I AM
HARENDRA

Problem Statement

Financial institutions needed a systematic way to identify loan default patterns and borrower risk profiles to improve credit decision-making and reduce default rates.

Data Source & Scale

148,671+ loan records encompassing approval patterns, borrower demographics, loan purposes, regional data, and repayment behavior.

Methodology

• Performed extensive data cleaning including null handling, feature selection, and data transformation to ensure dataset reliability.
• Built an interactive Power BI dashboard with 10+ KPIs and custom DAX measures for dynamic risk analysis.
• Applied segmentation analysis to identify high-risk borrower groups by loan purpose, geography, and demographic factors.

Tools Used

Key Results

Impact

• Improved dataset reliability and analytical accuracy by 30% through systematic data cleaning.
• Identified high-risk segments (loan purpose p4, North/South regions), delivering insights that can potentially reduce loan defaults by 20–25% through targeted policy improvements.
• Reduced manual reporting effort by 40% with automated interactive dashboards.

Problem Statement

A retail business needed to identify at-risk customers and segment its customer base to enable targeted retention strategies and optimize lifecycle marketing investments.

Data Source & Scale

1,000,000+ retail transaction records covering customer purchase history, frequencies, monetary values, and behavioral patterns.

Methodology

• Engineered customer-level features including Recency, Frequency, Monetary value, AOV, purchase intervals, and return rates.
• Built an RFM-based segmentation model classifying customers into 5 actionable segments: Champions, Loyal, At-Risk, Lost, and New.
• Developed and compared Logistic Regression and Random Forest models for churn prediction, selecting Random Forest based on superior performance metrics.

Tools Used

Key Results

Impact

• Uncovered that the top 10% of customers contributed a disproportionate share of total revenue, informing a high-value retention strategy.
• Enabled targeted lifecycle marketing interventions through 5-segment RFM classification.
• Recommended Random Forest model for deployment based on superior F1-score and ROC-AUC performance over Logistic Regression.

Problem Statement

A Direct-to-Consumer (D2C) e-commerce brand needed a comprehensive customer intelligence system to understand customer behavior, identify at-risk segments, and generate actionable insights for data-driven marketing and retention strategies.

Data Source & Scale

Transactional dataset containing customer purchase history, demographics, and behavioral patterns from a D2C e-commerce platform — cleaned and prepared as a team deliverable.

Methodology

• Performed extensive Exploratory Data Analysis (EDA) to uncover purchasing patterns, seasonal trends, and customer demographics using Python (Pandas, Matplotlib, Seaborn).
• Built RFM-based customer segmentation and machine learning models for churn prediction within a Jupyter Notebook pipeline.
• Wrote SQL scripts for structured querying, data aggregation, and business KPI extraction from the transactional dataset.
• Designed an interactive Power BI dashboard to visualize customer segments, revenue trends, and key performance indicators.
• Developed a GenAI-powered component for natural language insights generation and conversational data exploration.

Tools Used

Key Results

Impact

• Delivered a complete customer intelligence system combining EDA, segmentation, ML prediction, and dashboard visualization as part of the EY Capstone programme.
• Enabled natural language exploration of customer data through a GenAI-based conversational interface, making insights accessible to non-technical stakeholders.
• Produced a comprehensive project report and interactive Power BI dashboard as final deliverables for industry evaluation under EY Learning Solutions.