Enterprising Investor – Brian Pisaneschi, CFA: Demystifying AI for Smarter Investing

Podcast: Enterprising Investor (CFA Institute)
Date: August 18, 2025
Host: Mike Wahlberg
Guest: Brian Pisaneschi, CFA, Senior Investment Data Scientist at CFA Institute

Episode Overview

This episode explores the evolving role of artificial intelligence (AI), specifically large language models (LLMs), in investment management. Brian Pisaneschi shares insights from his research series on retrieval augmented generation (RAG) and other advanced AI workflows, aiming to demystify how AI operates and to highlight how investment professionals can leverage cutting-edge tools for smarter, more nuanced, and lower-risk decision making.

Key Discussion Points and Insights

1. What is Retrieval Augmented Generation (RAG)?

(01:00–03:06)

• Definition & Function:
  • RAG is a technique that connects LLMs to external databases, allowing them to return more current and accurate information than their “frozen” training cutoffs allow.
  • This approach reduces AI “hallucinations,” providing more grounded, real-time answers.
  • Example: When analyzing proxy statements for CEO compensation, RAG pulls relevant text chunks from the database and provides this to the LLM for context and analysis.

“A retrieval augmented generation system … is a way that we've, essentially us in the machine learning community have been able to provide more updated information or also more real time information.”
— Brian Pisaneschi (01:13)

2. Qualitative vs. Quantitative Analysis: Where AI Shines and Struggles

(03:06–06:49)

• LLMs excel at qualitative tasks:
  They’re built for interpreting and generating text, so they perform best when answering single, focused queries about qualitative data.

• Shortcomings in quantitative analysis:
  LLMs aren’t inherently suited for calculations. When extracting data from tables, their efficacy drops due to inconsistent structures and formats.

  • They do better with simple numerical queries than with complex, multi-source analyses.

“Large language models aren't meant to essentially, you know, perform calculations. That's never what they… were intended [for].”
— Brian Pisaneschi (03:24)

• Complex queries introduce more errors:
  When queries span multiple companies or datasets, accuracy drops unless tasks are broken down with agentic workflows.

3. Agentic Workflows and Hybrid Automation: Best of Both Worlds

(06:49–12:36)

• Hybrid approach:
  Combine LLMs (for language) with external tools (calculators, APIs, Python) for tasks like data analysis or financial metrics.
  • “Agentic workflows” break down complex tasks into steps, leveraging the tool best suited for each step. Automation platforms such as n8n and Zapier are highlighted as low-code/no-code solutions for building such workflows.

“You can break that up into steps and then at each step you can… have the large language model connected to APIs… a calculator, to a Python interpreter...”
— Brian Pisaneschi (07:12)

• Coding automation vs. human expertise:
  While AI is improving in code generation, human oversight and understanding of underlying technology remain crucial.

  • “Vibe coding”: Informally relying on LLMs for code snippets can lead to frustration if the user doesn’t understand the structure needed for automation.

• Process structuring matters:
  Even in no-code solutions, the ability to conceptualize automated workflows in a computer-scientist mindset is a competitive advantage.

4. The Enduring Role of Human Creativity & Curiosity

(12:36–16:30)

• Key human skills for the AI era:
  Creativity and curiosity are becoming even more valued, as they differentiate analysts from AI’s capabilities and biases.
  • The future edge comes from finding value “pockets” and leveraging domain expertise, not just from technical prowess.

“Human creativity and curiosity are going to differentiate us from AI's capabilities… we need to be giving teams a lot of space to exercise that part of our brain.”
— Brian Pisaneschi (12:52)

• Beware of model bias and consensus:
  Overreliance on LLMs can lead to groupthink based on training data. Injecting individualized prompts and guiding LLMs (via RAG and domain expertise) restores edge.

5. Prompt Engineering: An Overlooked Art

(16:30–19:07)

• Art more than science:
  Prompt engineering is about effective communication—good prompts are shaped by domain knowledge, creativity, and specificity.
  • Engineers can inject “secret sauce” (proprietary insights) into prompts to customize LLM outputs.

“Most of what I think to be really good prompt engineering stems from your domain expertise or your ability to communicate.”
— Brian Pisaneschi (18:45)

6. Common Misconceptions About AI in Investing

(19:07–21:25)

• Dismissal due to misunderstanding capabilities:
  Many investors try ChatGPT, don’t see results, and “brush it off,” missing the true value in custom workflows and connected data sources.
  • True potential comes from carefully defined, auditable, step-by-step workflows—not runaway autonomous agents.

7. Building Auditability and Nuance into Automation

(21:25–23:28)

• Audit trails and stepwise evaluation:
  Breaking down workflows with checkpoints helps prevent hallucinations and ensures output reliability.
  • AI can now add nuance and scale to previously simplistic processes—like improving binary filters in stock screening.

8. Career Reflection & Final Advice

(23:28–24:07)

• Brian’s first job: Corporate accountant at a property casualty insurance company.
• Advice to his younger self:

“Be as true to yourself as possible and take as much risk as you can and everything will work out.”
— Brian Pisaneschi (23:52)

Notable Quotes & Memorable Moments

• On hybrid AI-human automation:
  “That's what everything… VC right now is all trying to build… where if you have a highly complex workflow, you can break that up into steps…” (07:12)

• On prompt engineering as an art:
  “Prompt engineering is a lot more of an artist than it is a science.” (18:32)

• On creativity’s continued importance:
  “If you don't give the teams the ability to exercise that part of your brain, you know… we're really, you know, shooting ourselves in the foot.” (13:43)

• On the future of investing and AI:
  “Competitive advantage is found on the edge of that existing technology…” (09:21)

Timestamps for Important Segments

• RAG and AI basics: 01:00–03:06
• Qualitative vs. quantitative limitations: 03:06–06:49
• Agentic workflows, automation, and coding: 06:49–12:36
• Human creativity & future roles: 12:36–16:30
• Prompt engineering as a differentiator: 16:30–19:07
• Misconceptions about AI in finance: 19:07–21:25
• Auditability and nuanced automation: 21:25–23:28
• Brian’s career advice: 23:28–24:07

Closing Thoughts

Brian Pisaneschi’s central message is that while AI is transforming investment management—empowering investors to automate, surface new insights, and scale up their analysis—the real edge remains at the intersection of human creativity and domain mastery. Carefully structured workflows, hybrid AI-human systems, and thoughtful prompt engineering enable differentiated results. Automated tools are not a substitute for analytical imagination, but a force multiplier for those who know how to wield them.