Major accounting software companies view generative AI not as something that will enact a root and stem transformation in their products and services but as a powerful accelerant to enact their own longstanding ambitions. While hard at work building generative AI into the core of their offerings, the end result will be a more powerful suite of products that, at the same time, is more accessible and intuitive to the end user. The aim, then, is for AI to make them better, stronger, faster.
Vsu Subramanian, senior vice president of content engineering with tax compliance platform Avalara, said the company's AI ambitions are all about scale, from its recent rollout of a ChatGPT plugin for sales tax calculations to larger projects incorporating generative AI into its native software offerings. Subramanian conceded that the company's ultimate goal — namely for Avalara software to be part of every transaction on Earth — is certainly big and ambitious, but noted that tax compliance itself is a big and ambitious goal as well. Avalara already tackles complicated tax compliance issues, and the use of generative AI is seen as a way to increase its scope even more.
"AI for us means scale," said Subramanian. "We want to make sure what we build is done in a way that makes our products faster, easier to use, more accurate, more affordable and more reliable for customers. We do that today. How do I use AI for this? For us, it's primarily around scale. We can expand our coverage and that reflects as a better customer experience on everything we do."
In general, according to him, the company has been using AI for some time, even generative AI (ChatGPT is merely the most prominent example of generative AI, not the first) in ways the end user may not even realize. He said that, on the back end, there's a lot of classification, calculated, automated extraction and other data management processes that are currently guided by AI. The latest advancements in generative AI, then, are more about making these back-end processes even more efficient while making the end user experience even more intuitive.
"Generative AI is the next generation of natural language processing … but has some new tools," he said.
Intuit
Intuit similarly views AI as a way to achieve its lofty goal of being the financial platform of choice worldwide for any conceivable business activity — the role of AI in this goal isn't even that novel, as Nhung Ho, Intuit's vice president of AI, said they declared five years ago that they aimed to "be the AI-driven expert platform." She noted that Intuit products like QuickBooks and TurboTax first began their life as sets of floppy discs you bought at a brick and mortar shop. Now it's cloud-hosted online. The incorporation of generative AI, she said, represents the next step in an ongoing evolution.
In the immediate term, though, much of the focus is going to be on enhancing the user experience. A key part of this is the development of digital assistants who can not only walk people through business processes but also provide real-time insights, similar to how the company uses live experts today. The idea is to shift the products away from the user having to "do all the work" and toward having AI that works with the user side-by-side the whole time, showing them ways to be more efficient and effective.
"The assistant piece is really important," said Ho. "We see that as pairing AI with the power of human capabilities — really that is what will drive maximal benefit in the future. There is always this fear that SkyNet is coming for us to take over the world, but our view is AI is here to assist humans whether to do taxes themselves to assist an accountant preparing the taxes for that individual. It's about augmenting that expertise to deliver the best customer experience."
A key part of this is personalization and customization. While Ho said she is indeed an AI practitioner, she does not intend to build AI capacities for their own sake but instead to address specific business pain points. The use of AI in this respect is more about providing the right experience to the right user. It is not about making the AI but your AI.
"Having a unitary AI that knows all, does all, answers all — that is great in the future, but we really believe in AI for personalization, providing assistance and expertise regardless of where you are," said Ho.
Asked about a specific improvement, Ho said that when people contact experts for advice, generative AI is now capable of summarizing the call for the benefit for both the user and for future experts who might need to follow up on the issue. Summarizing previous conversations to gain a context for the issue was a highly time consuming process but now can be done automatically, meaning human experts can be more efficient and effective in delivering personalized, customized guidance.
Thomson Reuters
Thomson Reuters, similarly, can envision a future where it builds standalone AI products but, for now, is focused on ways that generative AI can be used to enhance its existing suite of products and services. In particular, according to Piritta van Rijn, global head of product, tax and accounting with Thomson Reuters, the company is working from the viewpoint of practitioners to develop ways they can use AI to add value to their services. This is not necessarily a new thing, though, and more an acceleration of what the company was already doing with its other automation efforts.
