2025 年 2025 巻 FIN-034 号 p. 75-78
Fermi estimation is a widely used technique in business analytics and market research for deriving approximate answers to complex problems where concrete data is insufficient. Business analysts employ this method to estimate market sizes and sales figures, which are critical inputs for drafting budgets and sales plans. This process requires advanced domain knowledge, logical reasoning, and precise formulation of estimation equations. In this research, we explore the integration of large language models (LLMs) to enhance the accuracy of Fermi estimation problems by leveraging causal structural reasoning. By constructing an estimation plan rooted in causal structure estimation, LLMs are used to systematically identify and utilize the relationships underlying the target event. Given the inherent challenge of obtaining precise ground truth data for such tasks, we propose an evaluation framework that leverages the ensemble consensus of multiple LLMs to assess the quality of Fermi estimations. Our findings demonstrate the potential of LLMs to improve the rigor and reliability of economic Fermi problem-solving, advancing their practical applications in business analytics.
