How Can GenAI Improve My Transfer Pricing Process?

Can generative artificial intelligence (GenAI) improve my transfer pricing process?

The answer is obviously yes. Right? I think so. Yes, definitely…but…how exactly?

These may be some of the thoughts running through your head when you hear about how GenAI and all its new glory is transforming the business world. It seems obvious that the latest technology, the one everyone is talking about, must be able to benefit you in your day-to-day tasks and operations. But when it comes down to really figuring out how GenAI can practically solve your real-world problems, sometimes it is hard to know where to start. The good news is that the fundamentals of process improvement can still be applied, with GenAI serving as an additional tool to help eliminate “waste” (i.e., any activity that does not add value) and optimize your transfer pricing processes.

GenAI is an evolving and dynamic technological force that is quickly changing the way we work, however it can still be unclear how to best integrate its strengths effectively into your business. Using a process improvement approach is an excellent way to identify where the strengths of this new technology can be aligned to meet the pain points in existing transfer pricing processes and ultimately increase efficiency. In this article, we discuss some of the questions and points to consider when using GenAI in your work. We suggest a three-step approach: Define, Discover, and Solve. We also highlight some of the struggles and the opportunities that GenAI presents.

What Is GenAI and Why Is It Different?

GenAI, at its core, is a system capable of creating new content, predictions, or solutions based on the data it has been fed. It’s a game-changer for tax departments because of its potential to streamline operations, augment efforts on rudimentary tasks, and make sophisticated predictions that can inform human decision-making. However, these advantages are balanced by challenges such as ensuring data quality, managing system complexity, and aligning AI capabilities with departmental objectives and processes. In addition, since GenAI is a probabilistic system, versus deterministic like most other technologies used in the tax department, there are unique considerations in where and how to integrate GenAI into your transfer pricing processes.

The fact is, GenAI is actually a big term that encompasses a lot of tools and capabilities. And while that might be overwhelming or intimidating at first, one thing that differentiates GenAI from other powerful technologies is that to get started, there are relatively low barriers to entry for a majority of users. The ability to interact with GenAI tools in one’s “natural language”, that is, users are not required to learn a new programming language to start using the technology, means that almost anyone can pick up GenAI tools like ChatGPT, Gemini, or CoPilot and start using them intuitively with very little experience to perform simple tasks that provide marginal benefit. Greater and more complex problems can also be tackled using Gen AI, but requires more concerted efforts put into the training, prompt engineering, and thoughtful integration. Just how much to invest in these efforts versus the potential benefit they can provide is a strategic decision that tax departments, with increasingly limited resources, are having to make.

Understanding both the capabilities and the limitations of GenAI tools is important as you think about building this technology into your transfer pricing processes. If the addition of GenAI is not improving the quality or efficiency of the process, are you really realizing a benefit? Applying a process improvement framework to identify how and where (or where not) to integrate GenAI into a process can help to make these decisions.

Process Improvement Approach

As long as there have been operations and processes, there have been opportunities to enhance them, and a plethora of methodologies to be deployed in order to realize these improvements. There is not a “one size fits all” solution. Various methodologies can be applied to varying degrees of success, dependent upon the facts and circumstances of the process, and the intended improvement or outcome. For example, Agile evolved to meet the needs of software developers, and is best suited to smaller teams working on flexible, dynamic projects, while Six Sigma aims to minimize defects, optimizes consistency, and is most often used in manufacturing. Lean methodology was developed to minimize waste, with a particular focus on ensuring that each step in the process adds value to the ultimate customer. While different approaches are better suited to different scenarios, in transfer pricing and other professional services, using combinations of these different approaches, including Lean Six Sigma (which is marriage of the value-focused Lean and defect-minimizing Six Sigma techniques) and Design Thinking (which takes a client/user centric approach to kick off a cycle of design, prototype, and test collaboratively, and is an ongoing exercise), are best suited as a framework to rapidly get started on your process improvement journey.

Define, Discover, and Solve

The fundamentals to many process improvement approaches can be summarized into three general phases: Define, Discover, and Solve.

Define: Identify the process, problem, or opportunity that is to be addressed, the scope and objectives of the project, as well as the stakeholders and resources involved.
Discover: Collect and analyze data about the current state of the process, identify the root causes of the problem or opportunity, and brainstorm potential solutions.
Solve: Evaluate and prioritize the potential solutions, design and test the future state of the process, and implement and monitor the changes.

