As artificial intelligence (AI) gains traction throughout the business world, tax is in the thick of it, aiming to take advantage of AI’s benefits while navigating the attendant complexities. At TEI’s Tax Technology Seminar held this spring in Orlando, one session spoke to those issues at length—an engaging conversation and one of the big highlights of the event. Moderated by Belinda Buvens, senior director of US indirect tax and global tax technology at Procter & Gamble, the panel was rounded out by Anli Chen, managing director of tax automation and innovation at BDO USA; Neal Schneider, chief technology and product officer at K1x; Nancy Hawkins, vice president of product management at Thomson Reuters; Benjamin Alarie, CEO of Blue J and the Osler Chair in Business Law at the University of Toronto; and Jamie Eagan, vice president of tax product management at insightsoftware. With the exceptions of Buvens and Eagan, the panel recently came back together for a follow-up conversation, which was moderated by Sam Hoffmeister, Tax Executive’s managing editor.
Sam Hoffmeister: Thank you all for joining the discussion today. What are you seeing as the three greatest barriers to AI implementation in the tax world, and why?
Anli Chen: In my view there are three I see as the largest barriers here. One is on the data quality of tax functions. I think the amount of information, the quality of the data, and also the disparity of it and the compression of the information coming through into the tax function, all different dimensions of the data that the function needs to have access to, are just very, very challenging for AI implementation. Oftentimes that information comes in on an ad hoc basis; the quality needs to be enhanced by the tax professionals who know exactly what they are looking for. The access challenges are everywhere for tax professionals because they know exactly what they’re looking for, but [are] not necessarily always ready. So, these data quality and accessibility challenges are what I see as the first and key challenge that our tax professionals are dealing with almost on a daily basis. The second one is the regulatory landscape. In the AI world, that means you basically are implementing a solution with a moving target. On one side you have built out all the training, all this history, that you’ve put into your AI model so you can leverage the AI technology to help you get to the answer. On the other side, the regulation’s changing regularly; you have a lot of unknowns, which is another norm of the tax professional’s day-to-day. And so the big disconnect between the need of training AI versus the dynamic, let’s say, of tax technology is really what I see as another barrier for the implementation. And the third, coming directly from this client session I’ve been sitting in this week in Nashville, is just our tax professional responsibilities and the reality of day-to-day. The amount of work they are managing is really preventing tax professionals from thinking strategically, from upskilling strategically, simply spending more time with exploring the new technology. And that’s what I see as another key barrier. It definitely requires our tax professionals to think more broadly and more analytically and also with more open ideas, and that’s what I see as a key for implementing AI, but unfortunately oftentimes we see tax professionals do not have that luxury of exploring because of the amount of responsibilities on their hands.
Neal Schneider: I one hundred percent agree with Anli and what she laid out there. To piggyback on the data quality piece of her response, with disparate systems and multiple source formats you need standardization of data to ensure proper AI training and quality. Many times, especially with regulatory compliance that she’s talking about, there’s different interpretations of results, different methodologies, processes, etc., across different facts and circumstances of clients, this is hard to standardize. But in the same vein, you’d argue that’s what AI should help you with as well, right? It should be able to look at all these different permutations, fact patterns, etc., to help you get to some standardization. There are just multiple steps along the way, where maybe you’re not using AI to get to the answer or a solution that you’re trying to hit on immediately tied to some tax problem or facts or circumstance because of this data quality component of it, but you could actually leverage AI to standardize the data instead. I think one other challenge is the sensitivity around the data that is collected from a tax perspective. It’s private data, right? There’s a lot of sensitivity around it—regulations, etc. And with accounting firms, rightfully so, they have a concern of mixing their clients’ data with others and bringing that stuff into common training sets and training models. This requires infrastructure controls, governance and other components you have to do around protecting data that you may not have the appetite or capability for out of the gate.
