Analyzing over 360 billion emails across 4,361 organizations, Zuzul and colleagues showed that after the onset of Covid-19, intra-organizational communication networks became more siloed (increased modularity) while membership within silos became less stable (lower year-over-year adjusted Rand index) — a pattern they termed 'dynamic silos'. The study could not observe organizational outcomes or in-person communication, and explicitly left open whether these structural shifts causally affect productivity, efficiency, learning, or innovation (including the type of innovation), which mechanisms generate them (task-focused essentialist communication, loss of serendipitous cross-community interaction, or household-driven scheduling reallocation), and how the phenomenon varies across geographies, industries, and functions. This core remains unresolved: existing causal evidence covers either the remote-work-to-structure link (Yang et al. 2022) or the remote-work-to-outcomes link (Bloom et al. 2024), but never the structure-to-outcomes chain, and mechanism identification at the community level has not been attempted.

Background

Modern organizations communicate through digital traces — email, instant messages, video calls, and calendar meetings — that can be aggregated into networks whose nodes are employees and whose edges are communications between them. The subcommunity structure of such networks (teams, workgroups, and functional areas, inferred from interaction patterns by community-detection algorithms such as Leiden — an algorithm that partitions a network into groups by maximizing a fit score — rather than taken from the formal organizational chart) is a central object of organizational network research. Two summary statistics are standard. Modularity, QQ, is the fraction of communication falling within inferred communities minus the fraction expected under a degree-preserving random null model; higher QQ means more siloed communication. Membership stability is measured by the adjusted Rand index (ARI), a chance-corrected measure of agreement between two partitions of the same set of nodes — here, the community assignments of the same month in consecutive years (a year-over-year comparison that controls for seasonality); lower ARI means employees are moving between subcommunities more fluidly.

Zuzul, Cox Pahnke, Larson, Bourke, Caurvina, Parikh Shah, Amini, Weston, Park, Vogelstein, White, and Priebe first posted this work as a preprint in 2021 and published it in Management Science in 2025. Comparing month-to-month and year-over-year network metrics over 24 months before and after Covid-19 — across 4,361 organizations worldwide and, within one global organization (Microsoft), across email, instant messaging, video calls, and calendaring — they found that in 2020 organizations became more siloed (increased modularity) while membership within silos became less stable (decreased ARI). They called this joint pattern 'dynamic silos'. The finding complements individual-level causal evidence: Yang, Holtz, Jaffe, Suri, Sinha, Weston, Joyce, Shah, Sherman, Hecht, and Teevan (2022) used the fact that some employees were already remote before the pandemic as a natural-experiment control and showed that firm-wide remote work caused collaboration networks of 61,182 Microsoft employees to become more static and siloed, with fewer bridges between disparate parts and a shift from synchronous to asynchronous communication — effects they noted may make it harder to acquire and share new information, without measuring any organizational outcome.

Organizational theory gives competing predictions about what such restructuring should do. Tighter within-community communication can speed information sharing and tacit-knowledge transfer, so siloing may raise efficiency and productivity; conversely, novel recombination of knowledge held in distant communities is a classic source of innovation, so siloing may reduce it — and, paradoxically, membership instability that might have promoted cross-pollination was observed to coexist with greater siloing. Whether siloing favors a particular kind of innovation (incremental improvement versus 'competence-destroying' innovation that renders existing organizational capabilities obsolete) is likewise contested. On the outcomes side, Bloom, Han, and Liang (2024) ran a six-month randomized trial showing hybrid work-from-home improved retention without detectably affecting performance — but measured no network structure at all, so whether community restructuring is the channel through which work arrangements matter remains unidentified. What no study has done is connect measured community-structure dynamics to measured organizational outcomes within the same organizations, or identify which mechanisms generate the restructuring. The source study also introduced a generative model for such networks — a Barabási–Albert augmented hierarchical stochastic block model (BA-HSBM), combining preferential-attachment growth (new nodes link preferentially to already well-connected nodes) with hierarchical block structure (nodes grouped into blocks with distinct within- and between-block connection probabilities).

Problem Statement

Zuzul et al. (2025) established that, following the onset of Covid-19, intra-organizational communication networks worldwide exhibited 'dynamic silos': modularity increased (communication became more confined within subcommunities — teams, workgroups, functional areas) while community-membership stability decreased (year-over-year adjusted Rand index fell; employees moved between subcommunities more fluidly). The source study explicitly could not observe organizational outcomes or in-person communication. The open problem is the causal and explanatory completion of that finding, with one overarching objective — establishing whether and how remote-work-induced community restructuring matters for organizations:

  1. Consequences: do the concurrent increases in modularity and decreases in membership stability causally affect organizational outcomes — productivity, efficiency, learning, and innovation, including the kind of innovation produced (e.g., incremental versus competence-destroying or architectural innovation)?

  2. Mechanisms (explanatory layer): what generates dynamic silos — task-focused 'essentialist' communication, the loss of serendipitous in-person cross-community interaction, or time- and household-driven scheduling reallocation — and with what relative contributions?

