Turning data into guardrails - decoding financial vulnerability through behavioural signs

Abstract

This report presents the work I did during my internship at Hongkong and Shanghai Banking Corporation (HSBC), Kolkata. As a financial institution, the strength of the bank is fundamentally rooted in the behavior and reliability of its customers. Understanding this behavior is not only desirable; it is essential for the security, risk mitigation and future strategic planning of the bank. To do this, banks must invest in a thorough analysis of the financial behavior of their customers to detect early signs of risk and act accordingly. I worked in the Finance Support Team within the Data and Analytics division, where our focus was to build data-backed tools and insights to help identify what can cause financial vulnerability. My internship focused on feature engineering and dashboard making to support and strengthen financial decision-making frameworks. Key areas of my work included • Developing features like overdraft utilization, debt-to-income ratio, and essential spend indicators • Automating SQL pipelines to extract and aggregate large-scale banking data • Designing interactive dashboards in Looker to visualize customer behaviors (e.g., gambling, BNPL overuse) • Analyzing correlations, outlier patterns, and vulnerability signals in transac-tional datasets • Presenting a theoretical exploration of NLP embedding techniques including TF-IDF, CBOW, Skip-Gram, and LSTM The project blended technical depth with domain knowledge to build insights from financial data, aiming to identify early risk markers and guide responsible lending. The internship not only enhanced my analytical and engineering skills, but also offered a valuable experience in solving real-world problems collaboratively within a data science team.