When Artificial Intelligence Meets Organizational Intelligence

Why does the promise of predictive artificial intelligence (AI) so often diverge from its organizational reality? Prior research alternately treats AI as a tool that augments decision making or as an agentic technology capable of autonomous discovery, yet struggles to explain how these views play out in practice. Drawing on a longitudinal ethnography of predictive AI initiatives in two large Indian banks, we examine how predictive AI is organized amid competing expectations for what AI is and should do. We reconceptualize AI integration as a problem of assemblage formation rather than technology adoption. We show how organizational actors assemble configurations of humans (and their expectations), data, algorithms, and models that enact distinct orientations toward AI, and how these assemblages collide and are reworked over time. Rather than converging on a single configuration, AI initiatives stabilize through partial and uneven reassemblies in which selected data-driven capabilities are absorbed into durable procedural and monitoring infrastructures. By unpacking the internal composition and biographic trajectories of AI assemblages, our study explains why predictive AI can be simultaneously transformative and disappointing, and advances an assemblage-based account of AI integration centered on organizational intelligence and organizing processes.