Within each phase, there are a variety of tools and techniques that can be used to meet each objective, depending on the nature of the project or problem. For example, you can use a Project Charter and a SIPOC diagram to summarize the key information about the project and the process in the Define element, or you can use process maps and fishbone diagrams to visualize and analyze the process flow and the root causes of the problem in the Discover element. The choice of tools and techniques is not fixed, but rather depends on the situation and the preferences of the team.

Project Charter

SIPOC is an acronym for Suppliers, Inputs, Processes, Outputs, Customers. The SIPOC helps describe and gain consensus on high-level processes, and identifies who the Suppliers and Customers are in the process.

Process Maps, such as this example swim lane process map, visually illustrates the process step by step, with a distinct “swim lane” row for each individual and/or team.

Fishbone (or Ishikawa) Diagrams can be used to investigate complex problems that may have multiple root causes, and helps illustrate the relationship between the outcome and the factors that influence it.

In addition to these elements, there are some other factors that are essential for effective process improvement. These include:

The presence, involvement, and support of management and leadership;
The inclusion and participation of the individuals closest to the work during the analysis and improvement stages;
The alignment of the process improvement goals with the strategic goals of the organization; and
The continuous monitoring and evaluation of the process performance and the outcomes of the improvement.

By using a process improvement approach, you can more effectively identify and address the problems and opportunities in your existing transfer pricing process highlighting areas where GenAI and other technologies might enhance the future state process. While there are ways to deploy GenAI to more rapidly facilitate your process improvement approach, that is beyond the scope of this particular review, which is instead focused on how to use a traditional process improvement framework to determine where to optimize the use of GenAI in existing transfer pricing processes.

GenAI: Another Tool in the Belt

As you proceed through the process improvement framework, you will likely find that GenAI is not an all-encompassing solution, but rather one optional tool (amongst many others) that can help you improve certain aspects or tasks within the existing process. During your review of the current state process map, you will undoubtedly identify areas of waste in your process that could be reduced through the automation of manual tasks, or significant timing delays due to resource capacity. (To hear what GenAI use cases are top of mind for transfer pricing professionals, check out KPMG’s podcast Will GenAI Revolutionize Transfer Pricing?.) For instance, when looking at the process for conducting interviews for management service allocations, you may identify that significant manual effort is being exerted to summarize and transform the information collected through the interviews into a usable format that can be evaluated and incorporated into a cost allocation model. In the same process, you may identify waste related to resource capacity where a single person is the bottleneck for conducting functional interviews with dozens of individuals.

Identifying first the nature of the waste, and then further defining the root cause, can lead to more robust solutions that can better align the strengths of the technology to the weaknesses in the system. This approach also allows you to take small, manageable steps in the adoption of GenAI. GenAI is not a magic wand that can do everything for you, but a tool that can augment your own skills and knowledge. For example, in the management services allocation process, where the time spent summarizing notes and the bottleneck of performing one interview at a time are two areas of waste, you may determine that the natural language chat capabilities of GenAI may be a good replacement for the human interviewers, conducting interviews simultaneously across multiple interviewees with the ability to quickly summarize the results into a defined and useable output.

When determining the best solution to meet the needs of the process you are aiming to improve. GenAI has its own limitations and challenges. Depending on the use case, GenAI may require a large amount of high-quality data to produce the desired output, requiring you to ensure that the data that you feed to GenAI is accurate, complete, consistent, and relevant. GenAI is also a predictive (versus predetermined) system that can generate different outputs depending on the inputs and the parameters that you set. Challenges with incorporating and working with GenAI compared to other types of technology and automation is this variation in outputs. Process improvement is about maintaining a level of quality, and one of the major impediments to quality is process variation. While output variation can be beneficial in creative endeavors such as brainstorming, users and organizations should be prepared to set strict guidelines when operationalizing GenAI within a process that requires consistency. In certain situations where consistency is key in upholding quality, other types of automation technology may be more effective.

GenAI is also not a substitute for human judgment and expertise, and you need to ensure that you review and verify the outputs that you receive from GenAI are valid, reliable, and aligned with your objectives and expectations. The same level of professional skepticism should be applied to GenAI-generated content as would be applied to content generated by a human member of the team. Considering this, you’ll want to ask yourself: are you spending as much time reviewing and correcting AI hallucinations as you were creating the content to begin with? Do you find the need to recreate or fact-check the content because there is no clear audit trail of where things came from? These are issues that can be improved with more complex prompt engineering (e.g., requiring citations and sources, asking to “show your work” for advanced problems, or building a multi-step prompt that includes the GenAI self-checking its work), enabling the opportunity for scalable solutions that optimizes human efforts in the process. However, it should be acknowledged that developing and verifying more advanced GenAI solutions require an investment at the outset, which is an important ROI consideration when evaluating the use of GenAI to reduce waste in your process.

