Artificial Intelligence Governance and Institutional Capacity in the Public Sector: Lessons from Estonia and Rwanda for Developing and Post- Conflict States: A Literature Review

Abstract

Governments across the developing world are adopting artificial intelligence (AI) in public administration faster than they are building the institutional capacity to govern it. This review synthesises twenty-six peer-reviewed studies to examine what determines AI governance readiness in the public sector, drawing on the Technology–Organisation–Environment (TOE) framework that recurs across the literature, on the specific and consequential problem of public-sector corruption, and on the contrasting digital government trajectories of Estonia and Rwanda, two small states frequently cited as models but built on different institutional logics: legally mandated interoperability in Estonia's case, and ICT-led policy leapfrogging in Rwanda's. The review finds consistent evidence that technological infrastructure, organisational capacity, and regulatory or environmental context jointly determine AI governance readiness, and that readiness, not technological sophistication alone, is what converts AI adoption into public value rather than accountability risk. Emerging evidence further suggests that AI can support corruption control specifically, but only where institutional readiness and citizen trust are already present, a conditionality directly relevant to the sequencing question this review addresses. Studies of low-income and fragile states report regulatory fragmentation, weak data governance, and limited human capital as recurring constraints, and argue for governance-before-technology sequencing rather than pilot-first approaches. The review closes by considering what these findings imply for public institutions in Somalia, a post-conflict state building administrative capacity largely from the ground up, and identifies the absence of Horn of Africa-specific empirical evidence as the clearest gap for future research.