AI
Unlocking the Power of AI in Accounting: Opportunities and Challenges
AI. Two letters that, when combined, elicit a weird combination of panic and hope.
For years, we’ve been talking about the talent crunch facing the accounting profession. Accountants leaving in droves, fewer majors at universities, and growing expectations have created a vicious cycle in which accountants are busier — and more stressed — than ever before.
The good news is that business leaders are starting to understand this. The only question is: Will it be enough?
The rapid evolution of artificial intelligence (AI) has sparked transformative changes across various industries, and using AI in accounting is no exception. As businesses look for ways to streamline operations and enhance decision-making, AI’s potential to revolutionize accounting practices is undeniable. However, the journey is not without its challenges, especially when it comes to leveraging generalized AI models for highly specialized tasks.
Recently, I sat down with my co-founders, Cullen and Chris, for a special episode of Blood, Sweat & Balance Sheets — available for free on FloQademy — to break down:
- What AI really is, specifically focusing on LLMs and neural nets
- The current limitations and strengths of AI
- How the perspective of accounting departments has shifted from viewing AI as a threat to seeing it as a significant opportunity
- Ways AI can alleviate the workload of overworked accountants by automating routine tasks and providing valuable insights
Here’s a teaser:
The Generalization Challenge
Most people associate AI with large language models (LLMs). Generative pre-trained transformers (GPTs) are a specific type of LLM.
Cullen sheds light on why GPTs — as they stand now — aren’t equipped to handle the nuanced, highly analytical nature of much of the work accountants tackle.
“GPT is a generalized model, which means that its job is to try to give you a good enough answer about almost anything, but you kind of lose something there,” Cullen said.
While these models excel at providing broad answers across diverse topics, their ability to handle nuanced, company-specific accounting queries is limited.
“As an example, an LLM can provide answers for general accounting questions — ones that might appear on an exam — but struggles with specific queries about unique company practices. GPT is designed as a generalized solution, so out-of-the-box LLM models can’t easily give precise, job-specific information.”
The Path Forward: Tailoring AI to Accounting
Despite these challenges, there are promising techniques to harness the power of AI for accounting. One such approach is Retrieval-Augmented Generation (RAG), which involves preloading the AI with company-specific information.
“As an example, there’s something called RAG, which is basically when you pre-seed the LLM with a bunch of information that’s really specific to you and your company…it can do that pretty darn well,” says Zandstra.
Thus, by narrowing the scope and providing relevant context — and a lot of it — AI models can generate more accurate and useful responses tailored to specific needs.
However, Cullen cautions that even with techniques like RAG, achieving nuanced answers remains a complex task.
“Even with something like RAG, it’s not easy. It’s actually rather difficult to get an LLM to give you really nuanced answers…it’s trained to give you generalized answers,” he said.
Embracing the Future of AI in Accounting
What we’re not factoring in here — though we do touch on it in the full episode — is what’s next for AI, and what’s next for AI and accounting.
As AI continues to advance, the accounting industry must balance leveraging its potential while still taking into account its limitations. Customizing AI models to better fit specific accounting needs and continuously refining techniques like RAG will be crucial steps forward.
Check out the full episode only on FloQademy. This episode explores the latest trends, challenges, and success stories of AI in the accounting world. Visit learn.floqast.com for free and earn CPE credits.
