Artificial Intelligence in Accounting and Finance: Benefits and Pitfalls

July 10, 2025

Use of artificial intelligence (“AI”) across business functions, including accounting and finance, is rapidly becoming a reality for businesses of all sizes. When deploying AI, it is critical to understand how the artificial intelligence model is operating, and where it may falter, to truly derive value for your business. 

Common AI Models

Some of the most well-known artificial intelligence models are known as LLMs, or Large Language Models. This type of AI, which includes OpenAI’s ChatGPT, Microsoft’s Copilot, and Google’s Gemini, are designed to understand and generate human-like text. They essentially function as a highly complex mathematical model to predict and produce language in a coherent way. The models are trained on huge amounts of text data, including books, websites, and news articles, to identify statistical patterns in language. As a result, the reply to any query in one of these models simply represents language that would statistically be most likely to appear as a response.

Other AI models used by businesses today include machine learning models, used for tabular prediction tasks, and computer vision models, used for quality control, security monitoring, and facial recognition, among others. Although there are a large variety of AI models available for business use, this article will focus on LLMs, given their low cost, widespread use, and flexibility in use cases when compared to other AI models, which tend to be specialized in nature.

Benefits and Use Cases for AI

LLM artificial intelligence models have exhibited beneficial use cases for generating text which is adequately supported by the model’s training data. Beneficial use cases may include:

• Nontechnical Writing: Using AI to write initial drafts of content that does not require highly technical or specialized information can drastically reduce the amount of time spent on these tasks. AI can often provide quality initial drafts of emails, memos, website content, product descriptions, and advertisements. However, AI may falter with highly technical material, such as accounting guidance, legal citations, and scientific analysis, as it lacks true subject-matter comprehension.

• Computer Programming: Thanks to the large amount of existing code available for training, along with the logical nature of computer programming languages, LLMs are highly effective at assisting in writing code for small- and mid-sized tasks. Accounting and finance functions may be able to realize significant value by using LLMs to write simple code, such as macros and basic Python functions, to automate tasks. Likewise, outside of the finance function, businesses may find success in using AI to assist in HTML coding for website updates.

• Summarization of Content: When provided with specific content, such as a news article, email, or notes from a meeting, AI can be used to generate quick summaries of the content.

• Proofreading: AI can be helpful in proofreading and providing feedback on existing content.

Overall, LLMs are most effectively used when they are supporting your existing work, making you and your employees more efficient.

Pitfalls of AI

While AI can be a powerful tool to increase the efficiency of a human, it is not a replacement for human judgment. The effective use of AI in accounting and finance requires the understanding of its limitations, which include:

• Accuracy: As described above, responses provided by an LLM simply represent language that is statistically likely to appear as a response to the user’s query. As a result, LLMs often “hallucinate,” or confidently generate text that is factually incorrect, made up, or misleading, while sounding plausible. Two lawyers previously made headlines for submitting a filing to a Manhattan federal court that cited several fake cases in support of their arguments, caused by hallucinations in the model’s responses.

• Incorrect Conclusions: Even if the model is not directly hallucinating, AI can frequently fail to understand the nuance required when performing accounting research. Accounting and tax rules often contain subtleties that require human judgment to determine the appropriate treatment. AI can fail in interpreting these details by providing plausible solutions from incorrect sections of the related regulations.

• Mathematical Abilities: LLMs are notorious for their inability to interpret numbers. While results have seemingly improved since their initial releases, they frequently fail at even the most basic math problems. Even when prompting an LLM to write a paragraph on any topic with exactly 30 words, the model will most likely not produce a result meeting the criteria. In practical applications, we recommend against using LLMs for math problems, table generation, or any type of modeling, such as cash flows, due to the frequent errors identified in the results.

• Prioritization of Information: While AI can be used as an effective tool for performing initial research and summarizing existing information, users should understand the AI’s lack of human judgment in its responses. AI responses can sometimes prioritize the wrong information in a summary due to its lack of understanding in the content, providing a bigger focus on the less important details. The output of any AI response should always be reviewed by a human for accuracy and appropriateness. 

• Data Security: Caution must be exercised when inputting data into LLMs. Differing tiers of AI models can provide different levels of privacy and data security. The free versions of the major LLMs available generally offer no expectation of privacy; any queries or information provided to the model will likely be stored and used to further train the model, which can provide significant limitations on its uses for businesses. There are differing paid tiers of certain models available, including Copilot and ChatGPT, that are designed for enterprise use and claim to not retain or share your data for training purposes. However, even with these guarantees, many businesses still maintain policies to not share confidential, highly sensitive, or proprietary information with the models.

AI in Your Business

AI can be an effective tool for businesses of any size to improve productivity and quality, but improper use can pose serious business risks. Businesses will be most successful in deploying AI when they have an understanding of how the models work, what they do well, and where they fall short. Stradiot Consulting & Advisory can consult with your business to identify process efficiency opportunities and AI use cases to improve your business’s cost structure.

This article is for informational purposes only and should not be relied upon as financial, investment, or business advice. If you would like personalized guidance and support, please contact us.