How do teams integrate expertise, respond to disagreement, and coordinate as their work unfolds? I develop a pipeline that combines multimodal AI and concept bottleneck model to analyze these processes across more than 1,000 hours of recorded interaction in scientific, entrepreneurial, and sports settings. In this pipeline, I develop codebooks grounded in organizational theory, validate AI annotations against human judgments, and relate interaction patterns to collective outcomes.
This stream received $75,000 from Microsoft’s AI and the New Future of Work program for Building Behavioral Process Models from Real Teams with Multimodal AI. I independently wrote the proposal, with Brian Uzzi as PI. Our related TeamLens proposal was also recognized as a finalist in the Stanford HAI and Google DeepMind AI for Organizations Grand Challenge.
Measuring what matters in teams
I develop a methodological agenda for studying team interaction at greater scale, resolution, and temporal continuity. Linking language, vocal cues, and visible behavior could help researchers connect organizational constructs to enacted behavior, trace sequences of coordination and repair, and test which patterns generalize across settings.
This perspective article establishes the research agenda. The empirical projects below develop and test applications, with measurement validity assessed for each behavior and setting.
Evey Huang, Brian Uzzi, and Matt Groh. Measuring What Matters in Teams with Multimodal AI. [Revise and Resubmit] at Journal of Organization Design
Science teams
Manuscript in preparation. We examine integration behavior: how members build on, connect, evaluate, and extend one another’s contributions. Current analyses suggest that effective teamwork is distinguished by what the team does with its members’ ideas, alongside the ideas they contribute. We study how these patterns relate to collective outcomes, providing an empirical foundation for AI that can help people integrate existing knowledge.
Evey Huang, Max Chalekson, Matt Groh, D. Abrams, and Brian Uzzi. Multimodal AI for Large-Scale Scientific Team Behavior Analysis.
Entrepreneur–investor interaction
I extend the approach to study the back-and-forth interaction between entrepreneurs and investors. We investigate how questions, challenges, and responses reveal preparedness, build rapport, and relate to evaluators’ judgments. The focal interaction is the exchange between founders and evaluators, which allows us to study evaluation beyond a prepared presentation. Findings and their boundary conditions remain to be established.
Huang, E. J., Grey, A., Uzzi, B., & Groh, M. Measuring Catch and Toss and Its Impact in Entrepreneurial Pitching. [Submitted] to Organization Science Special Issue: AI-Enabled Frontiers in Organizational Science.
Professional volleyball
We use team huddles to examine how cohesion is enacted through observable interaction and how those dynamics relate to subsequent performance. Professional volleyball provides a setting where actions are tightly interdependent and coordination takes place under time pressure. Together with the science and entrepreneurship studies, it helps us ask which patterns of collaboration carry across settings and which depend on the demands of the work.
Grey, A., Huang, E. J., Uzzi, B., & Groh, M. How Teams Recover After Setbacks. [Submitted] to Organization Science Special Issue: AI-Enabled Frontiers in Organizational Science.