At an executive leadership meeting earlier this year, we devoted a block of time to discussing ChatGPT. Leaders from our AI team demonstrated a range of capabilities, including its ability to write marketing content, generate computer code, engage in customer support dialog, and as a grand finale, to answer a very complex accounting question. At the end, our young, high achieving global corporate controller gasped “I’m going to be replaced by AI!” To which, another of our leaders, known for his logic and dry wit, responded “no, you’re going to be replaced by someone who uses AI better than you do.” That’s the question of the day. Will ChatGPT and other forms of generative AI replace entire industries, like accounting, or will they, as we like to say at Sage “elevate human work?”
To be clear, the accounting industry is already using AI extensively and to great effect. AI that can read and categorize data from business documents is entering the mainstream. Accountants are using AI to automate financial modelling and forecasting. And as a corollary to all this automation, accountants are relying on AI to identify anomalous transactions – identifying accounting errors, revealing changes in business performance, and even catching fraud.
So, what makes generative AI different? First, a little explainer. Generative AI refers to a category of artificial intelligence that can create new content or outcomes, often based on patterns or structures it has learned from existing data. This includes generating text, images, music, and even complex designs. GPT, the foundation for ChatGPT, is a form of generative AI called a Large Language Model (LLM). LLMs are machine learning models designed to understand, interpret, and generate human-like text. These models learn patterns and structures in language from massive volumes of text data, enabling them to complete prompts, answer queries, or even create entire articles in a remarkably human-like manner.
Many of you have already had your magic moment with ChatGPT or one of the other new chatbots where you realized AI has leapt forward. Mine was a very simple interaction, but the implications hit me like a bolt of lightning. I told ChatGPT a dad joke: A skeleton walks into a bar and says to the bartender, “I’ll have a beer and a mop.” Many humans don’t get this joke right away, but ChatGPT quickly responded, “Haha, that’s a good one! The joke plays on the fact that a skeleton, being made of bones with no organs or tissues, would not be able to consume the beer like a living person. Instead, the beer would simply flow through its bones and make a mess on the floor, hence the need for a mop.” After picking my jaw up from off the floor, I surmised OpenAI, the makers of ChatGPT, had solved two big AI limitations – an understanding of the world and the ability to reason. Whether these abilities are real or imagined is irrelevant – AI is now able to perform complex tasks with minimal instruction.
Research conducted by OpenAI in partnership with the University of Pennsylvania reveals around 80% of the workforce will have at least 10% of their work tasks impacted and the remaining 20% will have at least 50% impacted. Not surprisingly, accountants, auditors, and tax preparers are among the jobs expected to be most impacted. The research generated a lot of headlines. My favorite was a bit cheeky – “Stonemasons Should be Fine in the Brave New World”. Apparently, journalists, who are also among the jobs expected to be most impacted, use sarcasm as a natural defense mechanism.
What does all this mean for the accounting industry? Let’s start with a massive disclaimer: generative AI is moving fast and will be profoundly disruptive. I believe generative AI is as transformational as the Internet or the smart phone but is racing through the adoption curve faster than any technology we’ve ever seen. ChatGPT reached a million users in five days. It reached 100M users in two months. In the span of six months, its capability, as measured by performance on the Bar exam, improved from performing in the bottom 10% to performing in the top 10%. It is impossible to predict how this will play out, but I can give you a few likely scenarios.
AI will progress from automating tasks to automating entire workflows. I mentioned how we’re able to use AI today to read data from an invoice, categorize the expenses, and identify problems like over-billing. Today, that AI is deployed within strictly defined workflows where pre-configured rules orchestrate activity. With LLMs, we’ll be able to hand the orchestration off to “AI Agents” that can operate autonomously. These agents could own and operate a workflow with the same type of instructions you might give to an accountant: “don’t let the cash balance fall below X; make sure to pay vendors of type Y early; let me know if you see any bills with unusually large amounts.”
Digital Assistants will change the way we interact with technology. Today’s business applications guide user interaction through pre-defined menus, carefully designed data entry forms, lists for sorting and filtering records, and “canned” reports or dashboards for data analysis. These UIs are both limiting and hard to learn. LLMs promise to replace all this with simple human language interaction. Imagine working within a spreadsheet and asking a digital assistant to gather the accounting data you need to analyze. Imagine reviewing and approving purchase orders from within Teams. Tools like Excel and Teams are familiar and suited to the job to be done, enabling more accessible “serf-serve” capabilities that reduce dependency on the accounting team. Further, imagine LLMs integrated with enterprise systems enabling employees within the business to get their own answers or do their own data analysis on enterprise data via a simple chat interface. In fact, as chat interfaces become the norm, employees won’t settle for today’s crude tools.
