A conceptual ai-based adaptive model for fraud detection in international transactions: innovations, emerging risks, and applicability to Latin American SMES

Abstract

The growth of digital international transactions has increased the exposure of small and medium-sized enterprises (SMEs) to financial fraud, in a context where artificial intelligence (AI) can be used both to combat and to commit this crime. In Mexico, losses of approximately 17.4 billion pesos were reported during the first half of 2025. This paper proposes, at a conceptual level, an adaptive fraud detection model under the Software as a Service (SaaS) modality aimed at Latin American SMEs. A mixed-methods approach with an exploratory-descriptive scope was employed, combining a narrative documentary review, a descriptive analysis of aggregated operational records from a remittance organization (October 2024 to January 2025), and the design of an input, processing, and output data architecture. In the case analyzed, 103.720 calls were referred to the fraud department and only 33,53% were confirmed as fraud, which illustrates the friction created by fixed-rule systems. The proposed model combines supervised learning, unsupervised learning, and human review, and is designed to reduce that friction without weakening security. Its performance was not measured in this study and requires subsequent empirical validation.