This study aims to understand complex and dynamic group coordination, wherein three or more members interact over time through role-sharing to achieve a common goal. We proposed a conceptual model of information sharing and interaction in role-sharing for high team performance. This model will guide designs for coordination between humans and autonomous AI. As a pilot study, we analyzed interview data recorded in a field experiment observing offensive coordination in 3-on-3 basketball. Six top-level female players from Shizuoka Prefecture who competed at the National Sports Festival of Japan were divided into offensive and defensive teams. A mini-game designed in a previous study was repeated. The offensive players explained crucial interactions, shared ideas, and anticipated teammates’ plays, etc. The constructive concepts of offensive strategies were extracted from each player’s report, and their co-occurrence relationships were networked. The results confirmed shared and individual-specific concepts, respectively. The shared concept was prominent across the players. The individual-specific concept was prominent for each player. The shared concepts acted as hubs and tended to have higher co-occurrence relationships with some individual-specific concepts. Considering quantitative and qualitative results, we propose a conceptual model comprising three elements: (A) sharing a core strategy related to a mental model for efficient interaction, (B) leaving to other roles without information sharing, which may trigger flexible interaction, and (C) relying on implicit anticipation and trust to support (A) and (B). This study provides fundamental findings for understanding complex and dynamic coordination. Future studies should analyze group behavior to investigate the three elements.