David Plohn

AI Summary

→ WHAT IT COVERS David Plohn, founder of Portrait Analytics and former investor at Baupost and Slate Path Capital, explains how investors practically apply AI tools across their research workflows. He covers specific use cases for position monitoring, pre-buy research, idea generation, prompt engineering techniques, and emerging capabilities like agentic AI for investment analysis. → KEY INSIGHTS - **Position Monitoring Beyond Individual Names:** AI enables investors to efficiently track sparse data points across entire value chains rather than just watching individual companies. For an Expedia position, investors can now automatically surface relevant commentary from Marriott or Hilton transcripts about OTA distribution strategy or travel demand trends without manually reading every hotel company's earnings materials, capturing insights previously only available to dedicated sector specialists. - **Prompt Engineering Framework:** Effective prompts follow an email-to-overseas-analyst structure with five components: specific task definition, context explaining why the task matters, desired output format, specific guidelines for execution, and domain knowledge transfer. Include skeptical instructions like reminding models that management teams are positively biased, since AI models are trained to be helpful and tend toward overly positive interpretations of corporate commentary. - **Pre-Buy Research Acceleration:** AI moves time-intensive analyses earlier in the research pipeline, enabling faster idea triage. Tasks like mapping five years of CEO compensation metrics across proxies or cataloging every piece of management guidance to assess credibility now happen instantly rather than taking hours. This allows analysts to kill flawed ideas faster and spend deep research time only on ideas passing initial quality screens. - **Idea Generation Through Mental Models:** AI excels at finding companies matching nuanced qualitative frameworks, like identifying previously high-performing businesses with attractive fundamentals facing temporary headwinds where the market is revaluing franchise value. This requires translating implicit pattern recognition into explicit queries, but when successful, surfaces compelling opportunities that fit an investor's specific edge rather than just screening on quantitative metrics. - **Documentation as Future AI Infrastructure:** Firms should document all investment thinking, trading decisions, and research memos now, even without immediate use cases. As model context windows expand and agentic reasoning improves, this historical data becomes valuable training material for firm-specific AI systems. Models will eventually need this context to operate like analysts who understand a fund's unique decision-making patterns and past experiences with specific names. → NOTABLE MOMENT Plohn reveals that when Portrait first launched using GPT-4, their agentic AI system required a 30,000 token system prompt with extreme constraints because models could not self-correct when going down wrong paths. Modern models now successfully run multi-hour autonomous tasks in software engineering, dynamically adjusting plans based on errors, suggesting similar capabilities will soon work for investment research workflows. 💼 SPONSORS [{"name": "Portrait", "url": "https://portraitresearch.com"}] 🏷️ AI Investment Research, Prompt Engineering, Position Monitoring, Agentic AI, Investment Workflows