Desktop version

Home arrow Business & Finance

  • Increase font
  • Decrease font

<<   CONTENTS   >>

Case Study #2: Use of AI for Tax Transfer Pricing Services (KPMG)


As part of its Al strategy, KPMG Ignite is a global Al platform designed to accelerate the development of internal and client solutions focused on unstructured data (text, voice, and image) problems. It brings together KPMG professionals' in-depth domain knowledge with a toolbox of capabilities from open source, strategic technology alliances, and KPMG-developed IP. KPMG Ignite applies Al-based automation patterns to create intelligent workflows to solve business problems made complex by unstructured data. Ignite is deployed at multiple KPMG member firms to serve client bases globally. It is a containerized platform solution that provides benefits of scale and flexibility for applications supporting various business issues across multiple industries.

The KPMG Ignite platform is used by professionals of KPMG's Global Transfer Pricing Services (GTPS) practice to provide Al solutions for member firms' tax clients, specifically for transfer pricing. Transfer pricing refers to the rules and methods used to establish prices for goods and services sold between enterprises under common ownership or control. According to Komal Dhall2, Global Leader, KPMG Global Transfer Pricing Services, one of the challenges involved in the transfer pricing process is benchmarking relevant market prices. By using Al, Dhall explained that they have successfully trained the KPMG Ignite platform to read subscription databases, analyze financial statements, and read company websites to assist with benchmarking market prices for transactions.


The application of Al saves time while simultaneously increasing quality and consistency. “This type of qualitative review for humans is incredibly time-consuming and labor-intensive, but obviously the machine isn’t limited in that way,” said Dhall. “So what would take typically several hundreds of hours is now shortened into a simple,

Contemporary Case Studies 75 manageable timeframe.” Consequently, tax transfer professionals “can spend time on the qualitative results and identify quality benchmarks,” said Dhall.

According to Thomas Herr, Principal and National Leader, Innovation, of the Economic & Valuation Services Practice of K.PMG LLP, the primary driver of Al implementation for transfer pricing is not cost savings. “The bigger benefit will be the improvements in quality and consistency,” said Herr.

Herr also noted that in addition to addressing quality and consistency with transfer pricing, they want to change the nature of the process fundamentally. “Because of its manual nature, there is a well-established process in place to winnow down the amount of data that actually has to be reviewed by a human,” Herr said. He added, "But it’s a trade-off: in winnowing it down, you lose information, and you lose out of necessity to gain efficiency.”

It is important to note that Dhall views Al as a tool to complement the work of their professionals, not as a substitute. “We’re not trying to replace humans, but we’re trying to enhance the quality and capability of the human in the loop in the benchmarking process,” said Dhall. “The outcome is intended not only to be a more efficient process but also a higher quality analysis with more meaningful results. For example, the solution, in my view, functions as a single source of history on a many country basis, so this is the tool used globally. It’s used in many countries and territories, helping to make our collective global analysis smarter.”

Lessons Learned

According to Herr, one of the key lessons learned from the adoption and application of Al is good communication between subject matter experts in transfer pricing and the machine-learning specialists. “We had meetings, and things would be explained, and there was a sense that people understood it. And then we would go away and come back together again and realized we need to delve in further to gain clarity,” Herr said.

This challenge was addressed by bringing the teams together more frequently. Herr describes the need “to have tighter integration with teams working much more closely together. Your subject matter experts need to work closely with the AI/ML team to ensure a good exchange of information.”

Another important lesson learned was ensuring the quality of the data. “While we had a lot of data to train the models, a good chunk of that datawas not of the best quality. So, we had to do quite a bit of data clean-up. And that was a larger effort than we had expected,” said Herr. Also, Hen-said that training K.PMG professionals to properly label the data “in a way that makes it usable for machine learning” was a challenge.

Finally, Dhall recommended that companies looking to implement Al start with some well-known, repeatable processes. “We started with something manageable, doable, and something we know we can build on, because it is the backbone of many of our analyses,” said Dhall. In other words, companies should be realistic and start with a relatively small-scale and straightforward project, learn from their experience developing it, and then take on more complex business problems. “From the simple use case, we’re able to look across K.PMG and find different applications for it” said Dhall.

<<   CONTENTS   >>

Related topics