Olist Customer Retention Analysis

🚧 This project is currently in progress. The analysis, modeling, and insights are being continuously refined.

Project Overview

This project analyzes the Brazilian E-Commerce Public Dataset by Olist to understand customer behavior and identify patterns related to customer churn and retention. The goal is to translate data-driven findings into actionable business insights that support customer retention strategies in e-commerce.

Project Summary

From a business perspective, a key question for e-commerce operations is: "Which customers are unlikely to return, and what behavioral patterns can indicate churn risk?"

This project follows an end-to-end data science workflow, including data cleaning, exploratory data analysis (EDA), feature engineering, and churn-oriented customer analysis. Predictive modeling is planned as a next step.

Problem: Low customer retention directly impacts revenue and long-term growth in e-commerce.
Data: Real-world Olist dataset with 100k+ orders (2016–2018), including customer, order, product, payment, delivery, and review information.
Current Approach: Exploratory Data Analysis and customer behavior profiling using RFM-style metrics (recency, frequency, monetary value).
Preliminary Insight: A large portion of customers make only a single purchase, suggesting that churn is often structural and related to product type rather than dissatisfaction.
Next Steps: Feature engineering and supervised learning models (e.g., Random Forest or XGBoost) to estimate churn risk.

Exploratory Data Analysis

The exploratory analysis focuses on understanding purchase frequency, customer recency, order value distributions, and product category behavior. Visualizations and statistical summaries are used to highlight patterns related to retention and churn.

Feature Engineering

Planned features include recency, frequency, monetary value (RFM), delivery delay indicators, review scores, and product category aggregation at the customer level.

Modeling

The modeling phase will focus on supervised classification techniques to estimate churn probability. Model performance will be evaluated using metrics such as precision, recall, F1-score, and ROC-AUC.

Insights & Recommendations

Business-oriented insights and recommendations will be derived from both exploratory analysis and model outputs, with a focus on actionable retention strategies.

Conclusion

This project aims to demonstrate how data analysis can support customer retention decisions in e-commerce. Final conclusions and results will be added as the project progresses.