From digitisation to digital transformation: a stage-sensitive actor–network theory of AI/ML diffusion

Purpose This paper aims to explain how artificial intelligence/machine learning (AI/ML) systems diffuse and stabilise across societally consequential contexts by integrating the digital change literature with actor–network theory (ANT). While existing research differentiates digitisation, digitalisation and digital transformation, it often remains descriptive and stage-based. At the same time, AI/ML adoption studies tend to privilege technical readiness and organisational determinants over the sociotechnical mechanisms through which systems become embedded, contested or rejected. Design/methodology/approach The paper adopts a conceptual theory-building approach. It integrates three literature streams: digital change, ANT and AI/ML adoption. Digitisation, digitalisation and digital transformation are treated as analytically distinct diffusion environments rather than as a deterministic ladder. ANT provides the mechanism vocabulary, including translation, inscription, black-boxing and irreversibility. On this basis, the paper develops a stage-sensitive conceptual framework, specifies the unit of analysis as an AI/ML sociotechnical assemblage-in-use and advances five propositions. Findings The paper finds that the two literature streams offer complementary explanatory strengths. The digital change literature clarifies where diffusion unfolds, whereas ANT explains how embedding is achieved or disrupted. The framework shows that translation is more likely to succeed when aligned with the dominant inscription site of each environment: data in digitisation, workflows in digitalisation and infrastructures in digital transformation. It further distinguishes between mere stabilisation and socially preferable stabilisation, with distributed contestability identified as the key differentiating condition. Social implications The paper argues that widespread and durable AI/ML use should not be equated with desirable outcomes. Systems may stabilise while remaining opaque, difficult to challenge and weakly accountable. By distinguishing mere stabilisation from socially preferable stabilisation, the study foregrounds the social importance of preserving contestability, correction and redress across all stages of digital change. This has implications for due process, public trust, legitimacy and the distribution of harms and benefits in AI/ML-mediated decision environments. Originality/value The paper contributes by bringing together two literature streams that have largely developed in parallel. It extends ANT by making it stage-sensitive and extends digital change research by supplying a mechanism-based account of diffusion. It also introduces the distinction between mere stabilisation and socially preferable stabilisation, showing how contestability and accountability may be understood as materially and institutionally assembled properties of AI/ML sociotechnical systems rather than as external ethical add-ons.