Chapter 3: Artificial Intelligence in Financial Fraud Detection Lessons from the United States and Applications to South Africa

Abstract

Artificial intelligence (AI) has revolutionised financial fraud detection in developed markets, particularly the United States, where machine learning (ML), natural language processing (NLP), and network analytics now underpin real-time transaction monitoring, anomaly detection, and suspicious activity reporting (SAR) triage. However, the transferability of these AI systems to emerging markets like South Africa is not straightforward. Differences in data availability, fraud typologies (e.g., insider-driven vs. external fraud), regulatory frameworks, and forensic capacity create implementation barriers. This chapter systematically extracts lessons from the US experience – including the use of supervised and unsupervised learning, federated learning for privacy preservation, and explainable AI (XAI) for regulatory compliance – and evaluates their applicability to South Africa’s financial sector. Using a comparative institutional analysis, we identify three core lessons: (i) AI effectiveness depends on high-quality labelled data, which South Africa lacks for many fraud types; (ii) regulatory sandboxes can accelerate AI adoption but require trust; (iii) human-AI teaming (rather than full automation) is essential where forensic resources are scarce. We propose a phased roadmap for South African banks and the Financial Sector Conduct Authority (FSCA), including a shared fraud data consortium, open-source AI models for smaller institutions, and capacity-building partnerships with US fintechs. The chapter concludes that AI can bridge the fraud detection gap, but only if adapted to local institutional realities