AI will handle routine interactions with customers and vendors. Most emails an accounting team receive fall into a small set of repetitive interactions. For instance, customers ask for tax information, ask for their latest statement, or remit payment information; vendors send invoices, ask when to expect payment, and provide compliance details. LLMs are exceptionally good at understanding the intent of these emails, determining the steps required to respond (i.e., send an invoice for processing or get an invoice from the accounting system for a customer), and generating a response email if called for.
AI will be an expert resource. Accountants will have access to an expert resource for understanding and applying accounting standards. What’s the correct accounting treatment for a complex sales contract? What does tax law say about capital asset write-offs? LLMs are also helpful data analysts. Give an LLM access to balance sheet data for the last 24 months and ask it to identify the most significant trends. Give it access to detailed ledger data and ask it to forecast future cash flow. More generally, AI will make every knowledge worker more productive. A few paragraphs ago, I was really struggling to write a concise, understandable definition for generative AI and LLMs, so I asked ChatGPT for help. Its suggestion wasn’t perfect, but it gave me something to work with and unblocked the writing process.
Of course, LLMs come with well-known risks. They’re known to hallucinate, where their responses are articulate, well-reasoned, and entirely make-believe. They can be tricked into misbehaving. There are questions about data privacy. They’ve introduced a long list of legal questions over both the content used to train the models as well as the ownership of content the models create. The technology is outpacing regulation. The situation could call for a moratorium on their use, but the genie is already out of the bottle. We need the accounting industry to guide businesses on how to leverage AI safely and effectively.
I went to New York City earlier this year to visit my friends at the AICPA. I wanted to look into the whites of their eyes for signs of fear. I saw the opposite – where most see a threat, they see opportunity. We aligned together on two overarching opportunities with generative AI: 1) solving the industry’s labor shortage, and 2) enabling accountants to deliver more strategic value. As we further discussed the risks and challenges posed by generative AI, we discovered a third opportunity I believe to be greater than the first two. As we delegate more and more tasks to technology, especially to technology that can be unpredictable, the value of human accountability will increase.
Solving the industry’s labor shortage. It’s no secret the accounting industry faces a labor shortage. Bloomberg Tax calculates that the number of accountants and auditors employed fell by 17% between 2019 and 2021. University enrollment in accounting courses fell by four percent between 2016 and 2019. According to Deloitte & Touche LLP, 82.4% of hiring managers for accounting and financial positions in public companies believe that recruitment is a “big challenge.” We can argue over how to attract more people to accounting careers, but a better answer may be to simply make accountants more productive. AI is particularly well suited to automating repetitive tasks like data entry and reconciliation. Sage customers using our AI powered invoice processing are reporting a 2X – 3X improvement in productivity, creating more capacity within their accounting teams. And this automation has a happy side-effect – automating low value, repetitive work may make an accounting career more attractive.
Enabling accountants to deliver more strategic value. In the list of “jobs to be done” by accountants, there’s no negotiating over compliance related work. The accounting team can’t postpone an audit to make time to evaluate a potential acquisition. There’s no give on tax filing deadlines. Any time and resources devoted to strategic work can only follow the required work. Automating these non-negotiables naturally frees time for more strategically valuable work. Equip accountants with LLMs, and not only will they have more time for this work, but they’ll also have an invaluable, hyper efficient research assistant that can analyze massive amounts of information and structure coherent, professional presentations of the results.
The value of human accountability. We have a maxim at Sage – we should match our investments in automation with equivalent investments in trust. Human users will only transition their work to technology if they trust the technology to do the job safely and competently. Not to be too dramatic here, but while there’s an obvious benefit to automating compliance work, the risks of non-compliance are ruined careers, or worse. We can transition the work to technology, but accountability will always reside with humans. Humans sign audit statements – not AI.
The accounting industry’s greatest contribution to society is confidence in our markets. There are various ways to put a price on that confidence. For example, the total revenue in audit fees for the global big four accounting firms in 2021 was $271B. I believe AI will drive an exponential increase in this value, in part because AI increases risk, but more importantly because a more efficient and productive accounting industry will lead to more efficient and productive markets.
Getting back to our question of the day, I’m quite optimistic. We have the opportunity with generative AI to not only elevate human work, but also to elevate the entire accounting industry. Yes, we should proceed with caution, but we should also proceed with pace and determination. The winners will be the ones who use AI better than anyone else.