  3. Boundary conditions (scope): how do the phenomenon and its consequences vary across geographies, industries, and functions?

Resolving the question requires connecting community-structure dynamics to outcomes within the same organizations, supporting causal claims with a credible identification strategy, providing evidence that discriminates among the candidate mechanisms rather than evidence consistent with all of them, and establishing how the phenomenon and its consequences vary across geographies, industries, and functions. The question is one of identification and analysis design — how structure and outcomes within the same organizations are to be connected, not what the empirical findings are — and it asks how the candidate mechanisms would be distinguished and their relative contributions quantified. The acknowledged measurement-robustness issues — the sensitivity of modularity and stability estimates to the temporal aggregation window and the unobservability of in-person communication — bear on whether such a causal claim can be established.

References

  1. Tiona Zuzul, Emily Cox Pahnke, Jonathan Larson, Patrick Bourke, Nicholas Caurvina, Neha Parikh Shah, Fereshteh Amini, Jeffrey Weston, Youngser Park, Joshua Vogelstein, Christopher White, Carey E. Priebe (2021, revised 2023). Dynamic Silos: Increased Modularity in Intra-organizational Communication Networks during the Covid-19 Pandemic. arXiv:2104.00641, DOI 10.48550/arXiv.2104.00641, https://arxiv.org/abs/2104.00641 (preprint of the Management Science article below)
  2. Tiona Zuzul, Emily Cox Pahnke, Jonathan Larson, Patrick Bourke, Nicholas Caurvina, Neha Parikh Shah, Fereshteh Amini, Jeffrey Weston, Youngser Park, Joshua Vogelstein, Christopher White, Carey E. Priebe (2025). Dynamic Silos: Increased Modularity and Decreased Stability in Intraorganizational Communication Networks During the COVID-19 Pandemic. Management Science 71(4):3428–3448. DOI 10.1287/mnsc.2022.02797, https://doi.org/10.1287/mnsc.2022.02797
  3. Longqi Yang, David Holtz, Sonia Jaffe, Siddharth Suri, Shilpi Sinha, Jeffrey Weston, Connor Joyce, Neha Shah, Kevin Sherman, Brent Hecht, Jaime Teevan (2022). The effects of remote work on collaboration among information workers. Nature Human Behaviour 6:43–54. DOI 10.1038/s41562-021-01196-4, https://www.nature.com/articles/s41562-021-01196-4
  4. Nicholas Bloom, Ruobing Han, James Liang (2024). Hybrid working from home improves retention without damaging performance. Nature 630:920–925. DOI 10.1038/s41586-024-07500-2, https://www.nature.com/articles/s41586-024-07500-2
  5. Dong-Nghi Pham, Luis Rios, Maciej Workiewicz (2022). Remote Access Memories: How Remote Work Will Impact Organizational Learning and What Firms Can Do About It. SSRN preprint, DOI 10.2139/ssrn.4032764, https://doi.org/10.2139/ssrn.4032764 (also in Academy of Management Proceedings 2022, DOI 10.5465/ambpp.2022.14485abstract, https://doi.org/10.5465/ambpp.2022.14485abstract)
  6. Kyle Schirmann, Miguel Espinosa, Prithwiraj (Raj) Choudhury, Tarun Khanna, Christos A. Makridis (2026). Assessing Information: The Content of Asynchronous Communication in Hybrid Work. AEA Papers and Proceedings 116:212–217. DOI 10.1257/pandp.20261010, https://doi.org/10.1257/pandp.20261010
  7. Jenni Kantola, Maria Pajuoja, Outi Vanharanta (2025). “I miss those chance encounters”: Innovative work behavior in hybrid and remote work. Computers in Human Behavior Reports 20:100823. DOI 10.1016/j.chbr.2025.100823, https://doi.org/10.1016/j.chbr.2025.100823
  8. Avanti Athreya, Zachary Lubberts, Youngser Park, Carey E. Priebe (2025). Euclidean Mirrors and Dynamics in Network Time Series. Journal of the American Statistical Association 120:1025–1036. DOI 10.1080/01621459.2024.2392912, https://doi.org/10.1080/01621459.2024.2392912
  9. Tianyi Chen, Zachary Lubberts, Avanti Athreya, Youngser Park, Carey E. Priebe (2026). Euclidean mirrors and first-order changepoints in network time series. Electronic Journal of Statistics 20:3658–3706. DOI 10.1214/26-ejs2572, https://doi.org/10.1214/26-ejs2572
  10. Cencheng Shen, Jonathan Larson, Ha Trinh, Xihan Qin, Youngser Park, Carey E. Priebe (2024). Discovering Communication Pattern Shifts in Large-Scale Labeled Networks Using Encoder Embedding and Vertex Dynamics. IEEE Transactions on Network Science and Engineering 11(2):2100–2109. DOI 10.1109/TNSE.2023.3337600, https://doi.org/10.1109/TNSE.2023.3337600