People are an integral part of any team. They form a network with edges representing their relationships. Network attributes such as robustness of the network structure, flow of information, and network balance change over time with performance on sequential tasks and feedback processes. A well-structured team can deliver a good performance even in the presence of node/edge failures and external shocks.
In this work, we used a field dataset extracted from a hedge fund company to investigate the relationship between network structure and performance. The dataset contains instant messages between employees in the hedge fund and their performance over a period of five years. The performance metric of an employee is his/her gross profit. A message classification method was first built to classify messages based on content (business or social). Then, two networks between employees were constructed. We conducted extensive experiments to investigate the relationship between network structure and collective performance. We observed the following: 1) High network robustness and balance lead to high performance. 2) Performance of a node is related to the sentiment of its messages. 3) Nodes with a similar performance tend to cluster. These findings can help in the design of good teams.