Author: Olga Seymour

Date: May 2025

GitHub: https://github.com/AI-Data-Space/BankingRetentionOptimizer

Customer segmentation analysis showing distribution across digital transformation readiness and churn risk dimensions

Note: Visualizations and charts are generated when running the notebooks and will appear in the reports/figures/ directory.

Atlantic Regional Bank (fictional case study) is undergoing a major digital transformation, shifting resources from physical branches to digital channels. The bank must simultaneously reduce branch operating costs while preventing customer attrition during this transition. With a limited retention budget of $1.5 million, the bank needs to identify which customers are both at high risk of leaving and most likely to respond positively to retention efforts. Traditional retention approaches that target all high-risk customers are not financially viable during this transformation period, requiring a more sophisticated, ROI-focused approach.

This project uses the Churn Modelling Dataset originally from Kaggle, which contains customer information from a European bank including demographics, account details, and churn indicators.

Original Source: Churn Modelling Dataset on Kaggle

Note: The dataset is included in this repository (data/Churn_Modelling.csv) for convenience and reproducibility.

The dataset includes 10,000 customer records with features such as:

• Customer demographics (age, geography, gender)
• Account information (balance, tenure, number of products)
• Engagement metrics (active membership status, credit score)
• Target variable (customer exit/churn indicator)

While the original dataset provides standard banking customer attributes, this project transforms it into a digital transformation context through custom feature engineering and business application development.

While customer churn prediction is a well-established field, this project introduces several novel approaches specifically designed for banking digital transformation scenarios:

Rather than relying on standard demographic and transactional features, we developed banking-specific indicators that directly address digital transformation challenges:

• Digital_Readiness: Quantifies customer adaptation potential to digital channels using age-based technology comfort combined with current engagement patterns
• Retention_Score: Weighted loyalty index combining relationship depth, product diversification, and active usage patterns
• High_Risk_Age: Identifies the 40-60 demographic bracket showing pronounced digital hesitancy during transformation

Innovation: Combines departure risk with digital adoption capability in a matrix approach, moving beyond traditional risk-only segmentation.

Business Impact: Enables targeted retention strategies that address both customer departure prevention and successful digital transformation—solving two business challenges simultaneously.

Applied Innovation: Transforms probability thresholds into a "financial dial" that directly controls retention budget allocation. Strategic Insight: Analysis reveals that blanket retention approaches generate negative ROI (0.87x overall), but selective targeting of profitable segments can achieve 1.25x-1.45x returns, fundamentally changing retention strategy from broad-based to precision-focused.

Business Impact: Provides marketing teams with precise control over spending levels while maintaining optimal precision-recall balance for different budget scenarios.

Business Innovation: Segment-specific intervention costs reflecting the varying complexity of digital adoption support, from automated digital-ready customers to high-touch branch-dependent transitions.

Business Impact: Realistic ROI calculations that account for the operational complexity of helping different customer types successfully navigate digital transformation.

This approach directly addresses the strategic challenge facing banks today: how to retain customers while transforming operations. The solution framework is immediately applicable to any financial institution undergoing digital transformation with constrained retention budgets.

• Key Results Summary

• Machine Learning Model: Random Forest with 59.4% precision, 68.1% recall optimization
• Advanced Feature Engineering: Custom ML features for digital transformation context
• Predictive Analytics: 68.1% of at-risk customers identified through ML-driven segmentation
• Strategic Segmentation: $219,702 net benefit with 0.87x overall ROI, identifying profitable segments (1.45x ROI) vs unprofitable segments (-0.29x ROI)
• Implementation Ready: ML pipeline with A/B testing framework

This project delivers a complete customer retention solution that enables banks to:

• Predict which customers are likely to leave with 59.4% precision
• Identify different customer segments requiring tailored retention approaches
• Allocate limited retention budgets strategically (0.87x overall ROI with profitable segments reaching 1.45x ROI)
• Implement a phased transition strategy supporting the shift to digital channels
• Continuously refine retention strategies through A/B testing

1. Banking Executives: Gain a strategic framework for managing digital transformation with reduced customer attrition and optimized spending.

