End-to-end CRM analytics pipeline: RFM segmentation, churn prediction (XGBoost AUC 0.997), CLV modelling, and A/B campaign evaluation on 100K+ orders — Python, SQL, scikit-learn, XGBoost, Power BI.
Online retailers generate vast amounts of customer transaction data, yet many struggle to translate it into actionable CRM strategies. This project simulates the work of a commercial data analyst tasked with:
- Understanding who the most valuable customers are
- Identifying customers at risk of churning
- Predicting future customer lifetime value
- Evaluating the effectiveness of promotional and re-engagement campaigns
The analysis is built on 100,000+ real e-commerce orders from the Olist Brazilian E-Commerce dataset (Kaggle), covering orders, customers, products, payments, and reviews.
| Layer | Tools |
|---|---|
| Data Processing | Python (pandas, NumPy) |
| Machine Learning | scikit-learn, XGBoost |
| CLV Modelling | lifetimes (BG/NBD model) |
| Statistical Testing | scipy |
| SQL Analysis | PostgreSQL / SQLite |
| Visualisation | matplotlib, seaborn, Power BI |
| Environment | Jupyter Notebooks |
ecommerce-crm-analytics/
│
├── data/
│ ├── raw/ # Original Olist CSV files (not tracked in git)
│ └── cleaned/ # Processed master dataset
│
├── notebooks/
│ ├── 01_data_cleaning.ipynb
│ ├── 02_rfm_segmentation.ipynb
│ ├── 03_churn_prediction.ipynb
│ ├── 04_clv_modelling.ipynb
│ └── 05_campaign_evaluation.ipynb
│
├── sql/
│ ├── customer_metrics.sql
│ ├── cohort_analysis.sql
│ └── campaign_performance.sql
│
├── dashboard/ # Power BI .pbix file and screenshots
├── reports/ # Summary findings
└── README.md
- RFM Segmentation: 9 customer segments identified across 93,349 customers — Champions represent the highest avg CLV at 310 BRL, while Loyal Customers contribute the largest revenue share at 22%
- Churn Prediction: XGBoost model achieved AUC of 0.997, identifying 55,852 high-risk customers with recency as the dominant churn driver
- CLV Modelling: Total historical CLV of 15.4M BRL across all customers — top 20% of customers contribute 55% of total revenue
- Campaign Evaluation: Re-engagement campaign delivered 105.6% conversion uplift vs control group (11.9% vs 5.8%), generating 200,507 BRL additional revenue (p < 0.0001)
1. Clone the repository
git clone https://github.com/Toyor12/ecommerce-crm-analytics.git
cd ecommerce-crm-analytics2. Install dependencies
pip install -r requirements.txt3. Download the dataset
Download from Kaggle: https://www.kaggle.com/datasets/olistbr/brazilian-ecommerce
Place all CSV files into data/raw/
4. Run notebooks in order
jupyter notebookRun notebooks 01 through 05 in sequence.
| Notebook | Description |
|---|---|
| 01_data_cleaning | Load, merge, and clean all Olist tables into a master dataset |
| 02_rfm_segmentation | Build RFM scores and segment customers into CRM personas |
| 03_churn_prediction | Train and evaluate churn classification model |
| 04_clv_modelling | Calculate historical CLV and predict future customer value |
| 05_campaign_evaluation | Simulate campaign cohorts and run A/B test analysis |
The dashboard is built in Google Looker Studio and contains 4 pages:
- Customer Overview — total customers, orders, revenue, monthly trend, category breakdown
- RFM Segments — segment distribution, revenue share, avg CLV by segment
- Churn Risk — risk distribution, churn rate by segment, high-risk customer count
- Campaign Performance — A/B test results, conversion rate and revenue uplift by group
[View Live Dashboard →](https://datastudio.google.com/reporting/6689c78c-483f-4700-9f73-e4186cd20d9a
d20d9a)
Olist Brazilian E-Commerce Public Dataset Source: https://www.kaggle.com/datasets/olistbr/brazilian-ecommerce License: CC BY-NC-SA 4.0
Oyewole Oluwatoyosi — Data Analyst | Commercial & CRM Analytics
- GitHub: github.com/Toyor12
- LinkedIn: linkedin.com/in/oluwatoyosi-oyewole-3a75b2109
- Portfolio: toyor12.github.io