The rise of tax technology owes much to the development and implementation of AI technology. The digitalization of the economic transactions taking place on cloud platforms brought changes to our workplace and daily lives. The digital economy also brought an unprecedented amount of information previously unavailable to everyone. Many tax authorities and taxpayers are moving toward using technology solutions to address the challenges and opportunities brought by these changes.
My tax technology journey starts with transfer pricing, since much of the typical transfer pricing work involves tailored data analytics and visualization. For technical correspondence with tax authorities, transfer pricing specialists also provide econometric-based statistical analyses. For multinational companies, transfer pricing is a highly visible area of tax with governments agreeing to exchange data, and as a result, many potential technology applications in this area. It is therefore desirable to establish modern, data-driven transfer pricing management in-house with a systematic overarching design. The diagram below helps to demonstrate how different pieces could fit together:
Transfer Pricing Data Store
Much of the AI technology involves teams building models, training, and testing data to validate model results. The bulk of the teams’ time and efforts are spent on improving the code, the model, or the algorithms. However, when tracing back the root cause of the issues, the underlying data is often the problem.
Data is the foundation for any transfer pricing technology solution, and in this regard, for any tax technology projects. Good quality data needs to be consistent, complete, timely, and accurate. A secure, clean, and efficient database is fundamental to successfully implementing any technology solution in transfer pricing. For transfer pricing and tax specifically, such data system also needs to be repeatable and systematic to answer potential audit queries or reproduce retrospective results on demand. To find the right type of system for any transfer pricing technology application, one has to consider many aspects of the requirements, for example:
- the type of data one may use currently and in future, e.g., numerical, textual, multimedia;
- the quality and consistency of data;
- the remediation workflow if needed to fundamentally improve data quality;
- for prediction or classification models, the kind of feedback your data could provide;
- who by and how the data should be managed and maintained in the long run;
- what querying mechanism, including the frequency, one may require;
- the desired data structure or data model; and
- where the data should be stored
I termed the database a “transfer pricing data store,” but its actual form and substance would depend on specific business requirements, and to a large extent, the existing enterprise resource planning (ERP) system. Broadly, a transfer pricing data store needs to satisfy both back-end and front-end requirements. As tax professionals, we often have specific front-end requirements: the transfer pricing data store should enable a series of workflows on technology applications, from data collection to transformation, reporting, analysis, review, approval, submission, and archiving. Based on the relevant requirements, tax professionals may also need to work with IT professionals in many areas, such as database architecture, information security, and platform and application services, to complete the back-end requirements.
In addition, the reason I described it as a “store” is because ideally the transfer pricing data is organized in a modular way, so that each component could be used and reused independently and be available on demand. “Independently” could be interpreted in multiple dimensions. For example, different users with access to the transfer pricing data store could access the same attributes built by others; or one could query the same attribute over multiple years to create a tailored trend and anomaly analysis. Consequently, careful design with a holistic view of the database needs to be communicated clearly between the tax and IT professionals.
Transfer Pricing Technology Applications
With the transfer pricing data store in place, we will have a closer look at different technology applications available in transfer pricing. There are broadly five main areas of transfer pricing technology applications:
- transfer pricing strategy
- transfer pricing audit and rulings
- transfer pricing implementation
- transfer pricing documentation and benchmarking
- transfer pricing operation
The last two categories, “documentation and benchmarking” and “operation” in transfer pricing, typically rely on structured data, i.e. data in pre-defined formats, such as the ERP system data. Some examples of technology projects in this area include automating the transfer pricing documentation report generation process using robotic process automation (RPA), and automating transfer pricing data extraction from a centrally managed data lake for further analytics and reconciliation. Clean, good quality data at source remains the biggest challenge in this type of application.
For the first three categories, on the other hand, much of the data remained unstructured and may require much more manual efforts. For example, to design and implement a web-based application to monitor transfer pricing policy exceptions worldwide may utilize no code tooling such as Microsoft’s Power platform, to automate the process, even with an audit trail of every action that took place. Nonetheless, during this process one would still rely on human judgments for the final review and approval.
It is certainly possible to build some natural language processing elements into the process of reviewing the unstructured data, but the built process itself requires subject matter expertise to identify domain-specific terms and concepts, as well as meaningful interpretation of the results. The biggest challenge in this type of application may be noisy, messy data and difficulties involved to meaningfully extract entities and relations within the vast number of transfer pricing-related files and obtain insights from the extractions.
In this article, I discussed technology in transfer pricing in two main parts: data and applications. From experience, to set up a transfer pricing specific data store could be very difficult, given transfer pricing data is often nested within several ERPs, from finance to supply chain, requiring buy-in and coordination from several teams. As a result, a top-down approach from prototyping specific business requirements may prove the better option to acquire stakeholders’ buy-in and scale up thereafter. As we become more familiar and comfortable with using technology solutions to address various challenges in transfer pricing, the most important success factor for a transfer pricing technology project would be high quality and consistent flow of data. The true power of AI is yet to come.
This column does not necessarily reflect the opinion of The Bureau of National Affairs, Inc. or its owners.
Shan Sun worked in a variety of tax areas for 14 years before joining Unilever in 2019 as the Lead for Tax Innovation. She has a particular focus on data and technology in tax. Shan is a member of ICAS and CIOT. She is also a member at the CIOT Commerce and Industry Group and Women in Tax in London.