Portugal's Tribunal de Contas — a Machine-Learning Risk Model to Target Public-Procurement Audits

Public procurement equals roughly 10.3% of Portugal's GDP, and the Tribunal de Contas, like most supreme audit institutions, historically relied on sampling-based manual review to check contracts for irregularities, limiting how much of total spend could realistically be audited.

From around 2023, the Tribunal de Contas partnered with NOVA Information Management School, with OECD technical support and European Commission funding, on a two-year project to build a data-driven risk-assessment model. The system applies a risk matrix of 37 indicators — covering award procedures, single-bidder awards, contract modifications, execution delays and cost variation — fed by data integrated from the public procurement institute IMPIC, the registries institute IRN, the Tax Authority and the Competition Authority, alongside the Tribunal's own procurement platforms.

The tool was publicly presented in early 2025 and the OECD cited it as a model example for other audit institutions in a dedicated methodology paper (Working Papers on Public Governance No. 83, June 2025). The OECD's own working paper is explicit that the Tribunal has not yet published hard performance figures — no public accuracy rate, count of contracts flagged, or funds affected.

This is a genuinely deployed practice, but one still at an early, unquantified stage rather than a mature system with a multi-year track record; the OECD recommends the Tribunal begin measuring and publishing impact indicators going forward.