As technological innovation accelerates, Artificial Intelligence (AI) has seamlessly integrated into daily routines, transforming how we work, communicate, and solve complex problems. The push to incorporate AI tools is strong, driven by fears of lagging behind in a rapidly advancing digital landscape, whether through platforms like ChatGPT, DeepSeek, or other emerging technologies.
Within the accounting and finance sectors, there is a continuous effort to enhance efficiency, accuracy, and productivity. Many professionals are exploring ways to embed AI into financial processes, particularly for automating sophisticated model development within spreadsheets. Such developments are compelling enough that I am currently writing a book focused on these innovations and their implications.
Despite high expectations, current AI capabilities in financial modeling remain somewhat limited. Tools like OpenAI’s ChatGPT and Copilot utilize large language models (LLMs) trained to produce human-like responses by recognizing patterns in extensive datasets. These models generate outputs based on statistical likelihood rather than a true understanding of financial concepts, which can be a challenge when dealing with the nuanced nature of financial data. Their success heavily depends on clear, well-structured inputs, often difficult to achieve with complex language and intricate financial scenarios.
When deploying AI for critical financial calculations, careful oversight is essential. A single miscalculation can have serious repercussions, making manual validation and review indispensable. Over time, this will likely lead to increased emphasis on rigorous model auditing within financial workflows.
Assessing AI-generated financial models also presents specific difficulties. Frequently, these systems produce static results rather than flexible, formula-based models, limiting the ability to perform dynamic “what-if” analyses vital in financial planning. Attempts to generate actual formulas through AI often result in inaccuracies, undermining confidence in both the outputs and the overall models.
While LLMs aim to deliver dependable initial answers, they require iterative tuning, referred to as “prompt engineering”, to enhance output quality. Despite ongoing improvements, tools like ChatGPT and Copilot still encounter recurring issues such as illogical steps, inconsistent calculations, and overly generic handling of nuanced scenarios.
Particularly challenging are specialized calculations common in financial modeling, like determining terminal values in discounted cash flow assessments or calculating final loan installments. These AI tools tend to apply universal methods across different contexts, which can be both inefficient and error-prone. For critical tasks, the manual creation of these models often yields more precise and reliable results.
Another limitation is AI’s difficulty maintaining context over multiple dialogue turns. ChatGPT may lose track of earlier exchanges, and Copilot’s responses can be narrowly focused, restricting comprehensive analysis. Additionally, AI models lack true domain expertise; they generate responses based on training data rather than specialized knowledge in taxes, accounting, or valuations.
As AI technology advances, the roles of accountants and financial professionals remain crucial. Human expertise continues to be essential for refining, validating, and ensuring the accuracy of models. Currently, AI serves best as a powerful aid rather than a replacement, supporting professionals in achieving more efficient and precise financial analysis.
About the Author:
