AI May Take Transfer Pricing Up a Notch but Carries Some Risk

Transfer pricing is unique among all tax areas in that it requires very different data from a wider variety of sources. Artificial intelligence has the potential to unlock this data. Tax authorities have access to data that taxpayers don’t. Combined with that data , AI could strengthen the hand of tax authorities in tax and transfer pricing audits.

As of July 2023, the IRS’s Large Business and International Division had a 190-person staffing deficit in its transfer pricing practice. This limits its ability to conduct more transfer pricing examinations of large multinational corporate tax returns.

LB&I is closing that deficit with funding from the Inflation Reduction Act. Some of these new hires may find themselves collaborating with a different kind of IRS agent.

The Government Accountability Office recently stated that AI could help the IRS close the annual tax gap of almost $700 billion. While transfer mis-pricing accounts for only a small part of the tax gap, it is a big area of concern. The IRS and global tax authorities are exploring how AI can help them improve their enforcement success in transfer pricing.

Done right, AI can simplify transfer pricing audits, streamline compliance, and improve dispute resolution. Done wrong, AI could lead to transfer pricing agents gone wild—scrutinizing every small transaction—and an AI arm’s race among tax authorities with multinational enterprises caught in the middle.

Data is the soil from which AI models grow and are nourished. In turn, AI models can unearth secrets hidden in the data swamp, especially when that swamp contains different data elements.

Transfer pricing data is different from other tax data because it involves more nontraditional financial data. The transfer pricing data swamp stems from the complexity of what must be processed: both taxpayer-specific data and third-party, quantitative, and qualitative data.

This is especially tough for tax authorities. While they often decry the information asymmetry in transfer pricing, they do have one large advantage: access to vast amounts of transfer pricing relevant taxpayer data.

Tax authorities haven’t always been good at using that data. There have been some shoots of data analysis, such as the efforts by His Majesty’s Revenue & Customs around nudge letters in the profit diversion compliance facility or the more recent IRS inbound distributor initiative. And while those are very simple data analysis exercises not involving much AI, they give a sense of what might be possible with AI unleashed on transfer pricing by tax administrations.

Tax authorities can access vast data from a few different areas:

• Corporate tax and transfer pricing filings: tax returns, financial statements, transfer pricing reports, and transfer pricing forms
• Other tax data: customs data, electronic payment data, employee tax returns
• Other required data filings by corporations: This could include a variety of required statistical reports that provide information about MNE activities

In addition to this confidential data, transnational and private actors are making more efforts to compile data sets and insights from publicly available data. For example, the Organization for Economic Cooperation and Development and the United Nations have developed a platform that tracks the legal entity structure of the largest 500 MNEs built purely off publicly available data.

AI enables the integration of these disparate data sources and new insights that were previously too hard or costly to come by. Tax authorities will be able to use this data to develop proprietary AI and machine learning-based algorithms that will improve their ability to target intercompany transactions they deem “high risk.”

The IRS noted earlier this year that it has already started using AI to target large, complex taxpayers. Other tax authorities—such as the Australian Taxation Office, HMRC, Austrian Tax Authority, and French Tax Authority—have also been using AI. And this all started before generative AI came on the scene. Future investments in and developments of tax-administration-specific AI tools may very well tip the information asymmetry in favor of tax authorities.

With the data theoretically available today, tax authorities already can gain deep insights into the operations and transfer pricing of an MNE even before issuing a single information document request. Of course, this assumes the models are well developed and not so opaque that the tax administrators don’t understand them.

Tax authorities could be drawn into an escalating war over intercompany transactions as their different and not harmonized AI models alert them to “risk” areas they otherwise might not have identified.

But it isn’t necessarily all bad news for taxpayers. AI also offers an opportunity for improvement in resolving tax disputes. For example, AI could enhance mutual agreement procedures and advance pricing agreements by:

• Improving case management, which would automate routine tasks, track case progress, and prioritize high-risk cases
• Enhancing negotiation support, which would provide data-driven insights and scenario analysis to support negotiations, facilitating quicker resolutions
• Predicting outcomes, which would analyze past cases to predict likely outcomes, aiding informed decision-making

Based on its historical MAP and APA data, the IRS and other tax authorities could develop models to identify trends and common resolution patterns and make them available to taxpayers via AI models to help them anticipate challenges and develop effective strategies.

For example, a taxpayer with losses at a distributor could provide the AI with relevant data and the model could help predict the likelihood of achieving resolution, potential timelines, information they should be ready to present to the tax authorities, questions they should expect, and how answers to those questions could affect the outcome. This could be an AI pre-pre-filing conference.

We might also hope that AI will help the IRS and other tax authorities minimize targeting false positives. And we may even look at a future where virtual AI agents of different tax authorities harmonize their approaches and automatically negotiate resolutions. This might sound like science fiction today—but so did software that generates podcasts off tax rulings, until recently.

Like any transformative technology, AI comes with opportunities and risks. But there can be little doubt it will have a significant impact on transfer pricing and transfer pricing audits. Taxpayers should start to leverage AI to understand their own data better and how tax authorities may interpret their data with AI’s help.

This article does not necessarily reflect the opinion of Bloomberg Industry Group, Inc., the publisher of Bloomberg Law and Bloomberg Tax, or its owners.

Author Information

Thomas Herr is the national leader of transfer pricing services, for KPMG’s economic and valuation services practice.

Write for Us: Author Guidelines