When more detection means less performance: the detection–validation imbalance in AI-enabled fraud detection

Artificial intelligence (AI) has become a vital tool for fraud detection, enabling financial institutions to identify suspicious transactions at greater speed and scale. However, improved AI capability does not always enhance organisational performance. This study investigates this paradox using a qualitative interpretive approach guided by socio-technical systems theory, information processing theory, and absorptive capacity theory. Semi-structured interviews were conducted with 18 professionals across 12 financial institutions in Ghana. The findings show that fraud detection effectiveness depends on aligning AI-driven screening capacity with human validation capacity. AI primarily enhances control and directs attention, while forensic accounting expertise enables interpretation and validation of algorithmic outputs. A detection–validation imbalance emerges when alert volumes exceed validation capacity, creating bottlenecks and false-positive burdens. The study also identifies objective-function mismatches between technical optimisation and organisational performance goals. It contributes a mechanism-based explanation of AI-enabled fraud detection.