AI benchmark study of transfer pricing
Case 14/16 of applied Artificial Intelligence
If a company forms part of an international group, the prices and conditions applied to the sale of goods and services within this group must be similar to those of third parties. This is intended to prevent improper diversion of profits between countries. However, it saddles companies with a problem, says Martijn Krassenburg, Manager Transfer Pricing at Deloitte. After all, which prices of which companies are deemed ‘similar’?
A benchmark study is required in order to answer that question. One variant of such studies looks at what companies undertake identical activities in a similar industry, and the reported margins are checked against those of the company in question. That is quite a laborious process, in which many repetitive tasks are carried out manually. “This is why we decided to find out whether it was possible to automate this work,” explains Krassenburg.
AI email alert
Get 16 cases of applied AI in your mailboxSign up here
Thousands of screenshots
A benchmark study first examines which companies are similar. An initial selection of a few hundred companies that are possibly similar is made by applying filters to an international database. They have to be screened manually, such as by searching for the websites of all of those companies and taking screenshots of them. Krassenburg: “Robotic Process Automation is the automation of simple, repetitive tasks, which in this case are searching websites and saving screenshots, and it has enabled us to speed up that process considerably.”
But the team’s ambitions extend further. “We are busy developing artificial intelligence that will automatically estimate the extent to which a company is similar,” he adds. The system is already able to perform rough screening, and its self-learning ability means it will become increasingly accurate the more it is used. The technology is now being trained using large quantities of data from previous benchmark studies. and technologies such as natural language processing, neural networks and ultra-precise entity recognition algorithms are being employed for this purpose.
Soon, a percentage will appear next to the name of each company, indicating the likelihood of that company being similar. This will mean that an ever-smaller number of companies will need to be checked manually. Ultimately, the aim is to complete part of the task fully automatically.
The finishing touch
The project is still in the testing stage, but Krassenburg believes that the technology can be rolled out widely once it is up and running. “In this way, you can test how the intercompany prices relate to the market, and substantiate this with detailed documentation. All internationally operating companies that sell products or services are affected by it. In this regard, AI Benchmark helps them maintain consistency and is already saving them time.” Deloitte Global has already expressed an interest in this tool. As soon as AI Benchmark can be implemented, it will be employed more widely.
When it comes to the question of whether AI Benchmark will render his own job surplus to requirements, Krassenburg is not concerned. “Performing a benchmark study is not what creates our added value. It is the finishing touch, one part of a much larger and more complex process. The really interesting work, understanding exactly how a company functions and how it relates to tax rules, is not something that can be simply outsourced.”