Transfer Pricing in the AI Era — Why It Matters More Than Ever

Introduction: Transfer pricing and its role in modern economy

Transfer pricing, the pricing of goods, services, and intangibles between related entities within a multinational enterprise, is probably one of the most important tax regulations of recent decades. From first attempt of pricing control in intragroup transactions in 1930s aimed at responding to the increasing complexity of multinational corporations (MNCs) and fighting early offshore tax practices it developed into a complex OECD Base Erosion and Profit Shifting (BEPS) Guidelines for tax authorities, whose core role is to facilitate fair distribution of taxes in accordance with country’s input into creation of value. Every day millions of intragroup transactions are being analyzed by both companies and tax authorities with one single aim – to ensure companies make fair tax input to countries where they produce goods and services.

Transfer pricing involves a range of analysis that boils down to 5 methods recommended by OECD Guidelines, and widely used by governments around the world. Transfer pricing methods are used to determine the arm’s length price for transactions between related parties (such as divisions or subsidiaries of a multinational company). The most commonly used methods are divided into Traditional Transaction Methods and Transactional Profit Methods. These methods are often outlined in the OECD Transfer Pricing Guidelines and local tax regulations.

Traditional Transaction Methods focus on specific transactions and try to determine the price as if the transaction had occurred between unrelated parties. Those methods include Comparable Uncontrolled Price, Resale Price Method and Cost Plus Method.  

Transactional Profit Methods include Transactional Net Margin Method and Profit Split Method.  

The choice of method depends on the nature of the transaction, the availability of data, and which method provides the most reliable result in determining the arm’s length price. In practice, the TNMM and CUP methods are among the most widely used due to their flexibility and applicability to a broad range of transactions.

Benchmarking study: search of independent comparables  

All methods mentioned above, except for CUP, require information on the profitability levels of independent companies. The most common approach to determining this is through a benchmarking study—a research process aimed at establishing the profitability levels of independent companies engaged in the same business as the analyzed transaction. In practice, a benchmarking study involves compiling statistical datasets containing financial indicators of independent companies operating within the same industry.

In other words, the researcher must identify comparable independent companies and analyze their financial indicators to determine the arm’s length profitability range. This process generally consists of two major steps:

• Creating initial dataset using specialized databases like Orbis, Amadues, Compustat, and filtering data based on several criteria like Industry classification, Geographical market, company size etc.
• Refining this dataset to ensure comparables have same business, function and risk profile.

The second step involves manually researching each company in the dataset. This process often includes reviewing the company’s website to assess key business parameters such as market conditions, business strategies, and geographic relevance. Based on this analysis, a decision is made on whether the company is comparable to the analyzed transaction.

In other words, the researcher must gather relevant information about each company and make a professional judgment on its level of comparability.

To emphasize the importance of manual research in a benchmarking study, it is crucial to highlight that decisions on whether a potential comparable is included in the final dataset directly shape the margin range considered arm’s length. This margin, in turn, has a direct impact on the price deemed to be at arm’s length.

Manual element in benchmarking study 

The manual element allows for a nuanced understanding of business models and operational strategies, and detailed research of analyzed business situation. At the same time manual element has a lot of drawbacks. In practice those manual searches almost always involve multiple researchers in order to speed up the process, which can bring slight differences in approach and judgment.

The search process is highly repetitive and can take weeks of human working hours. As a result, researchers may experience fatigue, increasing the likelihood of errors.

In practice, these challenges often lead to the need for revisiting completed work or even disputes between companies and tax authorities, as both sides may challenge each other’s benchmarking studies—and, consequently, the arm’s length price determination.

The industry’s response to this challenge has been the unification of data and multiple layers of verification for both information and its sources. Major statistical data providers have developed their own manual search results and strive to keep their company descriptions up to date.

While these efforts are valuable, a transfer pricing professional can never rely solely on this information for two key reasons: (1) company descriptions may not be specific enough for the analyzed transaction, and (2) data must be updated as of the date of the benchmarking study. These factors ensure that the manual component remains essential, requiring significant human involvement.

AI Automation of manual search 

With the advancement of AI models, there is now a practical possibility to automate the manual search process in benchmarking studies. While maintaining the core principles of benchmarking, AI has the potential to replace human involvement in this aspect of the process, significantly improving efficiency and accuracy.

Aside from freeing humans from a highly repetitive and often undesirable task (trust me, I’ve been a junior consultant working in a major transfer pricing practice), AI-driven manual search can bring several additional advantages to the transfer pricing analysis process:

• Consistency: An AI model’s performance tends to be more stable compared to humans, who may become fatigued by repetitive tasks.
• Elimination of Variability: Traditional benchmarking studies often involve multiple researchers, which can lead to slight differences in the applied approach. AI can fully eliminate this variability.
• Speed: The process can be completed much faster with AI, significantly improving efficiency

Formulating a task for an AI model makes the process highly technical and eliminates potential distortions caused by human perception. The task becomes clear and objective, similar to writing a computer program for the AI model. If a human were to execute the same algorithm, the instructions would be clear enough for someone with no prior experience to understand. In other words, an AI model would perform the task with the lowest possible bias.

In the work of professionals, bias often arises from their experience and the tendency to make judgments about the source of information or the description of a company before completing the study. This can sometimes lead to the exclusion of unique or emerging businesses from the set of comparables, even though they may, in fact, be relevant comparables.

Indeed, assigning manual search in a benchmarking study to an AI model is a highly challenging task for several reasons:

• Rapid Development of AI Models: The fast-paced development of AI models results in frequent changes to their behavior, making it difficult to predict their performance over time.
• Hallucinations of AI Models: AI models sometimes provide incorrect answers or generate information that does not align with real-world facts.
• Stability of AI Model Performance: The consistency and reliability of AI performance can fluctuate, which is a critical challenge for integrating AI into professional tasks.

These challenges are widely recognized as industry-wide issues for AI projects based on large language models.

Quantum TP (https://www.quantumtp.xyz/)  is a California-based company focused on automating transfer pricing processes, with its benchmarking study AI assistant being one of its first products. Our testing results have shown a significant increase in the speed of performing benchmarking studies, allowing the entire process to be completed in approximately one hour. We are currently testing its tools and comparing the results to human-made benchmarking studies. Quantum TP has confirmed that our approach is effective: large language models (LLMs) are capable of performing the manual search required for benchmarking studies. Our company is now offering a beta version of its product to customers for testing.

I built my career in transfer pricing from intern in a major consulting firm to transfer pricing manager in big corporations. I could feel myself how exhausting benchmarking study is for junior consultants, who have to work weeks doing just one thing – searching companies in the internet, writing descriptions of their business, and documenting results. I also know how challenging it is to manage junior consultants doing this job, keeping their motivation, and maintaining quality of the search. I also know how important a benchmarking study is in the transfer pricing analysis. I have seen a number of examples where tax authorities challenged benchmarking studies, and accrued significant additional tax. In light of all that I am excited to create a tool that would help automate the process and help in the best possible way to make a right decision when choosing transfer pricing strategy for a company. I believe that AI will make work of transfer pricing professionals less routine, efficient, and will let them concentrate on creating the best transfer pricing strategies, ensuring every country receiving fair portion of taxes.

Written by Nikolai Portnov.
Have you read?
The World’s Best Medical Schools.
The World’s Best Universities.
The World’s Best International High Schools.
The World’s Best Business Schools.
The World’s Best Fashion Schools.
The World’s Best Hospitality And Hotel Management Schools.