Self-managed teams outperformed hierarchies in AI-assisted production: better and more frequent communication drove higher throughput and fewer errors, and statistical models indicate communication fully accounts for the organizational advantage.

Main Finding

Self-managed teams outperform hierarchical teams in handling automation failures in AI-assisted production: they communicate more (in frequency and quality), achieve higher productivity, and make fewer errors. Structural equation modeling indicates that the effect of team organization on output and quality is fully mediated by team communication.

Key Points

• Research question: Do hierarchical versus self-managed team structures differ in work performance (productivity and product quality) in highly automated, AI-assisted production—particularly during automation failures?
• Hypotheses:
  • H1: Self-managed teams develop better communication (frequency and quality) than hierarchical teams.
  • H2: Greater communication (frequency and quality) improves productivity and product quality.
  • H3: Communication mediates the effect of team organization on performance.
• Empirical result summary:
  • Self-managed teams communicated more often and with higher quality.
  • These communication differences led to higher productivity and fewer defects during simulated automation failure events.
  • Structural equation models show communication fully mediates the relationship between team organization and performance outcomes.
• Contribution: Bridges human–AI teaming (HAT/HAIT) literature with organizational sociology by experimentally testing how internal team design affects human–AI problem-solving in a realistic production simulation.

Data & Methods

• Design: Laboratory experiment simulating a realistic Industry 4.0/Smart Factory environment with highly automated machinery supported by an AI assistant; included a simulated machine failure to probe teams’ failure-handling.
• Treatments: Team organization manipulated as hierarchical (appointed leader, vertical information flows) versus self-managed (horizontal organization, autonomous role allocation).
• Key measures:
  • Communication: frequency and perceived/observed quality of team information exchange (distinguishing horizontal vs vertical flows).
  • Performance: productivity (output) and product quality (errors/defects) during failure scenarios.
  • AI role: assistant providing information/recommendations rather than high-autonomy agent.
• Analysis: Structural equation modeling to assess mediation by communication; comparisons between organizational conditions for performance outcomes.
• Methodological positioning: deliberately moved beyond common HAT lab paradigms (video games, Wizard-of-Oz) toward a more realistic simulated production task to improve external relevance.
• Limitations noted by authors: laboratory setting (trade-off between control and field realism); AI modeled as assistant rather than highly autonomous teammate; context-specific moderators (task complexity, leadership style) remain to be explored.

Implications for AI Economics

• Organizational form moderates returns to AI investment: simply deploying AI assistance in production will not automatically raise productivity—team organization and communication practices determine realized gains, especially under failure conditions.
• Value of communication capacity: investments that improve intra-team communication (training, decision protocols, communication tools, shared dashboards, explicit pull/push communication practices) can amplify the productivity and quality benefits of AI in production settings.
• Design priorities for AI in workplaces:
  • Prioritize transparency/explainability and concise actionable outputs that support team-level judgment rather than encourage overreliance.
  • Design AI interfaces to facilitate horizontal information sharing (easy broadcasting, querying, and joint situational awareness) to better fit self-managed team workflows.
• Labor and management implications:
  • Firms may realize higher effective productivity from AI when organizational designs shift toward greater team autonomy — suggesting potential returns to reorganizing work (and retraining managers).
  • Middle-management roles and incentive structures may need redesign; lean/hierarchical models could blunt AI benefits by suppressing communication.
• Policy and adoption considerations:
  • Policies promoting worker participation, team training, and support for organizational change could increase the social returns to AI adoption in manufacturing.
  • Productivity estimates used in macro/firm-level models of AI impacts should account for organizational heterogeneity (team structure, communication capacity) and the distributional effects of shifting to self-managed teams.
• Research directions relevant to AI economics:
  • Quantify how much of AI-related productivity gains are attributable to organizational change versus technology alone.
  • Field experiments to measure cost-benefit trade-offs of reorganizing teams around AI (implementation costs, transition frictions, impacts on labor demand and wages).
  • Explore interactions between AI autonomy level, team size/composition, and alternative leadership styles to refine predictions about firm-level returns to AI investments.