What is it about?
Financial reports are essential for investors and companies to understand business performance, but creating them manually is time-consuming and complex. Our research explores how AI, specifically large language models, can automatically generate these reports from companies’ earnings announcements. We compare two main approaches. The first uses an agentic style, where different AI agents collaborate to complete the full report. The second uses a “step-by-step” style, where the report is broken into smaller questions that the AI answers one by one before combining the results. By testing these methods, we find that the step-by-step approach produces more detailed and accurate reports, while the teamwork method is better at producing concise summaries. This comparison offers practical insights into how AI can be used to create structured financial reports more efficiently, saving analysts time and improving decision-making in finance and beyond.
Featured Image
Photo by Markus Spiske on Unsplash
Why is it important?
Financial reports are a cornerstone of business decision-making, guiding investors, analysts, and companies in evaluating performance and planning for the future. Yet, creating these reports manually is slow, costly, and prone to human bias. Our work is timely because LLMs are rapidly transforming how information is processed and summarized, but their use in producing structured, template-based reports remains underexplored. We are the first to directly compare two leading AI approaches: a multi-agent framework and a decomposed “step-by-step” prompting method. This comparison not only highlights the strengths and weaknesses of each approach but also provides practical guidance for real-world adoption in finance and potentially other domains such as climate reporting or risk assessment. By showing that decomposed prompting yields richer details while multi-agent systems produce concise overviews, our study equips researchers and practitioners with insights to design more reliable, efficient, and adaptable AI-powered reporting systems.
Perspectives
Working on this article was a rewarding experience because it brought together different ideas about how artificial intelligence can best support real-world financial analysis. I personally found it exciting to see how two very different approaches—multi-agent teamwork and step-by-step decomposition—can each reveal unique strengths when applied to the same task. For me, the most valuable part of this project was not only the technical results, but also the process of translating these abstract AI methods into tools that could genuinely help financial analysts. It made me think more deeply about how AI can shift from being a research concept to something that people in industry can rely on every day. I hope this work inspires others to explore practical and creative uses of large language models, not only in finance but also in other fields where structured, trustworthy reporting is essential.
Yong-En Tian
Read the Original
This page is a summary of: Template-Based Financial Report Generation in Agentic and Decomposed Information Retrieval, July 2025, ACM (Association for Computing Machinery),
DOI: 10.1145/3726302.3730253.
You can read the full text:
Contributors
The following have contributed to this page
