Shifting Work Patterns with Generative AI

Source: arxiv.org

The Challenge

While laboratory experiments demonstrate that generative AI boosts task-level productivity in controlled environments, empirical evidence regarding its actual impact on complex, everyday organizational workflows remains scarce. When workers are granted access to enterprise AI tools, they must independently navigate where technology fits into their workflows, adjust daily habits, and reorganize tasks. The central problem lies in understanding whether embedding generative AI into ubiquitous enterprise software genuinely reorganizes knowledge work, shifts task compositions, and improves operational speed, or merely introduces localized, minor time efficiencies without structural organizational transformation.

Core Findings

Drawing on a large-scale randomized field experiment across 66 enterprise firms and 7,137 knowledge workers, the researchers evaluated the rollout of Microsoft 365 Copilot over a six-month duration. Utilizing objective telemetry data across digital workplace tools (e.g., Outlook, Teams, Word), the authors analyzed time use and activity volume. The findings reveal that over 90% of treated workers used Copilot initially, stabilizing at roughly 40% active weekly usage after 12 weeks. Active users achieved an intention-to-treat reduction in Outlook time of 1.4 hours weekly, while local average treatment effect (LATE) estimates indicated a 2-hour (17%) decrease in email time alongside a 9% reduction in out-of-hours application usage. However, the study detected no significant changes in the total volume of emails replied to, meetings attended, or documents authored, indicating efficiency within routine tasks rather than structural task reorganization.

Strategic Takeaway

For digital leaders and socio-technical researchers, the study highlights that providing generative AI does not automatically lead to structural work transformation or immediate macro-productivity gains. Instead, enterprise AI primarily functions as an efficiency tool for information retrieval, synthesis, and communication triage. Leadership interventions must move beyond mere tool provisioning toward active managerial design, workflow restructuring, and capability building. To unlock transformative gains, organizations must intentionally redesign job roles and operational processes rather than expecting off-the-shelf digital assistants to spontaneously alter knowledge production.

Deep Dive Q&A

Did generative AI adoption alter the volume or composition of knowledge tasks?

No. The field experiment revealed no significant shifts in the volume or nature of completed tasks. Treated employees completed similar numbers of Word documents, attended equivalent numbers of Teams meetings, and sent comparable volumes of emails relative to control groups.

Where did the primary productivity gains occur during the trial?

The most significant gains appeared in email management and asynchronous communication. Active users saved approximately two hours per week in Outlook and reduced work outside regular business hours by roughly 9%, indicating localized operational time savings.

What explained the variation in generative AI usage across workers?

Firm-level differences accounted for the largest share of variance in AI adoption, followed by baseline individual digital behavior. This suggests that organizational culture, managerial practices, and internal training significantly govern technology assimilation.

Related Research & Implementation Playbooks

Cross-disciplinary frameworks and practical blueprints connected to this insight.