2. Marketing Teams: Receive actionable customer segmentation with targeted retention strategies rather than one-size-fits-all approaches.

3. Digital Transformation Leaders: Obtain a data-driven roadmap for transitioning customers from branches to digital channels with minimal attrition.

4. Customer Experience Teams: Receive insights to develop segment-specific experiences that address the unique needs of different customer groups.

5. Finance Departments: Benefit from quantifiable ROI projections and budget optimization for retention initiatives.

• Predictive model identifying customers at risk of leaving (59.4% precision, 68.1% recall)
• Customer segmentation with tailored retention strategies for digital-ready vs. branch-dependent customers
• ROI analysis revealing 0.87x overall return with profitable segments (1.45x) and unprofitable segments (-0.29x)
• Phased implementation plan for retention programs

The implementation of this solution provides:

• Financial Impact: $219,702 net benefit with strategic insights on profitable vs unprofitable customer segments
• Operational Efficiency: 59.4% targeting precision means reduced wasted resources
• Customer Retention: 68.1% of at-risk customers identified for proactive intervention
• Digital Adoption: Strategies for accelerating digital engagement across all customer segments
• Risk Mitigation: Early identification of high-value customers at risk during transformation
• Strategic Intelligence: Identification of profitable segments (1.45x ROI) versus unprofitable segments (-0.29x ROI)

This project followed a systematic approach:

1. Data Analysis & Feature Engineering: Understanding churn patterns and creating banking-specific predictive features
2. Model Development & Selection: Testing and optimizing machine learning models with focus on business metrics
3. Business Application: Developing targeted retention strategies with ROI assessment
4. Implementation Framework: Creating a practical roadmap with A/B testing approach

banking-retention-optimizer/
├── data/
│   └── Churn_Modelling.csv
├── notebooks/
│   ├── 1_data-analysis.ipynb              # Data exploration & visualization
│   ├── 2_model-development.ipynb          # Model selection and optimization
│   └── 3_business-application.ipynb       # Segmentation, strategies & ROI
├── reports/
│   ├── figures/                           # Visualizations
│   └── model_comparison.csv               # Model performance metrics                           
├── .gitignore
└── requirements.txt                       # Dependencies
└── README.md                              # Project documentation 

This notebook explores the customer dataset to understand key factors influencing churn during digital transformation:

• Identified significant age difference between churning (older) and retained customers
• Discovered inactive members are 3x more likely to leave
• Found geography plays a meaningful role in retention
• Revealed that customers with multiple products have lower churn rates
• Analyzed balance patterns showing zero-balance accounts have higher churn risk

This notebook focuses on developing predictive models for customer churn:

• Created custom features for digital transformation context (Digital_Readiness, Retention_Score)
• Implemented feature engineering pipeline with appropriate scaling and encoding
• Evaluated multiple algorithms (Random Forest, AdaBoost, XGBoost)
• Conducted hyperparameter tuning with F1 score optimization
• Performed detailed threshold analysis to balance precision and recall for budget constraints
• Selected Random Forest (Optimized) as best model with F1 score of 0.635

This notebook applies the model to create actionable business strategies:

• Developed customer segmentation combining churn risk and digital readiness
• Created tailored retention strategies for each segment
• Calculated ROI for retention investments by segment
• Designed a three-phase implementation roadmap
• Established an A/B testing framework for ongoing optimization

The project evaluated multiple algorithms, with the following performance on the test set:

Random Forest (Optimized) provided the best balance of precision and recall, which is critical for our limited budget scenario. The model achieves 59.4% precision (efficiency of retention spending) while capturing 68.1% of customers who would churn (reach of retention efforts).

The choice of machine learning algorithm directly determines retention budget efficiency and revenue protection:

Why Random Forest (Optimized) Solves Our Business Problem:

• AdaBoost Alternative: 76.4% precision would minimize wasted spending but 45.2% recall means missing 55% of departing customers
  • Business Impact: Save budget but lose significant customer value
• High Recall Alternative: Would catch more departing customers but exhaust budget on false positives
  • Business Impact: Protect more customers but waste retention resources
• Our Random Forest Solution: 59.4% precision + 68.1% recall = optimal ROI
  • Business Impact: Identify 68% of at-risk customers while maintaining budget efficiency

Result: Our systematic algorithm comparison identified the approach that maximizes net business value (revenue protected minus budget spent) rather than optimizing for statistical accuracy alone.