Nancy Hawkins: Totally agree with everything Neal and Anli have said. I think the only additional issue I’ll mention is tech debt. A lot of companies are dealing with tech debt and whether or not their systems can be enhanced with AI in the way they want it to be. Making that decision to work with what you have or to modernize technology to better implement AI can be a very costly decision as well.
Benjamin Alarie: One major challenge in AI implementation is ensuring ease of use. Successful AI systems are often characterized by their user-friendliness. For example, at Blue J we’ve seen clients quickly adapt to our solutions because they’re designed with the user in mind. A clear demonstration of this is ChatGPT, which gained 100 million users in a few months due to its simple interface. In the tax context, achieving this ease of use is complex because of data quality and integration issues. However, by involving tax professionals early in the development process, we can create systems that address these challenges effectively. This collaborative approach not only improves adoption rates but also builds trust in the technology. Addressing potential user concerns, such as job security and data privacy, is also crucial. By ensuring transparency and providing clear examples of successful AI integration, we can alleviate fears and demonstrate the tangible benefits of AI in tax functions.
Hoffmeister: Why is it important for tax to have a seat at the table when making decisions about AI implementation?
Schneider: I actually may draw on some of my past experiences working inside an accounting firm and being part of tax technology and tax transformation. I would look at this to draw a parallel to the question of inserting any technology here for that question. Why is it so important for tax to have a seat at the table? Remember in the 2010s it was a lot of RPA [robotic process automation], data analytics–type tools, etc. Putting AI in the context of this question, you get the opportunity to share in those challenges we just discussed with your other departments and colleagues. Those challenges of expenses, infrastructure, data security, etc.—you can share in that with a seat at the table, and then you’re also in that ground zero training ground with the other departments. You can look at their use cases they’re trying to do, draw parallels to that, look at leveraging data standardization, data quality–type tools to help you get to your use cases faster. I think if you go down your own path it makes it a lot more difficult to execute. Tax at the table creates that commonality and centralizes the discussion so the organization can drive forward [with] AI together. Then you can draw parallels on those use cases like I mentioned to then associate that to business value that you can create out of the tax function to then get your priorities, your projects, your things vaulted to the top of the list. I think you can look at AI similarly how you’d look to get on the RPA wagon or the data analytics wagon or other technology. It’s the same concept. There are just benefits there of having that seat at the table to understand, get the education, and share in the journey with your other capabilities within your organization.
Hawkins: The first thing I’ll bring up is piggybacking on what Neal said. It’s an objective of tax departments to be seen as trusted, strategic advisors for the C-suite, rather than simply a compliance department or number crunchers. I think any time there are strategic discussions and decisions to be made at an organizational level, it’s important for the tax department to be there at base to make sure that decisions are aligned with the organization’s tax strategies. And then for a whole host of reasons that Neal went through as well. From a risk management perspective, the tax department should ensure that decisions made around AI or other technologies include mitigations against errors in tax filings, overpayment, underpayment, etc., to make sure that decisions are taking into account the execution of the tax department and their accountabilities.
Alarie: Nancy and Neal have highlighted key points, and I’d like to add the value of involving tax early in decision-making—think of it as measuring twice, cutting once. If tax isn’t at the table, decisions might be made that conflict with tax strategies, leading to costly reengineering later. Ensuring tax has a seat from the start prevents these misalignments and integrates tax considerations seamlessly into AI implementations.
