Operationalizing Critical Data Approaches in Fraud Detection in Latin America

Abstract

This study investigates the challenges and sociotechnical dimensions of cross-border AI deployment, focusing on fraud detection products entering Latin American markets. Through semi-structured interviews with 18 industry practitioners, we examine how AI systems trained primarily on Global North data can reproduce and amplify existing inequalities when deployed in diverse cultural contexts from the one it was originally designed. By analyzing these dynamics through lenses of algorithmic fairness, critical data studies, and science and technology studies, we identify three key sociotechnical challenges: technical-cultural misalignment, socioeconomic stratification effects, and problematic model behavior assumptions. Our findings reveal how standardized approaches to fraud detection can systematically disadvantage certain populations, particularly those with limited digital footprints. We translate these theoretical insights into practical recommendations for industry practitioners, contributing to ongoing conversations about responsible AI deployment across diverse contexts. This work bridges the gap between abstract fairness principles and concrete implementation challenges, advocating for a contextual approach to AI governance that prioritizes local knowledge and inclusive design.