• Cross-validation: 5-fold CV with stratified sampling to ensure robust performance estimates
• Business Metric Optimization: Model selection based on F1 score to balance precision-recall for budget constraints
• Threshold Analysis: Systematic evaluation of decision boundaries for optimal business outcomes

1. Age is a critical factor: Customers 40-60 years old show significantly higher churn probability
2. Digital readiness varies: Customers with low digital readiness need specialized transition support
3. Value-based segmentation is essential: High-value customers require different retention approaches
4. Product holding matters: Multiple products significantly reduce churn risk
5. Targeted strategies outperform: ROI varies significantly across segments

The retention program analysis reveals strategic insights:

• Total Program Cost: $251,950
• Value Protected: $471,652
• Net Benefit: $219,702
• Return on Investment: 0.87x overall (1.45x for best-performing segments)

Key Strategic Value: Rather than generating uniform high returns, this analysis identifies which customer segments are profitable to target (Digital-Ready customers: 1.25x-1.45x ROI) versus which segments consistently lose money (Branch-Dependent customers: -0.29x ROI). This selective targeting approach prevents costly blanket retention spending while maximizing returns on profitable segments.

The retention strategy is designed for phased implementation:

1. Phase 1: High-Risk Intervention (Weeks 1-2)

   • Immediate focus on high-value at-risk customers
   • Personal outreach and premium offers

2. Phase 2: Proactive Engagement (Weeks 3-8)

   • Medium-risk customer engagement
   • Relationship reviews and cross-selling

3. Phase 3: Value Growth (Month 3+)

   • Long-term loyalty development
   • Referral programs and satisfaction monitoring

To continuously optimize retention efforts, the project includes an A/B testing framework:

• Advanced Machine Learning: Custom Random Forest optimization with hyperparameter tuning across multiple algorithms (Random Forest, XGBoost, AdaBoost)
• ML Feature Engineering: Created domain-specific features (Digital_Readiness, Retention_Score) using advanced feature construction techniques
• Ensemble Model Comparison: Systematic evaluation of multiple ML algorithms with cross-validation and business metric optimization
• ML Pipeline Development: End-to-end scikit-learn pipeline with custom transformers for preprocessing and feature engineering
• Predictive Model Deployment: Threshold optimization and model serialization for business application

• Python: Primary programming language
• Pandas/NumPy: Data manipulation and analysis
• Scikit-learn: Machine learning models and evaluation
• XGBoost: Gradient boosting implementation
• Matplotlib/Seaborn: Data visualization
• Jupyter Notebooks: Development environment
• Joblib: Model serialization and deployment

• Python 3.8+
• Jupyter Notebook

git clone https://github.com/yourusername/banking-retention-optimizer
cd banking-retention-optimizer
pip install -r requirements.txt

1. Start with notebooks/1_data-analysis.ipynb for data exploration
2. Continue with notebooks/2_model_development.ipynb for model training
3. Finish with notebooks/3_business_application.ipynb for business strategy implementation

• Training Data Limitation: Model trained on pre-transformation data; customer behavior may evolve as digital adoption increases
• Static Feature Engineering: Digital_Readiness metric may need recalibration as digital services expand
• External Factor Exclusion: Economic conditions, competitor actions, and market changes not captured

• Performance Tracking: Monitor precision/recall monthly during transformation
• Feature Drift Detection: Track changes in Digital_Readiness patterns as transformation progresses
• Dynamic Threshold Adjustment: Adapt probability thresholds based on budget availability and observed effectiveness

• Real-time digital engagement monitoring integration
• Customer lifetime value (CLV) incorporation for enhanced ROI calculations
• Conversational AI for personalized branch-to-digital transition support
• Economic indicator integration for external factor consideration
• Advanced A/B testing framework with multi-armed bandit optimization