Why does organised criminal activity persist after the economic or enforcement conditions that initially generated it improve? This paper develops a dynamic political-economy model in which criminal participation changes two objects that standard occupational-choice models treat as fixed. Participation erodes the social cost of joining crime and accumulates organisational capital (e.g. recruitment networks, coercive capacity, financing relationships, territorial knowledge, and related infrastructure) that raises future criminal returns. Their interaction can generate multiple local equilibria, basin-dependent policy effects, and hysteresis. A temporary intervention can therefore have a durable effect by moving the state across a basin boundary, while restoring the original wage or enforcement environment may be insufficient after organisational capital has accumulated. With convex implementation costs, the least-cost transition allocates effort across legitimate opportunities, effective deterrence, and network disruption. I examine the model’s empirical implications using an Armed Conflict Location and Event Data (ACLED) state-year panel for Nigeria. In 2019-2024, high-intensity states remain in the high kidnapping-rate tercile in 73.8% of year-to-year transitions. With state and year fixed effects, both the one-year lag of kidnapping intensity and a depreciating stock of deeper history are significant, whereas the 2018/19 state poverty headcount is only weakly related to recent cross-state intensity. For context using qualitative evidence, documentary and field materials additionally records reinvestment of kidnapping proceeds in weapons, market and logistical infrastructure used by bandit groups, and socialisation of younger cohorts into an environment in which banditry is treated as a business. This evidence is descriptive rather than causal, but together it supports a model of criminal persistence that extends beyond contemporaneous poverty alone.