Benchmarking—the process of screening, selecting, and analyzing comparable companies—is time consuming. Analysts can spend innumerable hours every year preparing transfer pricing documentation, with a substantial portion of that time dedicated to benchmarking. Even with improvements in the quality of databases (which offer a vast array of quantitative and qualitative data), the sets of potential comparables that analysts must sift through are often enormous.
With the applications of artificial intelligence (or “AI”) expanding by the day, it is time to start thinking about whether AI could automate parts of the benchmarking process.
Today, human analysts select comparables based on a multitude of factors, some of which cannot be reliably applied with the “screens” built into current databases. A case in point is the business description of a potential comparable company: descriptions in the databases frequently lack the desired granularity, are incorrect, or are simply missing. And business descriptions for private companies are often completely devoid of specific information on the companies’ functions and assumed risks. For these reasons, analysts usually must review the companies’ annual reports, or at least the information presented in the companies’ websites, to perform the benchmarking.
Parts of this process can already be automated. For example, using Visual Basic for Applications (“VBA”), an event-driven programming language developed by Microsoft, a transfer pricing professional can build customized applications to obtain a series of hyperlinks for company websites from an Excel worksheet, open them in a browser, and copy the opened pages into a PDF file. All that is left for the analyst to do is review the saved pages.
Benchmarking in the Future
While such applications, without a doubt, are of immense value to transfer pricing analysts, they are not AI. A program that would apply AI to benchmarking would need to be able to actually analyze the companies’ websites: it would need to replace, at least to some extent, the task performed by the analyst. An AI system in transfer pricing would comprehend the information presented in the companies’ websites and make some determination as to the comparability of this information to that of the tested party. This would be no small feat as, quite regularly, human analysts do not agree on the comparability of companies.
But there is reason to believe that, at least in the future, AI could perform these tasks. In litigation today, attorneys use AI systems to analyze large volumes of documents. These AI systems “score” the documents for their relevance to the subject matter of the litigation, often with very limited input from humans. Documents with a higher relevance score may be sent to the attorneys for their review—or they might even be produced to the opposing side without an attorney ever having looked at them. It is easy to see how the same technology could be applied to aspects of benchmarking (e.g., analyzing business descriptions on companies’ websites).
Indeed, the real issue might not be the technology—which is already quite advanced—but rather explaining it to taxing authorities and courts. For instance, in applying the Comparable Profits Method or Transactional Net Margin Method, being able to describe and re-create the search parameters and process is just as important as the search results themselves. How would one explain the AI’s selection process? If the AI were a “black box,” a judge or a wary revenue agent might not accept it.
Despite these obstacles, it is likely that AI will begin to make inroads into the transfer pricing world in the future. And the benefits for transfer pricing processionals could be significant. Although AI might eliminate some routine tasks, it will also yield time savings and allow analysts to focus their energy on the judgmental aspects of transfer pricing that a machine won’t be able to replace.