The Impact of Generative AI Adoption on Organizational Networks: Evidence From A Field Experiment
Citation. Büchsenschuss, R., Koch-Bayram, I., Biemann, T., & Puranam, P. (2026). The Impact of Generative AI Adoption on Organizational Networks: Evidence From A Field Experiment. SSRN. https://doi.org/10.2139/ssrn.6028034
Summary. This working paper reports a randomized field experiment in a European technology services firm testing whether access to a grounded GenAI assistant changes employees’ collaboration and knowledge-sharing networks. Across 316 employees in 42 teams, GenAI access increased collaboration and knowledge-network centrality, satisfaction with knowledge access, and project output over a three-month post-treatment window. The paper argues that grounded GenAI can act as a translator across expertise boundaries and a knowledge catalyst that changes who becomes valuable as a source of information inside the organization.
Key claims
- GenAI adoption increased employees’ network connectivity: individual-level treatment effects were positive for knowledge-network degree, in-degree, and out-degree and for collaboration-network degree, in-degree, and out-degree, with clustered-standard-error OLS coefficients reported as statistically significant.
- The intervention produced role-differentiated effects: specialists had a larger increase in knowledge-network in-degree than generalists, while generalists had larger increases in satisfaction with knowledge access and projects executed.
- The paper frames grounded GenAI as changing the local economics of human-human interaction, not only individual task productivity: AI can reduce translation and coordination costs while making some employees more valuable knowledge sources.
- The authors argue that network effects create managerial design issues beyond adoption and training, including possible information overload, coordination fatigue, and the need for forums where AI-surfaced information is discussed, vetted, and refined collectively.
Evidence & limitations
- Evidence comes from a field experiment in one Central European technology services firm; 42 teams were randomly assigned to receive a grounded GenAI assistant connected to organizational knowledge bases and communication data or to continue without it.
- Outcomes were measured over a two-week pre-treatment window and three months after introduction; network ties and knowledge-access satisfaction were self-reported, while project output was measured as number of projects executed.
- Role comparisons between generalists and specialists are not causal because roles were not randomly assigned, and the paper notes a strong negative baseline relationship between knowledge-network in-degree and project output that complicates interpretation.
- The single-organization setting, short post-treatment window, and survey-based network measures limit generalization; longer-term and digital-trace evidence is needed to test whether the network changes persist, overload key actors, or reduce other forms of direct mentoring and communication.