Mexico SAT — Machine Learning and Graph Analytics for High-Risk Taxpayer Detection (Plan Maestro 2024)

Mexico's Servicio de Administración Tributaria (SAT) has progressively embedded machine-learning and graph analytics into its audit and compliance pipeline since 2020. The SAT's Plan Maestro 2024 formalised this approach under a four-pillar strategy: Compliance Monitoring, Deep Surveillance, Coercive Collection, and Taxpayer Assistance. Graph models map transactional networks among 80 million-plus taxpayers in the CFDI (Digital Tax Receipt) database — one of the world's largest electronic-invoicing repositories — to expose concealed ownership structures and falsified invoice chains associated with Empresas Facturadoras de Operaciones Simuladas (EFOS, 'shell invoicers').
Statistical learning models trained on historical evasion cases are applied in real time to nine priority sectors: automotive, alcoholic beverages and cigarettes, construction, pharmaceuticals, hydrocarbons, logistics, technology platforms, real estate, and insurance/financial services. The system flags inconsistencies between declared income, third-party bank data, customs records, and CFDI flows, generating risk scores that trigger tiered interventions from automated taxpayer letters to full audits. SAT reports collecting 4.517 trillion pesos in 2023 — maintaining a cost-to-collect of 28 cents per 100 pesos, below the US IRS benchmark of 34 cents per dollar. The cumulative EFOS blacklist reached over 14,000 entities by mid-2026, with 7,300 digital seals suspended in the first half of 2026 alone.
Critics and academic researchers have noted risks of algorithmic bias, insufficient public disclosure of model architecture, and concentration of scrutiny on specific economic sectors. Notably, new EFOS additions collapsed from over 3,000 per year (2018) to 47 in 2023, interpreted either as deterrence success or as reduced enforcement intensity. No independent algorithmic audit has been published. Mexico's use of electronic invoicing as a compliance tool — mandatory since 2014 for all taxpayers — provides an unusually rich data substrate for ML applications that is rare among middle-income countries.

Read the full analysis: https://www.gob.mx/sat/prensa/plan-maestro-2024-sat-optimiza-procesos-de-fiscalizacion-recaudacion-y-atencion-al-contribuyente-007-2024