Bank-Customer-Segmentation

AllLife Bank Customer Segmentation

Business Case

AllLife Bank aims to grow its credit card customer base by improving market penetration.

• Marketing Team: Run personalized campaigns for new and existing customers.
• Operations Team: Enhance customer support by upgrading the service model.
• Data Science Team (this project): Deliver actionable customer segmentation using clustering algorithms to guide both marketing and operations strategies.

Objective

To identify distinct customer segments based on spending patterns and past interactions with the bank, and provide business recommendations that improve targeting, service efficiency, and customer satisfaction.

Data Profile

• Features:
  • Avg_Credit_Limit
  • Total_Credit_Cards
  • Total_visits_bank
  • Total_visits_online
  • Total_calls_made
• Observations: ~660 customers
• Nature: Mix of continuous (credit limit) and discrete count features (cards, visits, calls).

Data Cleaning & Preparation

• Removed irrelevant identifiers (Sl_No, Customer Key).
• Handled skewness in Avg_Credit_Limit via log transformation.
• Applied StandardScaler to normalize all features for fair contribution in clustering.
• Verified distributions with histograms and boxplots.

Exploratory Data Analysis (EDA)

• Correlation Heatmap: Identified strong relationships (e.g., credit limit ↔ number of cards).
• Pairplots: Visualized feature interactions and outliers.
• Boxplots per cluster: Validated spread and cluster-specific behavior.

Dimensionality Reduction

• Applied t-SNE (2D & 3D) to visualize high-dimensional data.
• Experimented with multiple perplexity values, finalized perplexity = 50 for optimal separation.
• Clear visual separation of clusters observed.

Modelling – K-Means Clustering

• Used Elbow Method to identify optimal k.
• Validated with Silhouette Score for cluster quality.
• Final model chosen with k = 3 clusters.

Cluster Profiling

Cluster 0 – Mid-tier Branch Loyalists (≈58.5%)

• Moderate credit limits (~₹33k), ~5–6 cards.
• High branch visits, low online usage.
• Recommendation: Promote digital adoption, hybrid campaigns, branch service bundles.

Cluster 1 – Mass Market Support-Heavy (≈34%)

• Lowest credit limits (~₹12k), ~2–3 cards.
• Moderate online usage, very high call dependency.
• Recommendation: Improve call center efficiency, credit-builder programs, incentivize app usage.

Cluster 2 – Digital Elites (≈7.5%)

• Very high credit limits (~₹141k), ~9 cards.
• High online usage, minimal branch visits and calls.
• Recommendation: Premium digital perks, loyalty rewards, personalized dashboards.

Approach Followed:

1. Data Cleaning & Transformation
2. EDA & Visualization
3. Feature Scaling
4. Dimensionality Reduction (t-SNE)
5. K-Means Modelling
6. Cluster Validation (Elbow + Silhouette)
7. Cluster Profiling & Business Recommendations

Tools & Environment

• IDE: Google Colab Jupyter Notebook
• Languages: Python (pandas, numpy, scikit-learn, seaborn, matplotlib), t-SNE
• Visualization: t-SNE, scatterplots, histograms, boxplots, heatmaps, pairplots
• Clustering: K-Means, silhouette analysis

Impact

This segmentation empowers AllLife Bank to:

• Marketing: Deliver personalized campaigns, increase penetration, and upsell products.
• Operations: Reduce support load, improve customer satisfaction, and optimize service channels.
• Strategy: Focus resources on high-value digital elites while nurturing mid-tier and support-heavy customers.

Conclusion

This project demonstrates industry-ready data science capabilities with business approach through:

• Rigorous preprocessing and scaling.
• Robust clustering validation.
• Clear visualization and profiling.
• Actionable business recommendations.

I approached this as a confident data professional delivering insights that directly drive business growth.