ChileCompra's Preventive Data-Intelligence System — Chile Flags Conflicts of Interest Before Contracts Are Awarded

ChileCompra, Chile's national public procurement agency (Dirección de Compras y Contratación Pública, part of the Ministry of Finance), runs a preventive data-intelligence system that screens the entire Mercado Público e-procurement platform for conflicts of interest before contracts are awarded, rather than auditing them after the fact.

Automatically cross-reference officials involved in a purchase against the declared beneficial owners of bidding companies, alerting the purchasing agency in real time when it detects an ownership or family tie so the process can be corrected, suspended or adjusted before adjudication.

The system interoperates with SIAPER, the Comptroller General's public-employee registry. A parallel workstream trains AI models to scan attached PDF bid documents for risk patterns such as collusion indicators and price anomalies. A nationally scaled version, monitoring 100% of Mercado Público transactions, went live in March 2026.

ChileCompra has published year-on-year figures for 2025: processes reviewed rose 10.2x (from 4,540 to 46,173), potential cases detected rose 19.6x (from 632 to 12,395), corrections made by public agencies rose 18.6x (from 185 to 3,434), and reports sent to oversight bodies rose 207%, with more than 13,000 notifications issued. By end-2025 the system had flagged potential conflicts across 1,216 purchase orders spanning 171 public institutions and 316 suppliers, covering $3.9 billion in reviewed procurement and $1.8 billion in linked commitments.

In August 2026, the Inter-American Network of Government Procurement (RICG), the Inter-American Development Bank and the OAS awarded ChileCompra their Innovation Prize for the system, citing it as a replicable preventive-integrity model for the region. The figures above are ChileCompra's own reporting of alerts and agency corrections, not an independent audit of corruption actually prevented, and the PDF-based collusion-detection models were still in training as of mid-2026 rather than fully deployed.