Applying scheduling techniques to manage activity instances (tasks) in executors' personal work lists is crucial in workflow management systems (WFMSs). To date, most practical scheduling techniques focus on process (control-flow) perspective of WFMSs but do not adequately address issues relating to work lists containing executors' to-do activity instances. Existing simple-dispatching-rule based techniques cannot fully capture the complexity in personal work list management under dynamic workflow environments. Given this motivation, this paper proposes an approach that applies genetic algorithm (GA) to manage activity instances in personal work lists. This approach is applied according to activity instances' probabilities of satisfying deadlines and costs of violating deadlines. The approach can schedule activity instances among multiple executors' personal work lists and recommend to each executor a list of activity instances that can be successfully executed, while minimising the overall deadline violation cost. Using real-world data collected from three manufacturing enterprises, the paper fits the probability distributions of activity instances' key time features such as starting time and execution durations, and then generates simulation experiment data. This paper also studies the setting of GA's parameters to fit the practical manufacturing scenarios. Finally, by comparing with other seven typical workflow scheduling algorithms, the paper evaluates the effectiveness of our algorithm using a large scale simulation experiment.