Chen: Just to add two quick points. One is from the tax perspective. It has such deep domain knowledge that really requires thinking and vision and the strategic alignment when any companies talk about AI implementation. Information technology is such a dynamic and specialized area. When the company is going through painting the road map of AI implementation, letting the tax professional think through the lens of their reality is really what I see as the key element for success. And another point I want to add is, as specialized as deeply in their area as they are, tax professionals are also highly integrated with other functions, such as finance, accounting, payroll, or other parts of the organization from the stream of data sources. If we do not have a seat at the table, I really feel we’re missing or potentially are missing the quick wins, easy wins, those items with high ROIs, simply because we’re not yet connecting the dots in the organization. I’ll give you an easy example, where again last week when I was going through an AI session with a client, and we’re talking about the art of possibility of utilizing AI for identifying their sales data pattern and helping them with the quality check of their sales data. One of the team members from finance was saying, “Well, why does tax need this data?” That’s a very fundamental question, and I think that really connecting not just for AI implementation, but also within this organization the data needs from different parts of the organization, and then simply articulating what the tax function’s desired data quality looks like. It does not need a transformative AI investment into solving that type of quick, easy win, because you’re simply connecting the needs with where the data comes from, and that’s a process improvement. That’s a quick win right there.
Hoffmeister: What other challenges in that realm does AI implementation present? More important, how can organizations get out ahead of those challenges and meet them where they’re being faced?
Alarie: There are concerns about bias and fairness when using historical data for AI predictions. In tax, this might not be the most acute issue but is still relevant. More pressing are the skill gaps and change management. Tax professionals need training to bridge their deep domain expertise with AI knowledge. Change management is crucial—organizations must support staff in adopting AI without expecting them to become AI experts. Trusting and carefully selecting vendors, like Thomson Reuters, Blue J, or K1x, ensures security and compliance, easing the transition.
Schneider: Yeah, one hundred percent agree on the change management element to it, again kind of similar to my response on the seat-at-the-table piece, insert any technology here. One of the biggest challenges is change management in the tax function. And I get it. You have to balance “When do I adopt, when do I learn this thing, when do I put it in practice?” and also not impact the deadlines that you’re tied to every single year. There are tight windows of time. So, it’s a planning thing. And then you’ve got to plan appropriately to cross that bridge on that skill gap that Ben talked about as well; completely agree. I think the other piece, too—and this may be a little bit broader, and again, it depends on how the implementation goes in your respective organization—this technology can be expensive. It’s expensive to run it, operate it, train on it, etc. And if you’re not taking the variability of the expense, similar to the “trust your vendor” concept, into consideration, you could break yourself into [operating expenses] jail pretty quickly. You’ve got to make sure there’s a balance there and good monitoring controls around that, because a lot of this AI stuff, just to get technical, it’s not running on CPUs. It’s running on GPUs. GPUs are more expensive to run and process all this stuff so that you get more powerful models, models that take on more tokens. You’ve got to be very conscious of that. Your IT department may help with all that, but it’s just something to be aware of and make sure we’re budgeting and planning appropriately. That might be another item that I would just throw into the mix there.
Hawkins: I think Ben and Neal really handled the responses I had. I’m going to double down on the human component of change management. It’s critical to understand what your people are worried about from a change management point of view. Are they worried about the security piece? Are they worried about maybe introducing more errors? Tax is a profession where your reputation is on the line, and your company’s reputation. Are they worried about that? For some people, are they worried, “Is AI going to take over my job?” To really help your team with the change management process, you have to understand their mindset and concerns so you can address them. This is Change Management 101: find champions on the team to say, “Hey, this is what’s in it for me. This is what’s in it for us.” And finding those quick wins where you can show this lower-value work that no one likes to do can be done by RPA or AI, and the staff can have greater capacity for higher-value work. We’re still talking about change management, but just thinking more on the human side.
Chen: And I think this group also mentioned just now about the challenge of transparency, auditability and traceability, explainability, all these along the lines of the myth of AI being the black box. Searching my memory for all the past conversations I’ve been having about technology, especially the emerging technology in the tax functions, I often get a question of how do I respond to my auditors’ questions: “If they need to see the year XYZ for Q1 and this data set, what is my response if this whole entire process is now automated, all packaged in one perceived ‘black box’? What would be my response?” I think it goes to the consideration of risk management and compliance, and also on the flip side how I would encourage tax functions to address that is, I think, a lot of the key success factors have been mentioned in this group. I’d also like to add that in implementing AI functions, please have the traceability and transparency as the key component for that implementation. The technology nowadays is far from “you package everything up, you throw things in, and magic comes out.” I understand K1x has that reference point, user put in your document, and you can trace back to where the source is, and same for Thomson’s product, right? So, a lot of this AI-empowered tax technology today by definition organically creates that audit trail, organically actually enhances rather than prevents that transparency that’s needed or demanded by the auditor. So, I wanted to point it out there. That’s just a lot of the questions I get from our clients, and that’s a constant education point I would like to put into our tax leaders’ minds as they think about the AI road map and their function.
