Group work is a commonly used method of working, and the performance of a group can vary depending on the type and structure of the task at hand. Research suggests that groups can exhibit "collective intelligence"—the ability to perform well across tasks—under certain conditions, making group performance somewhat predictable. However, predictability of task performance becomes difficult when a task relies heavily on coordination among group members or is ill-defined. To address this issue, we propose a technical solution in the form of a chatbot providing advice to facilitate group work for more predictable performance. Specifically, we target well-defined, high-coordination tasks. Through experiments with 64 virtual groups performing various tasks and communicating via text-based chat, we found a relationship between the average intelligence of group members and their group performance in such tasks, making performance more predictable. The practical implications of this research are significant, as the assembly of consistently performing groups is an important organizational activity.