Technology and Innovation Community

Good morning everyone,

As part of recent work at Aston University on financial‑analysis AI agents, I’ve been experimenting with Google’s Gemini to build a fully functional analysis agent capable of processing annual reports end‑to‑end.

I wanted to share with the community a short summary of the workflow I used while developing it. It also covers prompt‑engineering structure, ratio‑analysis reliability, and how to critically evaluate LLM‑generated financial insights using both academic and professional literature: A practical guide to building your first financial AI Agent with Gemini | LinkedIn

If you’ve been experimenting with agentic AI in a financial setting yourself, I’d be very interested to hear about your approach, especially concerning how you handled ratio accuracy and output interpretation.

Thank you in advance,

Best regards,

-------------------------------------------


------------------------------
Guillaume (William) RUQUILLA
MSc Fintech - MiM / Finance
Audencia BS x Aston University
------------------------------

Excellent article Guillaume. 

From my experience, I do believe a multi-agent approach will work best to separate the business strategy (PESTEL, SWOT) from the financial analysis part. I also like to use a metaprompting approach combines with a mixture of experts command that research has proved to minimize hallucination and improve scores in many cognitive tasks. In this way, two years ago I was involved in a project to conduct PESTEL and SWOT analysis and the prompt was lacking this mix of experts approach. Last year, I upgraded the model to factor the aforementioned approach and the results improved significantly. 

You can see below an example for PESTEL:

"Act as a strategic advisory board to conduct a comprehensive PESTEL analysis for [INSERT COMPANY/INDUSTRY].

Assign one specialist to each category:

• Political: A Policy Analyst/Lobbyist.
• Economic: A Macroeconomist.
• Social: A Sociologist/Trend Forecaster.
• Technological: A CTO/Innovation Specialist.
• Environmental: An ESG (Environmental, Social, and Governance) Consultant.
• Legal: A Corporate General Counsel.

The Process:

1. Specialist Reports: Each expert identifies the top 2-3 high-impact factors within their domain.
2. Cross-Impact Analysis: The experts must discuss how their domains overlap (e.g., how Environmental regulations are driving Technological shifts).
3. Risk & Opportunity Matrix: Based on the discussion, provide a final table of the most critical threats and strategic opportunities for the organization."

---

 

------------------------------
Carlos Salas
Portfolio Manager & Freelance Investment Research Consultant
------------------------------