Bank X: AI-Driven Personal Loan Campaign Optimization

1. Executive Summary

This project outlines a strategic initiative for Bank X to increase its asset base by converting existing liability (deposit) customers into personal loan borrowers. By leveraging historical campaign data and demographic information, a classification model was developed to identify customers with the highest propensity to accept a loan offer. The final solution, utilizing Decision Tree algorithms, achieved an accuracy of approximately 98%, providing a roadmap to significantly reduce marketing costs while improving conversion rates.

2. Business Context & Problem Statement

The Challenge

Bank X possesses a growing customer base, primarily consisting of liability customers (depositors). However, the number of asset customers (borrowers) is comparatively small, limiting the bank's profitability potential.

• Inefficient Marketing: Previous campaigns were broad and untargeted, resulting in a conversion rate of only 9.6%.
• Customer Experience: Mass marketing to uninterested customers risks engagement fatigue and dissatisfaction.

The Objective

The department wanted to build a model that would help them identify the potential customers who have a higher probability of purchasing the loan. The goal was to increase the success ratio while at the same time reducing the cost of the campaign.

3. Technical Architecture & Methodology

Data Strategy

The project analyzed a dataset of 5,000 customers, including demographic details (Age, Income, Family Size, Education) and banking relationships (Mortgage, Securities Account, CD Account, Online Banking, Credit Card).

• Data Cleaning: Identifiers such as ID and ZIP Code were removed as they provided no predictive power.
• Anomaly Handling: The dataset contained negative values in the Experience column (professional experience), which were corrected during the preprocessing phase.
• Exploratory Data Analysis (EDA): Analysis revealed that customers with higher income and higher education levels were significantly more likely to accept a personal loan.

Model Development Pipeline

Four distinct classification algorithms were implemented and compared to determine the best fit for this specific marketing problem:

1. Logistic Regression: Used as a baseline to understand linear relationships between features and loan acceptance.
2. K-Nearest Neighbors (KNN): Utilized to classify customers based on similarity to previous loan acceptors.
3. Naive Bayes: Applied to test probabilistic prediction capabilities.
4. Decision Trees (CART): Implemented to capture non-linear decision boundaries and provide interpretable rules for marketing teams.

Evaluation Metrics

The models were evaluated using a Confusion Matrix, assessing Accuracy, Recall, Precision, and F1-Score. Recall was prioritized to ensure the bank did not miss potential customers (False Negatives), while Precision was monitored to minimize wasted marketing spend (False Positives).

4. Key Results and Evaluation

Performance Highlights

• Decision Tree Superiority: The Decision Tree model outperformed other algorithms, achieving an accuracy of approximately 98% on both training and test sets.
• Key Drivers: The analysis identified Income, Education, and Family Size as the most significant predictors of loan acceptance.
• Comparative Analysis: While Naive Bayes performed reasonably well, it lagged behind the Decision Tree in overall accuracy and precision.

Strategic Insights

• Target Demographics: Customers with an annual income greater than $100k and those with Advanced/Professional degrees showed the highest acceptance rates.
• Account Holders: Customers who already hold CD accounts or Securities accounts were identified as high-potential targets.

5. Recommendations & Future Scope

To maximize the ROI of future campaigns, Bank X should adopt the following strategies based on the model's findings:

1. Segmented Targeting: Focus marketing budget strictly on high-income, highly educated segments identified by the Decision Tree rules.
2. Cross-Selling: Utilize the correlation between CD Account holders and loan acceptance to create bundled offers.
3. Digital Engagement: As internet banking usage is captured in the data, digital channels should be prioritized for tech-savvy high-income users.