How AI Can Help Both Tax Collectors and Taxpayers

IMF Blog, February 25, 2025

Source details

New generative AI tools can redefine the relationship between governments and citizens, but strong leadership and safeguards are fundamental.
Authors: Thomas Cantens, Herve Tourpe
Date: February 25, 2025
Generative artificial intelligence (GenAI) goes beyond simple automation by understanding and producing human language, opening possibilities for both internal administration support and external taxpayer interaction.
Early GenAI applications have focused on improving communication with taxpayers; most efforts by tax authorities remain at an early, experimental stage.

Distinction from existing AI:
- Most current systems used by tax and customs authorities are predictive and built for a single function, analyzing structured data to produce outputs such as risk scores.
- GenAI is a generalist system that understands almost all forms of information and is designed to interact with humans in any language.
Capabilities:
- Drafting letters, providing interactive guidance about tax regulations, assisting officers in investigations.
- Can be trained with legal texts, tax codes, operating procedures, and internal guidelines to create a dynamic system for both civil servants and taxpayers.

Internal effects:
- GenAI can help tax and customs officials focus on analytical and judgment-based roles, enabling them to become oversight specialists and increasing productivity.
External effects:
- GenAI can reduce the knowledge gap between administrations and taxpayers by aiding interpretation of complex provisions, navigating laws, identifying deductions, and auto-filling forms.
- Human-like interactions can personalize taxpayer service, as illustrated by:
- Singapore: a virtual assistant answers tax questions in multiple languages and has cut call-center inquiries by half.
- Korea: an AI guide helps citizens file and pay taxes.
- France: AI can analyze incoming emails and propose draft responses for civil servants to validate.
Opportunities for low-income countries:
- GenAI offers a path to drive organizational reforms and leapfrog into modern systems.
- Example: Madagascar’s customs authority intends to use GenAI to improve risk management, combat fraud and increase revenue, using data accumulated over 10 years to train its system.
Civic engagement:
- Citizens’ organizations, academics, and political parties can use GenAI to examine proposed reforms, compare scenarios, and engage in deeper policy debates, potentially increasing trust and making taxation feel more like a shared responsibility.

Data and knowledge requirements:
- Effective knowledge management is essential; revenue authorities must identify accurate, relevant, and suitable documents from extensive laws, regulations, case records, and operational manuals.
- Scattered archives and incomplete digitization can hamper training efforts.
Human oversight and workforce readiness:
- A human must determine which documents are appropriate for training.
- Employees need training to interpret, correct, and complement GenAI outputs.
- Policymakers must ensure that errors are reported and addressed promptly.
Risks to trust and necessary safeguards:
- Issues include data quality, ethics, privacy concerns, and hallucinations (i.e., incorrect results).
- Results must be explainable and perceived as fair.
- Sensitive queries may need routing to human agents (as in Korea) to preserve confidentiality and oversight.

Leadership and governance:
- Strong leadership and ethical policy frameworks are fundamental to shaping GenAI properly in revenue administration.
Data governance and privacy:
- Vigilant oversight of data quality, privacy, and accuracy is required to reinforce — not erode — trust.
Operational safeguards:
- Design systems so particularly sensitive matters are handled by humans.
- Implement procedures to report and fix errors promptly.
Capacity building:
- Train staff to use GenAI outputs critically and to manage the systems’ knowledge bases.
Inclusive deployment:
- Use GenAI to lower administrative hurdles, clarify tax obligations, and invite broader participation in policy debates while maintaining legal and ethical standards.

IMF Blog post: "How AI Can Help Both Tax Collectors and Taxpayers" — Thomas Cantens and Herve Tourpe, February 25, 2025.