From Pilot to Production: Microfoundations and Responsibility Patterns for Enterprise AI Projects

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Enterprise AI projects in firms frequently fail to transition from pilot to production. This study characterizes this transition through 14 projects across multiple industries. Using an inductive multiple-case design and a microfoundations lens grounded in the dynamic capabilities framework, we identify 38 microfoundations and 12 meso-level capabilities spanning sensing, seizing, and transforming. While sensing and seizing activities largely resemble prior digital transformation research, the transforming dimension diverges, centering on operational AI governance, knowledge institutionalization, and AI asset evolution. We further develop a two-dimensional framework of responsibility allocation across piloting and productionizing, identifying three responsibility patterns: domain-driven, handover, and tech-driven. For theory, the transforming dimension requires a microfoundation profile distinctive to AI projects, and responsibility allocation is a phase-sensitive structural variable rather than a steady-state governance property. For practice, AI competence distribution is the key precondition shaping which responsibility patterns are sustainable – and something firms can deliberately develop.