Governing Algorithmic Personalization
AI-driven adaptive learning systems are now deployed across a growing share of higher education institutions, yet the governance literature accompanying this growth remains largely descriptive: it reports that policies fail to translate into practice without theorising why. This paper addresses that gap theoretically. Drawing on institutional theory’s concept of decoupling, the tendency of organisations to adopt formal structures for legitimacy while leaving practice loosely coupled to them, the author argues that the policy-practice gap documented across adaptive learning governance research is a predictable organisational response to legitimacy pressure, made more severe by the technical opacity of vendor-supplied systems. It integrates this institutional lens with socio-technical systems theory to derive a five-dimension governance framework, Governing Algorithmic Personalization (GAP), structured so that no dimension can be strengthened in isolation without exposing weaknesses at its boundary with the others, and refines the Capability Maturity Model’s staged-maturity logic with a decoupling-aware stage distinguishing adopted policy from verified, enacted practice. The opacity-decoupling mechanism is articulated through four falsifiable propositions, and the framework is illustrated using real, publicly documented evidence from named United States universities rather than a hypothetical scenario. While these illustrative cases demonstrate the framework's practical applicability, they do not constitute empirical validation. Accordingly, a Delphi study is proposed to provide the expert consensus required to validate the framework. The contribution is theoretical: it extends institutional decoupling theory into a domain where opacity makes verifying enactment unusually difficult, and refines an established maturity-modelling logic to accommodate this risk. The paper concludes with a research agenda for executing this validation, particularly in resource-constrained and Global South contexts underrepresented in current evidence.