Business Breakdowns: How Investors Are Using AI

Podcast: Business Breakdowns (EP.240)
Host: Matt Reustle
Guest: David Plon, Founder of Portrait Analytics
Date: February 5, 2026

Episode Overview

This episode explores the evolving ways investors use AI tools to enhance investment research, decision-making, and workflow efficiency. Host Matt Reustle interviews David Plon, a former hedge fund analyst and founder of Portrait Analytics, about the practical applications of AI in investment processes. The discussion emphasizes real-world use cases, skills required to maximize value, the challenges and opportunities of implementing AI at both individual and organizational levels, and the feature frontiers such as memory and agentic AI.

Key Discussion Points & Insights

1. Guest Background & The Investor’s AI Perspective

• David Plon’s Journey:

  • Started as an investor at Barclays (special situations), then at Slatepath Capital (hedge fund generalist), most recently at Baupost in public markets (public equities and distressed credit).
  • First saw the inevitability of AI in investment research during Stanford business school (2015/2017), pre-transformers era.
  • Founded Portrait Analytics to address workflow bottlenecks he experienced as an investor.
    [04:52–05:54]

• The Investment Research Workflow—Pain Points Solved by AI:

  • High friction and time-intensive aspects can be enhanced by AI without undermining conviction-building:
    • Idea Generation: Hard to consistently find ideas that fit mental models.
    • Context Building: Getting up to speed on new names/industries was arduous, especially for generalists.
    • Position Monitoring: Challenging to track companies, sectors, competitors, and industry dynamics efficiently.
    • Historic difficulty in triaging relevant data points from vast, unfiltered information pools.
      [06:23–08:44]

2. Tangible Use Cases of AI in Investing

A. Position Monitoring

• AI Advantages:
  • Move beyond basic watchlists for specific companies; AI now enables a broader, more contextual net.
  • Example: For Expedia, tracking not just company-specific news but integrating relevant data from hotels, OTAs, demand trends, etc. was once almost impossible without reading every transcript—AI now surfaces these connections.
  • "Historically it was impossible to pick up all those data points unless you really consumed every piece of news... With AI today... you’re now able to pick that up much more efficiently as relates to your thesis."
    — David, [09:33]

B. Building Research and Pre-Buy Analysis

• AI Speeds Up Decision-Making:

  • Enables faster and more effective “triaging”—surfacing disqualifying factors early (e.g., existential risk, management compensation misalignments).

  • Moves deep-dive analyses (like CEO comp trends, credibility analysis, revenue recognition flags) earlier in the research funnel.

  • Analysts can now process more ideas and devote creative attention to higher-potential opportunities.

  • "A lot of times I could have killed an idea much quicker once I’d surfaced enough information... AI can be really helpful in that process is giving you enough baseline context to triage the idea."
    — David, [12:27]

  • Pattern Recognition:

    • AI can surface subtle, qualitative patterns (e.g., management’s track record of guidance and delivery) that previously required time-consuming manual work.
    • "Where AI can be really useful is surfacing patterns that historically were pretty painstaking to put together."
      — David, [15:03]

C. Sourcing & Generating New Ideas

• AI for Competitive Mapping & Screening:
  • Identify companies exposed to trends (e.g., tariffs) or nuanced mental models (e.g., high-quality business facing temporary headwinds).
  • Challenge: Many investors have “salt” intuition for what they want; AI can help formalize and systematize that.
  • "Finding ideas that fit the mold... is challenging for two reasons. One is that just requires a lot of qualitative reasoning... Two, even enunciating that clearly is hard. ... But when it works, it’s incredible."
    — David, [18:03]

3. Skills & Best Practices for Harnessing AI

A. Prompt Writing

• Humanize the Task:
  • Treat prompts like you’re instructing a smart but context-light overseas analyst via email: give background, desired outcomes, task guidelines, and domain knowledge (e.g., “Management is always biased—stay skeptical.”).
  • Balancing specificity (to reduce “hallucinations”) with flexibility (for synthesis/creativity).
  • "Defining the task, the context behind the task, some specific outputs if I want it, some guidelines and some kind of domain context... If you can send that to a human and they understand the task, you’re probably going to do pretty well with the model as well."
    — David, [22:29]

B. Supplying Context & Documents

• Task-Driven Choice:
  • Upload documents for accurate, structured outputs (e.g., numbers, model builds).
  • For exploratory or creative tasks, let the model iterate and discover using both context and web search, with explicit instructions for breadth and direction.
  • Be mindful of accuracy needs—the higher the stakes, the tighter the controls on inputs and verification.
  • "Some tasks, if you’re 99% accurate, you’re 0% useful… If there’s a factual error... but it’s more of a triage exercise, that’s okay."
    — David, [25:59]

C. Experimentation

• Continuous Learning & Edge Discovery:
  • Dedicate time (e.g., 15%) to trying features and approaches as AI capabilities shift rapidly.
  • Develop a template library of “10 things you wish a model could do” and continuously retest as new model versions are released.
  • "Having a process by which folks experiment and really spending some time pushing the models... gives you a better intuition of where the models are today and maybe where they’re going."
    — David, [28:43]

