Crowd-based open innovation communities have received increasing attention, based on the premise that leveraging the power and diversity of the crowd can lead to innovative outcomes. However, we still know little about how work is coordinated over time in this context, especially as the innovation process moves from idea generation to elaboration. Based on literature and theories of coordination and collaboration in traditional creative contexts and on emergent evidence from research on crowd work, we develop hypotheses about the unique interaction patterns that characterize co-creation and how these patterns impact, over time, submission quality. To test our hypotheses, we conducted a study of a crowd-based open innovation platform. We found that, in general, the diversity of contributors increased over time, but for high quality submissions, the number of contributors decreased and a small group of involved people became more dominant in providing feedback. Further, we observe that the creators of more successful submissions, while not dominating the discussion, were particularly engaged in the discussions in later stages. Our work contributes to understanding the temporal dynamics in open innovation communities by providing evidence that successful interaction patterns vary depending on the phase of the project.