AI in Transfer Pricing: Risk or Opportunity? – A Summary

Artificial intelligence is rapidly transforming transfer pricing (TP) practice and enforcement. Tax authorities worldwide are adopting AI-driven risk models for audit selection, while multinational enterprises (MNEs) apply machine learning to automate benchmarking, documentation, and operational TP monitoring. This article examines whether AI in TP represents a risk or an opportunity, concluding that it is a net opportunity only when deployed as decision support under robust human-in-the-loop (HITL) safeguards.

Current Applications

MNEs use AI to accelerate benchmarking, with platforms like KPMG's tpEngine and Moody's TP Catalyst using machine learning to match intercompany transactions with arm's length comparables with greater speed and accuracy. Generative AI and natural language processing automate documentation preparation, with case studies showing a 68% reduction in local file preparation time and an 81% decrease in documentation errors. Real-time transaction monitoring enables proactive adjustments before year-end close, shifting TP from reactive compliance to continuous governance.

Tax authorities are equally active. Austria's Predictive Analytics Competence Centre flagged 375,000 cases in 2023, yielding €185 million in additional revenue. Italy's VeRa algorithm detects high-risk discrepancies, while Poland's STIR model analyses bank transactions in real time. The US IRS now uses machine learning to analyse all large partnership returns, supporting efforts to close a nearly USD 700 billion tax gap.

Opportunities

AI offers five key opportunities: (1) increased efficiency and accuracy through automation; (2) improved compliance via predictive analytics that flag issues before they materialize; (3) greater consistency and fairness by applying uniform criteria across transactions; (4) strategic insights aligned with ESG priorities, enabling tax transparency and fair profit allocation; and (5) faster dispute resolution through rapid analysis of large datasets.

Risks and Challenges

Substantial risks accompany these benefits. The "black box" problem creates explainability challenges, as seen in the UK Elsbury tribunal where opaque algorithmic logic undermined taxpayer trust. Data and bias issues remain critical, exemplified by the Dutch toeslagenaffaire where an algorithm discriminated based on ethnicity. Legal and regulatory exposure under GDPR and the EU AI Act creates compliance uncertainty—notably, the EU AI Act carves out civil tax administration from high-risk classification, creating a regulatory gap. Antitrust concerns arise when many firms rely on the same AI pricing engines, potentially creating tacit collusion. Litigation risks are expanding, with taxpayers seeking disclosure of AI models in discovery.

Legal and Policy Frameworks

The OECD TP Guidelines require AI outputs to align with the arm's length principle, while the EU AI Act (2024) imposes risk-based obligations. GDPR guarantees rights against solely automated decisions. The OECD AI Principles emphasize transparency, accountability, and human oversight. Best practices include maintaining HITL governance, ensuring explainability, implementing rigorous data governance, conducting regular audits and bias checks, upskilling staff, and establishing clear accountability frameworks.

Conclusion

AI in TP is a major opportunity that outweighs its risks only with strong safeguards and human governance. It improves efficiency, accuracy, and consistency while enabling predictive risk management. However, uncontrolled AI can produce opacity, bias, and unlawful practices. With appropriate governance—including transparency, auditability, and accountable human judgment—AI serves as an ally rather than an enemy, modernizing the international tax system while preserving the arm's length principle and procedural fairness.