Model for Handling False Positive Alerts in Credit Card Transactions Qorus-Infosys Finacle Banking Innovation Awards 2025

Concept and Objectives: The project involves the development of artificial intelligence (AI) models for the automated handling of fraud alerts in credit card transactions. The main goal is to identify false positives — legitimate transactions declined by security mechanisms — and accelerate card unblocking, improving the customer experience and optimizing operational resources. Reasons Behind: With over 5 billion credit card transactions in Brazil in just the second quarter of 2024, the volume of alerts generated by anti-fraud systems is significant. Many of these alerts involve legitimate transactions that are declined as a precaution. Manual contact with customers is not always successful, leading to delays and dissatisfaction. AI was adopted to make this process faster, more accurate, and more efficient.

State of Competition: Although the use of AI in anti-fraud systems is a growing trend in the financial sector, Banco do Brasil stands out for its ability to integrate large volumes of sensitive data, automate the handling of alerts, and feed insights back into authorization models — significantly enhancing the effectiveness of the solution. Sources of Inspiration: The inspiration came from the need to improve the customer journey and from observing market practices in tech companies and fintechs that use AI for real-time decision-making. The approach was also influenced by academic studies on machine learning applied to anomaly detection. Departments Involved: The project involved several areas of Banco do Brasil, including: • Data and Analytics Teams • Information Security • Customer Service • Information Technology • Card Business Units Main Results So Far: • Significant reduction in card unblocking time • Potential release of up to R$74 million per month in legitimate transactions • Workforce optimization, with better allocation of up to 13 customer service agents per day • Positive impact on approximately 12,000 customers per month • Development of a second model to identify merchants complicit in fraud, enhancing overall system security