Rethinking fraud control: Fraud dynamics, effective prevention, and the limits of reactive detection-led strategies

Fraud against large private and public sector organisations remains persistent and difficult to measure. Despite sustained policy attention, some strategies prioritise detection and sanctions because prevention and deterrence effects are notoriously difficult to evaluate. This study argues that this evaluative difficulty is ontological: fraud is typically treated as a static quantity rather than as a dynamic system-level property arising from interacting controls and feedback processes within an open organisational system. An agent-based model (ABM) is developed to simulate five multi-layered organisational control regimes, testing how different configurations shape system behaviour over time. Within the specified feedback architecture, results indicate that the modelled prevention-oriented regimes generate more stable system dynamics than detection-heavy regimes, reducing fraud propensity, suppressing volatility, and constraining escalation. Conversely, detection-oriented regimes perform well on observable metrics but exhibit higher underlying propensity and greater exposure to peak harm. These findings challenge simplified deterrence assumptions centred on anticipated individual perceptions of punishment. Instead, fraud emerges from the interaction of opportunities, controls, and feedback. By distinguishing between behavioural dynamics and observable outcomes, the paper demonstrates how reliance on detection-based metrics can create a false impression of system health. Rather than aiming for prediction or prevalence estimation, the study advances ABM as a tool for under-labouring fraud policy within a complexity-informed critical realist framework. In doing so, it conceptualises fraud dynamics as an analytical approach that clarifies the systemic behaviours fostered by different control architectures, redefining success beyond the mere occurrence or non-occurrence of observable events.