evidoria

← Back to browse

Good practice Imported

Armenia's State Revenue Committee Pilots AI to Read Invoices and Flag Tax Fraud Rings, Results Still Unmeasured

Armenia · Yerevan · See the Armenia profile · See the Yerevan profile

Evidence: Descriptive / self-reported Top 97% 20/100 · Ask Evidence Copilot about this practice

Armenia's State Revenue Committee is piloting an AI tool, built by Interpretable AI and Daniam LLC with the American University of Armenia and the World Bank, to read invoices and flag fraud rings. As of mid-2025 only projected compliance gains (10-20%) had been published.

10-20 %
Projected rise in voluntary and enforced tax compliance (not yet measured in Armenia)
Armenia's State Revenue Committee Pilots AI to Read Invoices and Flag Tax Fraud Rings, Results Still Unmeasured

Details

Maturity
Pilot
Promoter
State Revenue Committee of Armenia, with the World Bank Public Sector Modernization Project and American University of Armenia
Period
2025–ongoing (pilot)
Keywords
tax administration, fraud detection, natural language processing, public sector AI

Context

Armenia's State Revenue Committee (SRC) is piloting an AI system to help detect tax evasion, developed by Boston-based Interpretable AI and Yerevan-based Daniam LLC with academic support from the American University of Armenia (AUA) and World Bank financing.

Objectives

The tool is designed to read invoices using natural-language processing, apply network analysis to identify fraud rings, and flag anomalies such as duplicate filings or identity mismatches, with a citizen-facing chatbot and faster refund processing planned for later stages.

Activities

SRC chair Eduard Hakobyan and AUA president Bruce Boghosian signed a memorandum of understanding on 17 April 2025, at a workshop attended by World Bank GovTech lead Khuram Farooq and MIT professor Dimitris Bertsimas, Interpretable AI's co-founder.

Results

The World Bank and IMF have each published accounts of the pilot, but both state clearly that the figures they cite — conservative estimates of a 10-20% rise in compliance, informed by comparable results in India and Brazil — are projections rather than measured outcomes from Armenia's own deployment.

Conclusions

The publishers themselves name bias, data privacy, security and job displacement as live risks, and describe safeguards such as explainability, ISO-aligned cybersecurity and a future Algorithm Impact Assessment as still to be built rather than already in place, underlining that this remains an early-stage, unverified pilot.

Implementation
Indicative cost
Medium (€50k–€500k)
Time to results
Medium (1–3 years)
Staffing & skills
State Revenue Committee of Armenia (chair Eduard Hakobyan), Interpretable AI (Boston), Daniam LLC (Yerevan), American University of Armenia (president Bruce Boghosian), World Bank GovTech lead Khuram Farooq, MIT professor Dimitris Bertsimas

Conditions for success

  • Formal memorandum of understanding between SRC and AUA
  • World Bank financing under a Public Sector Modernization Project
  • Academic partnership providing technical and oversight capacity

Common failure modes

  • Bias, data privacy, security and job displacement named as live risks by the publishers themselves
  • Explainability and ISO-aligned cybersecurity safeguards not yet built
  • Algorithm Impact Assessment still planned, not in place

Commonly funded by

National / regional programmes

Indicative funding routes for practices of this type — always check each programme's current calls and eligibility rules.

Do you run this practice? Claim it — verified implementers get a public contact pathway and can propose corrections.

Data sources

Where this practice's information was retrieved from, and when.

Attachments

Similar practices you may find useful