Team dynamics during the delivery of a large-scale, engineered system

Coordinated collective action refers to the synchronized action of agents towards achieving a predefined set of goals. Such activity is at the core of a wide range of social challenges, from organizational dynamics to team performance. Focusing on the latter, a novel dataset that captures the planned efforts to deliver a large-scale, engineered system is introduced. In detail, this dataset is composed of a total of 271 unique individuals, responsible for the delivery of a total of 721 tasks spread across a period of 745 days. The focus of this analysis is on the collaboration network between individuals, captured by their co-assignment in the delivery of particular tasks, and their dynamical patterns. Results indicate that the delivery of some tasks depends on disproportionately large collaborations, making them intrinsically harder to manage compared to tasks which depend on small, or no, collaborations. Similarly, some tasks require a disproportionately diverse set of skills to be completed, further enhancing their intrinsic complexity. Shifting focus to the topology of the contribution network, an abrupt emergence, and subsequent contraction, of a single large cluster is observed. This phenomenon corresponds to the emergence of an increasingly large and cohesive team, and its subsequent decomposition. In addition, the evolution of this cluster tightly follows the number of active tasks, suggesting that large teams are a natural way to respond to increased workload. These findings provide new insight on the underlying team dynamics that govern coordinated collective activity in general, and in the context of project delivery specifically.

💡 Research Summary

This paper presents a quantitative analysis of team dynamics and collaboration structures in the delivery of a large‑scale engineered system within the defense sector. The authors assembled a novel dataset comprising 271 unique individuals who were responsible for completing 721 interdependent tasks over a 745‑day period. By modeling the data as a bipartite “sociotechnical” network—where one set of nodes represents tasks and the other set represents resources (people)—they were able to project the bipartite graph onto two one‑mode networks: a task‑task “activity” network (weighted by shared resources) and a resource‑resource “contribution” network (weighted by the number of tasks jointly performed).

Temporal dynamics were captured by slicing the data into 20‑day windows, generating a sequence of network snapshots. For each snapshot, the authors applied a Newman‑Girvan modularity‑based community detection algorithm (run 1,000 times for convergence) to assign resources to clusters, which they interpret as functional teams. The average cluster size was then tracked over time, allowing the authors to relate team formation and dissolution to workload fluctuations.

Key findings include:

1. Task Heterogeneity – The majority of tasks are highly specialized, requiring only a single role, but a small subset is markedly interdisciplinary, demanding up to seven or more distinct roles. The number of resources assigned to a task (k_r) and the number of unique roles required (k_r^r) are strongly positively correlated (Spearman ρ ≈ 0.78) and exhibit a super‑linear relationship, indicating that certain roles become heavily concentrated among many individuals.

2. Ownership vs. Shared Ownership – The authors define “ownership” (the number of resources directly assigned to a task) and “shared ownership” (the average number of resources that other tasks share with it). Tasks with high ownership tend to have low shared ownership, suggesting they sit on organizational boundaries and act as communication bottlenecks. These bottleneck tasks are few in number but critical for the flow of expertise across the project.

3. Critical Pairwise Collaborations – By measuring the proportion of daily active tasks that a given pair of resources is responsible for, the study identifies a small set of pairwise links that account for a disproportionate share of daily output. Approximately 5 % of all links contribute to more than 50 % of the daily work, highlighting them as high‑risk points where failure could cascade through the project.

4. Abrupt Emergence and Contraction of a Giant Cluster – As the number of active tasks rises, the contribution network undergoes a rapid transition from many small clusters to a single giant component that encompasses roughly 60 % of all resources. This giant cluster persists while workload remains high and then gradually fragments as activity declines. The correlation between cluster size and active‑task count is 0.92, indicating that large, cohesive teams naturally form in response to increased demand.

The authors discuss practical implications: (i) early identification of highly interdisciplinary tasks and explicit definition of interfaces can mitigate coordination overhead; (ii) assigning dedicated coordinators to bottleneck tasks can reduce communication friction; (iii) monitoring and reinforcing critical pairwise collaborations can prevent cascading delays; and (iv) organizations should institutionalize mechanisms for rapid scaling up and down of team structures in line with workload spikes.

Overall, the study demonstrates that in complex engineering projects, collaboration networks and team structures are tightly coupled to task complexity and workload dynamics. By leveraging bipartite network modeling, temporal clustering, and quantitative measures of task and link importance, the paper provides a robust analytical framework that can inform both academic research on coordinated collective action and practical project management strategies.