This lesson provides a comprehensive examination of how tax administrations are using artificial intelligence (AI) and machine learning to improve efficiency, enhance compliance, and detect fraud. It covers the applications, benefits, and governance considerations for AI adoption.

• Widespread Adoption of AI: Over 70% of tax authorities are using AI to improve internal efficiency and taxpayer services . The adoption of artificial intelligence and machine learning has grown significantly, from approximately 30% of administrations using the technology in 2018 to 64% in 2022—a growth of 34 percentage points in just four years .

• Primary Applications of AI: The most common uses of AI in tax administration include :

  • Detecting tax evasion and fraud: This is the most frequently cited application of AI.

  • Risk assessment processes: Approximately 53% of administrations use AI in their risk assessment processes .

  • Enhancing risk scoring: Supervised machine learning models can score risk across millions of records, enabling auditors to focus on the highest-impact cases. Unsupervised learning can surface unusual clusters of behaviour that traditional rules might miss .

• Country Examples of AI Implementation:

  • France – Real Estate Rental Income Fraud: The Directorate General of Public Finances uses AI to identify tax fraud in rental income declarations. A machine learning model estimates the rental value of each property by analysing declared rent, property characteristics, and neighbourhood socio-economic data. The estimated rental amount is compared with declared rent to detect under-reporting. Cases selected for audit using this model resulted in additional tax being collected in nearly 50% of cases .

  • Japan – Targeted Compliance: The National Tax Authority uses AI to analyse data from tax returns, financial statements, third-party information, and audit history. In 2022, AI determined that the average amount of additional tax due per corporate and consumption tax audit for SMEs was 40% higher than the tax actually collected. AI is also used to predict the most effective method of contacting non-compliant taxpayers—whether by phone, in-person visit, or letter—based on historical data .

  • Australia – Data Modernisation: The Australian Tax Office has implemented a modernisation programme using the Nexus agile delivery methodology to facilitate improved collaboration. The ATO has established multi-disciplinary teams with cross-skilled members from data consumers, engineers, and producers, reducing skill dependencies and enabling continuous, timely design. New data patterns have been developed, including reusable configuration-driven patterns for ingesting streaming data across multiple use cases .

• Analytics Maturity Model: The OECD’s Forum on Tax Administration developed the Analytics Maturity Model to help tax administrations self-assess their current level of maturity in analytics usage and capability. Over 40 tax administrations have completed the self-assessment, which provides insight into areas of both weakness and strength, guiding analytics strategies .

• Data Science Skills and Infrastructure: Approximately 85% of tax authorities possess the talent, skills, and infrastructure needed to analyse big data. The purposes of big data analysis include improving compliance (89%), identifying trends (79%), and estimating tax revenues (64%) .

• Governance and Ethical Considerations: The adoption of AI by tax administrations must be subject to appropriate governance to ensure objectivity, data privacy, and trust. Key principles include :

  • Strong legal frameworks specifying purpose, scope, and retention

  • APIs that restrict access by purpose and jurisdiction

  • Encryption and role-based access controls

  • Model-explainability tools so decisions can be audited

  • Human-in-the-loop review for automated high-stakes actions

  • Transparency about how algorithms are developed and used

  • Explainable risk scores, clear redress mechanisms, and regular fairness audits