High interest but low adoption: Navigating organizations’ journey towards generative artificial intelligence implementation

Abstract

The rapid development of generative artificial intelligence (aka, LLMs) provides high potential to transform organizational operations, yet a pronounced high interest but low adoption gap persists. Hence, moving beyond individual-level studies to examine organization-wide implementation, we draw on Rogers’ innovation decision process and Engeström’s activity theory, and conduct in-depth interviews with 27 front-line experts, including LLM providers, adopters, and advisors. Our analysis uncovers ten key contradictions and corresponding practice-driven solutions that emerge across five implementation stages (agenda-setting, matching, redefining and restructuring, clarifying, and routinizing). These insights illuminate not only the multi-stage, socio-technical complexity of LLM deployment but also shifting priorities among activity subsystems and the collaborative mechanisms essential for success. Building on these findings, we offer actionable recommendations for practitioners: a tiered rollout strategy; the technical capability building including decision-support and trial platforms, agile modular architectures and multi-layer update pipelines; as well as an accountable governance framework that integrates internal controls with external accountability. By synthesizing theoretical and practical perspectives, our study intends to guide researchers and business leaders navigate the challenges of organizational LLM implementation and realize their transformative potential at scale.