Bright sides and governance costs of GenAI in business processes: insights from a multi-case study analysis

Purpose This study explores the governance tensions that arise during the adoption of generative artificial intelligence (GenAI), examining how the same mechanisms that create value also introduce frictions and risks. It identifies the Assure-Account-Align (AAA) routine as a process-oriented framework for navigating these tensions, in which observability and authorisation serve as gating constructs that mediate the progression from experimental to embedded GenAI use. Design/methodology/approach The study adopts a qualitative, exploratory multi-case design based on 14 semi-structured interviews with experts and users from three organisations in Northern Italy, complemented by internal documents and direct observation. Data were analysed using a hybrid deductive-inductive coding scheme in NVivo, structured around the integrated Technology–Organisation-Environment (TOE) perspective and the Dynamic Capabilities (DC) framework. Findings The findings reveal that GenAI adoption progresses not through model performance alone, but through a conjunction of observability (provenance, replayability, intervention capability) and authorisation (regulatory, contractual, professional and reputational approval routes). These mechanisms simultaneously enable value creation and introduce governance tensions – workflow overhead, approval bottlenecks and deskilling anxieties – that the AAA routine helps organisations navigate iteratively. Originality/value The study conceptualises governance tensions in GenAI adoption as inherent paradoxes arising from the same mechanisms that create value. It proposes the AAA routine, on an exploratory basis and grounded in three heterogeneous cases, as a candidate bridge between contextual conditions (TOE) and dynamic action (DC), offering a sensitising device for future research rather than a validated framework.