Artificial Intelligence in Financial Fraud Detection: A Conceptual Literature Review

Résumé

Artificial intelligence (AI) is increasingly applied to financial fraud detection in response to the growing complexity of fraudulent schemes and the exponential increase in data volumes. Despite abundant evidence of AI's technical performance, the literature has not yet proposed a model that simultaneously integrates its technological, organizational, and institutional dimensions: this is the gap the present theoretical research addresses. Based on an integrative review drawing mainly on studies published between 2009 and 2023 in auditing, information systems, and AI governance research, this strictly conceptual study draws on Agency Theory, the Resource-Based View, Institutional Theory, and the socio-technical systems approach to propose a model articulating five variables: AI adoption, data quality, institutional governance maturity, human-AI complementarity, and fraud detection performance. Four hypotheses posit that AI adoption influences detection performance both directly and indirectly, through two mediating mechanisms (data quality and human-AI complementarity), moderated by governance maturity. This model gives rise to a "Conditional Transformation Theory," in which each of these four hypotheses illustrates one condition of AI's partial, rather than revolutionary, transformation of financial control, contingent on the co-evolution of technological and institutional capabilities. Applied to the Moroccan context, viewed as an institutional transition laboratory, the study outlines directions for future empirical validation.

Classification JEL: M42, G28, O33

Paper type: Theoretical Research