ESCORT — Latvia's State Revenue Service ML risk-scoring system for tax compliance

ESCORT is a machine-learning fraud-risk-scoring system operated by Latvia's State Revenue Service (SRS/VID), assigning a risk score to every registered legal entity by combining SRS tax data with State Labour Inspectorate and other government data sources. Operational since 2019, it underwent a major technical upgrade in 2021 (consolidation of taxpayer accounts and migration to SAP HANA).

To detect VAT fraud, income-tax under-reporting, undeclared employment and irregular migrant labour more quickly, direct enforcement resources more efficiently, and, since February 2024, give the public visibility of company tax-compliance ratings via the Taxpayer Rating System.

A centralised data warehouse integrating tax, labour and other government data feeds a machine-learning risk-scoring model covering the entire registered taxpayer base; a public Taxpayer Rating System publishes company-level ratings on the SRS website.

Following the 2021 upgrade, fraud-detection lead times fell from approximately one month to minutes, and production of analytical risk reports fell from four weeks to one day, as reported by implementation partner Emergn. The SRS processes approximately €10 billion in annual tax payments. No independent algorithmic audit or revenue-recovery/audit-yield statistics have been published.