"By helping automate and streamline routine tasks for our professionals, we're going to enable them to focus on the more complex or strategic aspects of their work," said van Rijn. "This could look like new revenue streams or offerings for them. Thinking of advisory spaces really expanding, and getting rid of some of the routine tasks that burden folks in the accounting profession, leads to greater efficiency and greater job satisfaction."
The overall strategy is to use AI to provide transformative value that enables professionals to demonstrate more value to their customers and in their organizations through being empowered to provide better insights, advice and solutions for their clients. This includes using AI to augment their own expertise as well as find new efficiencies to leverage that expertise more effectively. This augmentation includes building more consistent and accurate workflows that can help an accounting practice do things faster. This implies a paradigm of humans working with AI versus AI taking over human tasks. Van Rijn said Thomson Reuters looks at these things from the engagement team level, and their thinking on AI comes from spending time with customers are their offices as well as hosting forums and panels to keep abreast of the common pain points for the professionals who use their products.
"I firmly believe that the best future solutions will be an enduring combination of AI systems and human expertise. We see it augmenting what can be done and how it gets done. I think we have to understand the value of generative AI and how this best combines with professional judgment, expertise and critical thinking brought out by the professional. We know tax and accounting is not black and white, there's a ton of judgment and expertise required and that will continue to be part of this profession," she said.
Wolters Kluwer
Brian Diffin, chief technology officer with Wolters Kluwer Tax and Accounting, said his company is taking the approach of looking at things from the CPA perspective and working outward from there. The company has been on a technological journey for many years, he noted, and the use of generative AI is just the latest step in that. The company began its use of generative AI by thinking about what CPAs actually produce and what clients look for in their CPAs and let that guide their development.
"A lot of work is going into recording and characterizing data in tax or audit," said Diffin. "We see that as some really low-hanging fruit to do that process with AI. We see AI looking at patterns of data to identify certain insights like anomaly detection … looking at a transaction volume that might be a lot higher than what you'd expect, or the transactions would only take place at 3 a.m. or a large number of those transactions happened in a short period of time. Producing reports, financial statements and audit reports. How do we automate that output?"
Part of this is about preserving the core of the accounting firm, which is client relationships. While AI can do a lot for accountants, it cannot and should not interfere with the human to human relationship between the professional and their client, according to Diffin.
"The relationship between the CPA and their client is discretely human," he added. It's a human-to-human relationship so it would be difficult to [completely replace]. Maybe you could someday in the far off future, maybe where digital humans are tough to differentiate from the real ones, but that is many years ahead."
Build, buy, partner? All of the above
As for where these AI models will come from for these companies, it will be a combination of building some of their own, and either buying or partnering with other companies for the rest.
Avalara's Subramanian said the plan is to build its own AI capacities for the most important parts of its offerings, particularly those related to compliance. Other areas are not as vital, and for those Avalara plans to either license or buy models from the outside.
"There are very common functionalities we don't need to build," said Subramanian. "Many of the cloud providers have certain functionalities we can use directly, even OpenAI's foundational models or the open source LLMs. We can use those. But we differentiate on things we are specifically good at, like compliance, and that is a case where we can build our own models and specific intellectual property that translates into better products for compliance customers."
Ho, from Intuit, similarly said there's no need to develop its own AI capacities for every single part of every single product. The company recognizes it can't really be all things to all people, so in areas where other solutions might be more appropriate for customers, Intuit plans to buy or partner to integrate them into its own products. At the same time, where it can be "best in class," Intuit has strong intentions to build its own models.
"We want to construct the best solutions so you can run your business," said Ho. "In terms of where we choose to build and where we see the advantages, we are always evaluating the technology to see what delivers the best experience."
Van Rijn, from Thomson Reuters, expressed a similar sentiment. On the one hand, it is taking what she called a "build by partnership" approach to integrate AI with its products, pointing to its recent acquisition of SurePrep, a tax automation company, as an example.
"At the same time, on the build side, we have 4,500 in-house experts with data science and AI expertise who have the tools and ability to build products to fit into our portfolio," she said.
