Algorithmic Government in a Decentralised State: Capacity, Accountability and the Procurement Gap

Abstract

Governments increasingly delegate consequential administrative decisions to algorithmic systems, and the accountability mechanisms available to citizens have not kept pace. This study examines where in the life of such a system accountability can actually attach, and what follows for a decentralised state in which adoption occurs at subnational level. The method combines a structured review of research published between 2017 and 2024 with comparative analysis of two extensively documented failures and a lifecycle-based design analysis. Three findings emerge. First, accountability is effective mainly ex ante: analysis of the Dutch risk-indication case concludes that accountability configurations depend predominantly on the prior sensitivity and responsiveness of political forums, and that where these prove insufficient, mitigation during or after deployment is very difficult. Second, the decisions that matter most are made during procurement rather than during policymaking, since vendors embed optimisation targets, data selection and thresholds that agencies treat as technical rather than political. Third, the legal remedies commonly assumed to exist are weaker than supposed: no general right to explanation of automated decisions exists in European data protection law, and administrative law has no settled account of how judicial review applies to automated public decisions. The Australian and Dutch failures show what follows, including a settlement returning over one billion Australian dollars to approximately four hundred thousand people. For decentralised states the implication is structural: the forum that must act ex ante is dispersed across hundreds of subnational bodies with uneven capacity, which is where the safeguards proposed here are directed.