Power systems are prone to cascading outages leading to large-area blackouts with significant social and economic consequences. Intentional controlled islanding (i.e. the separation of the system into sustainable islands) is an effective strategy to mitigate these catastrophic events. To ensure a correct separation, nonetheless, it is crucial to define a suitable time to split the system (i.e. to answer the when to island question). To consider the probability of the event, the reliability of the system components, the reliability of the information and communication technologies, and the potential economic costs of the event, answering the above question within a risk-based framework becomes critical. To date, however, this has not been done. This study proposes a risk-based methodology to define, in an adaptive manner, a suitable time to split the system following an event. This methodology complements the well-studied where to island question, resulting in an integral solution of the islanding problem. To illustrate the approach, the IEEE 118-bus dynamic system is adopted considering realistic security criteria. Simulation results demonstrate the effectiveness and flexibility of the methodology in identifying a suitable time for the creation of islands, which, in turn, results in the prevention of blackouts that would otherwise be obtained.