Diffin said Wolters Kluwer, doesn't see itself necessarily making large language models in the same way as big tech companies like Google or Microsoft, but a big part of its overall strategy is developing its own algorithms, using a large group of data scientists in the company who regularly experiment with adaptive algorithms based on specific business domains.
"So we see us owning that IP. We would never outsource that part of the intelligence in our software," he said.
Maria Montenegro, chief strategy officer and chief innovation officer at Wolters Kluwer, added that while eventually she does envision the company one day becoming more of a pure technology company, it would still work differently in this area.
"We're in the business of developing customer solutions through applied technology use cases," she said. "We don't do R&D for the sake of R&D like some of the big tech giants. We marry our deep domain expertise with technology to create specific algorithms… which is how we compete in the market."
When it comes to developing their own models and algorithms, all four companies are — like the Big Four accounting firms (
"Over the years we have collected a vast trove of customer data. If you hear anything about AI, it is all about the quality of the data, not just the volume. High-quality data allows you to train the best in class AI," said Ho from Intuit.
Being on Team Human
All the executives acknowledged the anxieties people had about the rise of generative AI but, at the same time, all were confident that AI won't completely replace humans. And, like leaders at the Big Four firms, they weren't entirely sure whether that's what they'd want anyway.
"It starts with the focus of us being able to augment, so 100% automation is not the focus," said Thomson Reuters' van Rijn. "The importance of human expertise, judgment and relationships, we do not underestimate that. It will be a combination, for sure. We're focused on mundane, repetitive, low-value tasks. Those are the things everyone wants to see automated."
Generative AI still has a lot of challenges that will need to be overcome before widespread adoption, such as security and privacy. Montenegro, from Wolters Kluwer, raised a similar point and added to these challenges the comfort people have with AI. Even if they could realize all their ambitions technologically, it does no good if people won't accept them.
"In the drive to adopt AI we have found, and I'm sure many fellow players have found, the human element is still super important. There's still the transition stage where the human needs to feel comfortable with the results and eventually you can automate more and more but we will not be able to go sit on the beach while AI does our jobs for us," she said.
Subramanian, from Avalara, noted that even in a world where a process becomes fully automated — an end to end, "fire and forget" process where the human is responsible just for starting the process and evaluating the output — people remain involved. He noted that Avalara's software has already gone fully automated on certain things.
"That level has already been achieved. We do have 'fire and forget' in a few cases because we are confident. We have checked the steps and rechecked … last year, I believe, 5.2 million returns filed, and a good amount of them were automatically filed. Humans were not involved in that at all," he said.
At the same time, there were other returns that were more complicated and needed a human to validate them and, in some cases, correct the information. Humans will always need to be in the loop, he said, because there will always be a need to vet information and validate processes, even if it's just part of an annual evaluation.
"So we do have a lot of validation checks and balances to make sure what we are doing can still be trusted, but where we can we bring in automation to speed up things," said Subramanian. "Sometimes systems can break down, and if they do break and there are issues, we need to fix it and we look at that part of human involvement always."
Ho, from Intuit, raised a similar point, noting there never seems to be an end to human involvement. Whenever one thing is automated, another related thing requires humans. Once that gets automated too, another task emerges requiring humans. Previously, someone might have spent a lot of time categorizing transactions. Now AI can do this. That same person might then move on to analyzing the cash flow, P&L and other points. "That is where the human can come in and assist with that decision making because ultimately the AI is not going to do that. The AI is not making the decision, but decision-making is the most important thing of all from us. It's great to crunch your data, but what do you do about it?" she said.
Diffin, from Wolters Kluwer, said that while ideally everything would be automated, he just doesn't see a scenario where humans aren't involved in the process. AI automating things will just shift where the work takes place. "On the tax and accounting side, it's critical to be accurate … so humans will need to validate whatever the [AI] is projecting," he said. "One kind of example of a changing job paradigm in this space is like for an auditor, where the audit assurance process gets automated and even the production of the financial reports [is automated]. The auditor may start to transition to actually auditing the algorithm. Is it explainable? Is it accurate? Is it biased? How did it get the answer it got to? Are there emergent properties, and if there are, what does this represent? This needs to be validated."