Hoffmeister: Why are conversations like this and the panel at TEI’s Tax Tech Seminar so important in helping to move the needle on AI and tax transformation in general?
Hawkins: I’m going to quote Anli. I love the “myth of AI as a black box.” For me, one of the most important things is to demystify the technology. These are technologies that, frankly, if not adopted and leveraged in a way to make the tax department more efficient, more effective, better able to make time to be more strategic, you’ll be left behind. Even at the TEI conference, when we were talking in the larger group, there were a lot of concerns around the things that we’ve already talked about today—security, compliance, implementation, change management—a whole host of things—I felt like it was really important for us to note those [are] very normal. People should be cognizant of the associated risks, but also that these can be mitigated risks. Do your due diligence on your vendors so you’re not in it alone. I think an important part of having these conversations is about leveling the playing field. One of the things I loved about [generative] AI when it came out, ChatGPT just democratized that technology instantly. Ben talked about the numbers of users in a crazy short amount of time. I love the fact that my then-twelve-year-old was playing with it before I was. It just democratizes that technology. Hopefully we’re doing our part and doing the same thing here.
Chen: Nancy put it so well. I would just want to simplify by saying we’re there to create conversations. Even your question says “move the needle,” and that’s really what the reality is when it comes to AI adoption. We are at the stage where we’re trying to move the needle. We know the AI technology as of today, its function and capability, has been far beyond how tax professionals are utilizing it as of today. I remember one of our polling questions in the audience was, How many of you are utilizing [AI] in your personal life versus how many of you are in your tax-related work? We need to create more conversations between tax professionals to help close the gap between technology readiness and the readiness for our tax function to embrace, the gap between personal use and professional use, the gap between technology advancement and the regulatory requirements. All these gaps, I think they start from the conversations between these masterminds in our audience like at the TEI Tax Technology Seminar. And then we need to question how to manage from both the top down on the strategic alignment standpoint and also bottom up. Many audiences that come to us are on the younger side of the profession, which is so exciting to see. And these conversations are those types of exciting topics that younger generations of CPAs are most excited about. So that’s another type of conversation that we are generating to retain those talents. Keep pushing the envelope for younger professionals to think outside the box.
Alarie: While the formal sessions at the Tax Tech Seminar are invaluable, the informal conversations between sessions are equally important. These interactions allow attendees to share experiences and insights, fostering deeper understanding and practical knowledge. Many attendees return year after year, building relationships and exchanging ideas that help them apply what they learn in real-world settings. These personal connections and ongoing dialogues significantly contribute to the event’s impact, making the Tax Tech Seminar a vital platform for advancing AI in tax.
Schneider: I think Ben was sitting on exactly what was in my head. Human beings by nature are social creatures, right? It’s about being together, bringing that information together, having that informal conversation. I don’t know if you’ve all experienced this as well from the AI stuff: the excitement is fantastic, but the noise is also through the roof. All the articles, all the sources of information out there, X, etc.—there’s just so much information out there, and everybody’s got their own different experiences. If you’ve got a good network and a good group of peers that you’re going back to like at TEI, you’re going to get information that you’re going to trust a little bit quicker to get you up to speed. It’s nice to know where everybody else is in their journey, because then you don’t feel like you’re alone, like you’re just sitting somewhere on your machine trying to learn and pick this up. You’re sound-boarding with other colleagues: “Actually, I went and did this training” or “I picked up this information really quick. It helped me. I’ll send you that article.” You need those networks to do that, because it helps you sift through the noise a lot more efficiently.
