Armenia SRC — ML-Powered Tax Compliance and CIT Gap Estimation

Armenia's State Revenue Committee (SRC) launched an AI-driven tax-compliance reform in 2024 with World Bank support under the Public Sector Modernization Project (Phase 4), partnering with the American University of Armenia (AUA) to build government machine-learning capacity.

The core application is a corporate income tax (CIT) gap estimation model intended to replace lower-accuracy random-sampling audit selection with machine-learning-driven targeting of high-risk companies.

IMF technical assistance estimated Armenia's CIT gap at 26.4-35.2% of potential liability, providing the baseline the ML model addresses; the ML system combines this gap analysis with taxpayer transaction data to select audit targets. A formal Memorandum of Understanding between the SRC and AUA was signed on 17 April 2025, and design principles include model explainability, transparency of risk criteria, and alignment with ISO cybersecurity standards. The system also informed plans to extend AI to customs risk management and X-ray image analysis at border checkpoints.

The SRC intensified risk-based audit selection in 2025, with company inspections increasingly driven by analytical tools. Based on international comparisons — India and Brazil are cited as having achieved 10-30% collection gains from similar approaches — the Armenian government projects a 15-20% improvement in voluntary and enforced compliance, though no post-pilot measurement of Armenia's own results has been published.

This is one of the first substantive AI deployments in tax administration in the Caucasus region, with formalised academic-government collaboration and explicit design commitments to explainability and transparency, but its compliance-improvement figures remain projections rather than measured outcomes.