silos. Zalando Annual Report 2017. Retrieved from https://annual-report.zalando.com/2017/magazine/the-ztypes-mission-breaking-down-the-marketing-silos/