Hoffmeister: What do we expect the conversation to be surrounding AI and tax a year from now—or at next year’s Tax Tech Seminar?
Chen: Out of all of these questions, I feel this one is the most challenging one. Because of just the pace of the change that we’re seeing right now, especially on the AI front. In my heart, I’m thinking, When is the time when AI becomes a fully adapted and commonly utilized technology in tax? I think that day is very close to us. I would venture a guess at least for next year’s Tax Technology Seminar when we gather again with this audience hopefully, but also a wider audience in the tax technology world, I would say I won’t be surprised if there will be much more success stories, many more use cases, and people can share not just between the speakers, or between the vendors and service providers, but actually in our audience. Because there are several projects that I’m aware of [of] certain clients of mine that are taking off. And I know that these transformative initiatives in organizations, and then given the pace and dedication of tax leaders in these organizations, I won’t be surprised that they’ll be sharing lessons learned, the impact that they were driving through these AI implementations in their organizations. There’s not necessarily going to be all smooth sailing. Like any kind of technology implementation, I would expect next year people will have a bit more transparency around “This is where we are in our AI exploration journey for tax, this is where we tripped up, and this is how we picked it up and plan for a successful road map from here.” So, definitely more tangible use cases and demonstrations and capabilities will become available next year.
Schneider: Always a difficult question to predict the future, but I will share what my hope is. If you looked at the theme, things that we talked about in the prior questions, and probably over the last year or so, we’ve been in that awareness/desire phase. The ChatGPT stuff going on the scene, all the articles, all the content, all the information—even this past year’s seminar—creating a lot of awareness, creating a lot of tooling to tap into that desire. I’d love to say this time next year we’re talking about experiences where folks have had success, even if it’s small. Because that means we’ve moved the curve into the knowledge and ability section, and that could be through a combination of off-the-shelf solutions that have AI on it, like a handful of us here that represent those types of solutions, or you’re doing your own tooling, you’re leveraging your own citizen/developer–type tooling within your organization that’s available to start to build that knowledge and ability. Sometimes with the technology that’s moving as fast as this—and you could call it a not-so-fun topic, but it just kind of depends on your lens—what are the jurisdictions, the public sector, going to start to do around this? We’re probably going to see more information come out around regulation, policy, governance around AI that we’re going to have to increase our awareness around and figure out how to move forward and work within those considerations so we’re not tied up. There’s always that kind of lag from the initial excitement factor, because just naturally those types of bodies tend to move a little bit slower. But I think as that’s getting caught up, we’ll be discussing a lot more around that too, so then you can future-proof your business and start to think about ways to put in the proper guardrails to continue to leverage the technology.
Hawkins: In addition to what was said about the evolving technologies, increasing utilization, the compliance and ethical considerations pieces, I absolutely had top of mind as well a workforce or talent transformation. I do think we will see change in not just titles, but roles and what people are doing. I was really pleased at the panel discussion when we started to talk about innovation mindsets and changes that would benefit any tax department in embracing these technologies. I just think seeing how the workforce transforms next year is going to be super exciting and a great topic of conversation.
Alarie: Reflecting on the past year, the rapid technological advancements are astonishing. For example, within three days of OpenAI releasing GPT-4o, Blue J had integrated it, robustly tested it for performance, and launched it into our core production platform for our users, enhancing our capabilities significantly. This pace of improvement will continue, making current technologies seem outdated quickly. Next year, I expect we’ll see more concrete success stories and a different kind of hype, as organizations move from awareness to practical implementation. This shift will underscore the importance of staying updated and prepared for these rapid changes, validating the relevance of events like the Tax Tech Seminar.
Hoffmeister: Thank you all so much for your time. This was an excellent discussion.