Therefore, when deciding where and when to incorporate GenAI in your transfer pricing processes, it is worth considering the following questions at each stage of the process improvement program:

Define:

What GenAI tools are available to you, and what are their capabilities and limitations?

Discover:

Are there tasks or activities in your transfer pricing process that are falling behind due to delays or dependency on a particular resource?
Where might technology solutions reduce each area of waste?
What are the expected benefits and costs of using GenAI for each potential solution?

Solve:

How will you integrate GenAI into your “ideal state” transfer pricing process, and what are the changes and adjustments that you need to make to your existing process, data, and systems?
How will you test and evaluate the quality and the relevance of the outputs that you receive from GenAI, and what are the criteria and the standards that you will use?
How will you use the outputs that you receive from GenAI, and what are the roles and responsibilities of the human and the AI agents in the process?

The answers to these questions can assist with determining how to deploy GenAI effectively and efficiently, and avoid wasting time and resources on tasks that are not suitable for GenAI or that will not result in reducing waste or improving the quality of your operations.

What to Keep in Mind When Adding GenAI to Your Team

Once elements of an existing process that align with the strengths of GenAI have been identified, you can begin training GenAI as if it were a new team member in order to transition these activities. This approach can eventually significantly alleviate the strain on resources, however, it should be acknowledged that this shift is rarely automatic. Just like any new team member, the GenAI resource must be directed with adequate prompts and context for the activity that it’s being asked to perform, and will continue to require thoughtful oversight. Responsibility for any defects remain with the human members of the team.

While there is an opportunity for GenAI to assist in making transfer pricing processes more efficient, being aware of and managing new types of waste that might arise from GenAI usage is essential. For instance, given the ease and volume with which GenAI is able to generate content, there is a real risk of potential overproduction and defects (e.g., from hallucinations of false information, or generating large volumes of repetitive or redundant data). The decreased barrier to preparing copious lines of lengthy written descriptions of transactions and functional analyses that have made bold assumptions and taken creative liberties based on fairly few facts can yield more paragraphs than a discerning human may care to inspect, running the risk of review fatigue, with a potential for decline in accuracy. Adding GenAI as a new member of your team will require additional checks, but these checks should not add to the original problem or ultimately consume more time than the current state process (after a reasonable window of time taken for development and training).

Returning to our management services example, we considered GenAI to address two areas of waste: the bottleneck and the time spent summarizing notes. Now if the summaries that GenAI creates are inaccurate, filled with hallucinations, and inadequately sourced, then a reviewer may spend more time after the fact figuring out what is factual compared to having had a single individual perform each of the interviews that can more quickly confirm and sign off on the truth / relevancy of what’s being presented. Employing GenAI should be a calculated step that adds real value, not a hasty decision. By following a process improvement approach, you have the opportunity to identify and proactively build in these safeguards into your future state.

Incorporating GenAI also presents unique challenges compared to other automation technologies. Maintaining consistent quality is key in process improvement, and GenAI’s generative nature introduces a risk of process variation. To mitigate this, establish strict operational guidelines, including ensuring a “human-in-the-loop” to review inputs and outputs for GenAI in processes requiring consistency. In some cases, like the calculation of transfer pricing adjustments, given the technology at present, traditional automation technologies might be more effective. Evaluating these factors during the Discovery phase can guide you towards the most suitable solution, ensuring the successful integration of GenAI into your transfer pricing operations.

Conclusion

We hope that this article has sparked your curiosity and creativity, and that you will continue to explore and experiment with GenAI in your business. Remember, GenAI is not a destination, but a journey.

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

Amanda Carey is a managing director and Brittany Hardin Tanguay is a senior manager with Washington National Tax.

The information in this article is not intended to be “written advice concerning one or more Federal tax matters” subject to the requirements of section 10.37(a)(2) of Treasury Department Circular 230. The information contained herein is of a general nature and based on authorities that are subject to change. Applicability of the information to specific situations should be determined through consultation with your tax adviser. This article represents the views of the authors only, and does not necessarily represent the views or professional advice of KPMG LLP.