How Generative AI will alter the role of the accountant

Arecent executive briefing from GlobalData noted that OpenAI’s ChatGPT only took two months to reach 100 million users in contrast to nine months for TikTok and five months for Twitter.  

Despite the obvious hype around it, the adoption rate shows that there is the potential for swift integration and change in accounting. Indeed, KPMG has referred to generative AI as ‘the fastest adopted business technology in history’. 

Speaking to International Accounting Bulletin, Robert McGillen, Chief Innovation Officer of Financial Services at CBIZ, comments, “The swift adoption rate in the accounting sector indicates an industry-wide recognition of AI's transformative potential. Firms that fail to adapt may risk obsolescence.” 

Large Language Models (LLMs) like ChatGPT, which can be interacted with using natural language, can be trained on large swathes of data, so have the potential in accounting to digest and summarise regulation and compliance documents and more. More and more firms are developing internal LLMs, understanding what to train the models on and how to educate professionals to interact with and get the most out of the technology. The generative tools dominating the market are the likes of ChatGPT and Microsoft Azure.   

Various analogies have circulated on how generative AI and LLMs should be viewed. Soon-Ee Cheah, GM of AI Products at Xero believes that, “AI has the power to augment human intelligence, not replace it.” 

James Osborn, KPMG’s Chief Digital Officer, largely agrees with that, “Imagine you've got a very smart, very capable trainee working with you on that engagement,” he explains. “And you wouldn't automatically take anything that trainee does and say okay, that's excellent, let's put that straight into engagements. But that doesn't mean they can't do fantastic things that then help you sort of, accelerate the outcome.” 

Echoing these sentiments too, Vsu Subramanian, SVP of Content and Engineering, from Avalara says, “I prefer to think of generative AI as a really smart, really good assistant. It will make you more productive and scale your capabilities, but you still need to have humans involved to review the work before it is finalised.”

McGillen believes AI has the potential to change the role of the accountant. “Utilised effectively, the role of an accountant with AI in hand transforms from a data processing expert to a true business advisor,” he argues.

Following on from this, though, Sarah-Jayne Martin Director, ICA Global AR Practice at Quadient, notes that, while there has been a huge buzz around generative AI in the last year, it “won’t necessarily replace traditional AI”, with more traditional AI still needed for tasks like data analytics.

The main use cases of AI at present in the accounting professions are its summarisation and drafting capabilities. Subramanian has highlighted its ability to draft compliance documents and identify regulatory risks by sifting through large databases of compliance documentation.  

Xero’s Cheah tells IAB, meanwhile: “AI has of course changed traditional accounting, in the way that it frees up time, helps with analysis and predictions; and enables accountants to spend more time on advisory services.” 

Xero has been focusing on how Generative AI will affect small businesses and advisors. Its most recent guide Future Focus: Xero’s AI guide for accountants and bookkeepers helps small businesses to ”navigate the impact of AI on their practices and clients”.  

Elsewhere, Grant Thornton has developed a new chatbot, GTAssist, launched  in October 2023.  

Explaining why the firm has dedicated time and resources on the project, Amy Smart, Digital Transformation Partner at Grant Thornton UK said, “Creating an in-house chatbot tool allows us to replicate the capabilities of public tools in a private way, giving us control over client and commercially sensitive information. Creating our own solution also allows us to add more capability than would be possible in a bought solution, for example better visibility of the ‘prompts’ being passed into the tools and feedback loops to help us address poor or incorrect responses. The GTAssist chatbot also makes it easier than ever for our people to instantly access corporate information, such as internal firm processes and policies, which are stored in multiple places.”

With many companies developing their own LLMs, Osborn believes there should be an emphasis on education and understanding of the technology. KPMG initially blocked the use of ChatGPT and then reached the conclusion that they needed to develop their own LLM internally that can access KPMG’s own data repositories. KPMG’s internal LLM was soft-launched in Q1 of 2023 and has been accessible enterprise-wide for about six months. Osborn adds that they are, “Still experimenting with the right boundary between internal proprietary knowledge and the external”, and over time they will allow the LLM to consume publicly available accounting standards.