D. Peer & Firm Adoption

• Bottom-Up is Best:
  • Top-down mandates for AI usage can backfire due to diverse individual workflows; best adoption comes from offering firm-wide tools but allowing individualized adaptation.
  • Documenting research processes and decisions enhances future model (and employee) learning and transferability.
  • "Adoption has to happen bottoms up and people need to feel comfortable with it to be able to continue making high quality investment decisions."
    — David, [33:07]

E. The Value of Documentation

• Invest in Process Memory:
  • As models grow more capable, having internal documentation (memos, trade rationales, decision logs, etc.) becomes a strategic asset for both humans and AI agents.
  • "Models... shift from being a helpful research tool to something that lives within a firm and can execute a research process. Their ability... is a function of how much data they have on how you and your firm operates."
    — David, [36:42]

4. The Future Frontier: Context, Memory, & Agents

A. Context Windows

• Rapid Expansion but Still Limiting:
  • Context windows (the amount of info a model can “see" at once) have grown (e.g., Gemini ~1 million tokens) and models are better at using that context intelligently.
  • For structured, targeted tasks, load only what’s necessary; for complex projects, full corpora are becoming practical.
  • "Models are generally if you’re using a ton of context, you can do better if you’re loading that in because you’re looking for very specific data points... I wouldn’t really hold back. I would give it pretty much the entire corpus."
    — David, [39:55]

B. Memory

• Short-Term Usefulness, Long-Term Potential:
  • Current memory implementations are inferior to humans but can add value for persistent, repetitive context.
  • In the future, AI models with learned “lived experience” across a firm’s investments could surpass human judgment and integrate pattern recognition/scar tissue—unlocking AGI-like attributes.
  • "Imagine having a model that has lived every single one of a firm’s investments... a world where it becomes so much more powerful than any given human."
    — David, [42:25]

C. Agentic AI

• Agents for Reasoning and Multi-Step Tasks:
  • Definition: Models capable of taking independent actions, reflecting, and iteratively working towards a goal.
  • Progress evidenced in software engineering: models iteratively examining, modifying, and debugging code bases.
  • For investment research, this is more challenging—but progressing: in the future, AI agents may operate like full-fledged analysts, running research projects autonomously.
  • "What’s really exciting... is agents have started to work... The form factor of can these models reason iteratively and arrive at complex answers...? The answer is yes..."
    — David, [46:51]

Notable Quotes & Memorable Moments

• On AI’s impact on research productivity:
  "There’s a ton of information out there... Consuming it in a way that was efficient and additive to the research process was challenging... AI can be really useful without replacing the parts of the process that are important for building conviction."
  — David, [06:23]

• Conviction through context and AI:
  "I was one of these guys that could never really outsource model building. I always had to build a model myself in order to feel conviction... But there were certainly aspects that... limited me in terms of how productive I could be..."
  — David, [06:23]

• AI and the magic of idea generation:
  "There are few feelings as exciting in this business as getting served up a pitch where you're reading it, you're like, oh my God. Yeah, this is really exciting. Clear my calendar..."
  — David, [20:51]

• On prompt-writing wisdom:
  "The real insight I think comes from... iterate on this. The cost of sending a single query is trivial... It’s really like a two-way dance to end up in a spot where the AI is acting like a useful analyst."
  — David, [31:33]

• On documentation as strategic edge:
  "Firms where they have documented memos and even write up short paragraphs on why they made a trading decision... It’s a pretty reasonable bet that having that data at a minimum is helpful for the humans, but will certainly be helpful for the machines."
  — David, [36:42]

Important Timestamps

| Segment | Timestamp | |------------------------------------------|--------------| | Guest intro/background | 04:52–05:54 | | Key pain points AI can solve | 06:23–08:44 | | Position monitoring w/ AI | 09:33–11:21 | | AI in early-stage research | 12:27–14:34 | | AI surfacing patterns & trust in mgmt | 15:03–17:02 | | Sourcing new ideas and mental models | 18:03–20:51 | | Effective prompt writing | 22:29–25:27 | | Experimentation as a skill | 28:43–30:42 | | Adoption/collaboration within firms | 33:07–35:27 | | Documentation and the future of AI | 36:42–38:31 | | Context window (limits and usage) | 39:55–42:03 | | Memory in AI—practical & philosophical | 42:25–44:53 | | Agentic AI and autonomous research | 46:51–49:23 |

Conclusion

This in-depth conversation bridges the tactical and the philosophical: it’s a masterclass in how investors are extracting value from AI right now—by using smarter workflows, writing better prompts, and embracing a culture of experimentation and documentation—as well as a forward glance at how features like agentic reasoning and machine memory could fundamentally change the investment process. David Plon’s advice: keep learning, keep documenting, and always leave space for creative experimentation as the frontier accelerates.

Find more episodes and resources: joincolossus.com