Artificial Intelligence and Fraud Detection in Digital Banking: Implications for Community Financial Institutions

Artificial Intelligence (AI) has proven useful for improving how financial institutions, and especially community financial institutions (CFIs), detect cyber fraud and build resilience in the U.S., but there is limited empirical research connecting the impact of AI-based fraud detection technologies with how CFIs manage cyber fraud risk. This paper aims to explore how AI-enabled fraud-detection capability (AIFDC) relates to fraud-detection effectiveness (FDE), fraud risk management (FRM), and operational resilience (OR) by analysing a quantitative cross-sectional dataset collected from a survey of 384 fraud-detection practitioners at U.S. community banks and credit unions. Partial Least Squares Structural Equation Modelling (PLS- SEM), serial mediation, PLSpredict as well as Necessary Condition Analysis (NCA) and moderation analysis were used for the data analysis. 77.3% of results were reported as ‘real-time transaction monitoring' (RTTM), and 71.9% of the results were reported as ‘anomaly detection'. There was a significant increase in the effectiveness of fraud-detection (β = 0.609, p < 0.001), fraud risk management (β = 0.647, p < 0.001) and operational resilience (β = 0.474, p < 0.001) as a result of AI capability for fraud-detection, as well as a significant serial mediation (β = 0.187, p < 0.001). The AI capability and fraud risk management (intentionally displayed by NCA) were the two items that showed a significant difference in the score of operation resilience. The augmentation phenomenon of AI usage on the score of operation resilience (β = 0.146, p = 0.001) implied that changes in the AI usage score had a stronger augmenting effect on the score of operation resilience and indicated that PLSpredict has high predictive relevance. The data suggest that AI capabilities – and AI governance – have multiple avenues to counteract fraud, helping make U.S. CFIs even more resilient.