A key challenge with generative AI is that it can fall victim to faults termed ‘hallucinations’, when the technology generates outputs that are inaccurate or that don’t make sense. This could be problematic in accounting, such as with regards to regulation and compliance. Osborn notes, “It’s not just giving access to the right data, it’s then ensuring we’ve got the right guardrails around and the right experts there as well.”

Another difficulty to navigate, he says, is how to impart to employees the fine line between caution and acceptance of the technology: “It's a tricky balance … if you over-emphasise the skepticism, then people won't use it.”

To address this, over the summer of 2023, KPMG had an internal education programme called the Summer of AI, which drew together 6,000 participants from 14 countries to learn more about the technology through over 50 in-person and online training sessions. Following this, an internal survey stated that 90% of participants said that they feel more confident in their understanding and ability to talk about AI.

Subramanian echoes the sentiment regarding the challenges of generative AI integration: “There are techniques to control and limit that risk. For example, designing better prompts can guide the model to think step-by-step, use guardrails, and provide more accurate answers. Even with these controls in place, generative AI should not be considered perfect.” 

He also speaks to the significance of good data repositories, emphasising that Avalara have a very large collection of compliance content, which can be used to train LLMs.

The importance of good data repositories, meanwhile, has been a core determinant of success within finance and technology, and it’s no different with AI. Osborn states that it is, “key that you've got good proprietary data stores as well, that you've got your own carefully curated data stores as well that, you've got your own, carefully curated data which add something on top of the more generally available models. I think that's really important.” 

There is clear interest and investment in the technology from the Big Four, with EY announcing at $1.4bn investment package in AI and blockchain technology, while Deloitte appears to be predominantly focussing on providing information surrounding AI use cases across a variety of industries and issues surrounding its legality. Deloitte’s AI institute has put out a Generative AI Dossier and has been reporting on the use of the technology in jobs. Its Generative AI practice is focused on the responsible growth of AI, which is an increasingly discussed topic by businesses and countries alike, with the UK government having hosted the AI Safety Summit at the start of November 2023.  

PwC is investing heavily in AI, with the US arm making a $1bn investment over the next three years to expand and scale up AI capabilities. PwC is working with both ChatGPT and Microsoft’s Azure OpenAI Service to support their clients. In 2023, at the International Accounting Forum and Awards, PwC was awarded Digital Innovation of the Year for the company's predictive analytics, AI-assisted financial statement disclosure checking and AI-enabled search functionality. This AI-supported function will be able to generate summaries to assist with the auditing process and find relevant information for auditors. This facilitates PwC auditors to handle large volumes of data more easily. 

KPMG has also announced a $2bn investment in AI, as well as a partnership with Microsoft for the next five years, showing the company's dedication to integrating and understanding the technology.  

McGillen says, “2024 will be a breakout year and likely to include several transformative players in audit, tax, and advisory solutions.” He also believes AI has the potential to affect the talent shortage in accounting, adding: “Many theorise that AI will solve (to some lesser degree) the talent shortage. My perspective is it changes the perception of the profession – and makes it more interesting as a career choice.”

For Osborn, the emerging technology will be pervasive. “We're increasingly of the view that LLMs will be everywhere and embedded in pretty much every piece of software that humans interact with in some way now, so there'll be lots of it,” he says. “But the real winners, I think, in the longer term are the ones who can enable their people to use the models and augment the work they do with them. I think that that's really key.”

Looking to the future, Osborn states that it is not a case of letting a model learn the accounting standards in one year and being done with it, he foresees the need for a permanent group of experts to continue interacting with the respective LLM models, helping to shape them into something constantly valuable for companies. He explains that LLMS need “that expert to interact with, almost like a librarian or a custodian and curator to make sure it's continually useful.”