How do you leverage AI and automation features in TP software to enhance efficiency and quality?

It is also important to highlight that TP related data might come in different formats and be compiled using different methodologies. Given the international aspect of TP, it is more likely than not that let's say transaction volumes can come in different curriencies (without it being indicated on a document itself, which might interfere with the algorithm). The key to success would be good data labels, most likely created by someone who's seen the formats that TP data comes in and make sure that the labels account for this.

Here are some ways in which AI and automation can be applied: 1. Data Analysis and Benchmarking: ? Automated Data Collection: ? Benchmarking Analysis: 2. Risk Assessment: ? Predictive Analytics: 3. Documentation and Compliance: ? Automated Documentation: ? Compliance Tracking: 4. Intelligent Audits and Reviews: ? AI-Powered Audits: ? Automated Review Processes: 5. Risk Mitigation and Scenario Analysis: ? Scenario Modeling: ? Automated Compliance Checks: 6. Machine Learning for Continuous Improvement: ? Machine Learning Algorithms: ? Adaptive Documentation: 7. Cross-Border Compliance: ? Automated Currency Conversion: 8. Data Security and Privacy Compliance: ? AI for Compliance Checks:

Using a TP software will add significant value in performing any type of TP analysis! However, it is important to note that most businesses have become more dynamic than ever, especially in the last decade or so with the advent of novel technologies. Dynamics lead to complexities that need to be fully understood to evaluate the usefulness of any technology used to perform TP analysis. The process followed in the prior years may not be the same due to changes that a business have undergone which may warrant revisiting vital elements of transfer pricing analysis - e.g., selection of method, selection of tested party (and therefore the tested party financials being tested), introduction of new revenue / expense items, etc.

Choosing appropriate TP methodology is highly subjective in nature. For intercompany transactions that are similar in nature, multiple methods may be applicable depending on a number of factors. E.g., in a transaction involving sale of products from a related manufacturer to a related distributor, either the distributor, the manufacturer or in some situations both may be used as tested parties depending on the functions, assets and risks profile of the related parties (a decisive factor in choosing TP methods) and as a result multiple TP methodologies may be applicable. Imho, AI/automation may work best for testing and compliance but only after defensible TP policies are first established which perhaps require human expertise.

The process of selecting a TP method is automatic to some extent, but also needs to take into consideration other aspects, such as data availability for each of the methods, which would be challenging for the taxpayer, as we wouldn't know what was the basis of the selection for the companies that made it into the training dataset (unless it was a big set with such details provided). Sometimes the most appropriate method isn't really one, because of the data availability. Having said that, if one was to create a unified machine learning approach that both taxpayers and tax authorities would use, who would create this model? We could go even more meta and have taxpayers and tax authorities have their own separate models.

As for the benchmarking, I agree that this could be an automatic process and should require much less legwork than it does now. But again, would it remove room for discussions on the appropriateness of comparables between the taxpayer and tax authorities?

Every time I prepare a piece of standard TP documentation, I ask myself why is this not more commonly automated yet. A model could be trained to scan the documents and recognise relevant information to populate the TP documentation with. It would remove the burden off the taxpayers of preparing something that should have been automated a long time ago.

This stage is perhaps where automation / AI features can come in extremely handy. As noted by Zuzanna below, I ask the same question myself - why isn't TP documentation automated? I've created numerous templates and processes in the past to increase efficiencies but I suspect automation will further increase efficiencies. Hours spent into writing and formatting standard reports can be easily saved with automation (welcome ChatGPT and perhaps Bard once I get to use it!). Having said that, I'll be remiss to not mention the need to constantly monitor the changes happening within a taxpayer's business which may have an impact on the transfer pricing arrangement of the taxpayer, and as a result may change certain elements of the documentation.

Absolutely. A lot of companies, like Klarna, have used ChatGPT APIs to create their own products. A TP Chatbot would be extremely helpful to tax teams that do not have a TP professional on board.

100% agreed that AI will reduce efforts for taxpayers in maintaining compliance. However, I cannot comment on the potential reduction in cost at this point of time. Simply going by the market rates for AI/Automation experts and the speed at which technology is evolving in general, the higher costs are simply going to be transferred to the user. Hopefully, with time and increase in the user base, it will be cost effective. I guess we'll figure it out with time.

Software can only be as good as the developer's inputs. There are consultants/taxpayers that build their own TP software, and there are tech companies that build software and license it to consultants and taxpayers. In any case, whoever is building the software needs to monitor the changes and constantly update the software to stay current per the market trends. Similarly, the user of the software needs to be aware of the the changes in tax landscape to evaluate the effectiveness of the software that is being used.

I wonder if the task of evaluating TP software's efficiency shouldn't be responsibility of the software provider. Overburdening tax teams could have a contrary effect.

It is helpful to do forecasting, but how can we really predict changes in TP policies? Once a draft is published, like in Pillar 2's case, that task becomes easier. A predictive model which forecasts a potential change in regulations would have an insane amount of features as well as it would be hard to compare its accuracy to the baseline. It's possible to do, but not sure how much worth would be in acting in accordance to this data.

AI in TP is absolutely something to be considered, but in my view at this stage it should be the compliance and mundane tasks that should be automated in the first place, such as Master Files, Local files or benchmarking. The problem with automating the rest is of similar nature to a problem with automating lawyers or other advisers. The best answer sometimes is not the right one.