Fintool's Nicolas Bustamante on using AI to improve in investing

Episode Summary

Executive Summary: Andrew Walker interviews FinTool founder Nick Wustamont about how AI can improve investor workflows. They focus on breaking research into discrete tasks, using AI for earnings memo drafting, screening, transcript/podcast digestion, and proxy/comp analysis. Nick argues AI will increasingly augment, not replace, investors—but only if paired with human judgment, especially for edge cases and fat-tail decisions.

Main Topics: AI as a workflow optimizer for investors (Priority: 5/5): Nick's central recommendation is to decompose investing into specific tasks and delegate repetitive, standardized parts to AI, such as memo drafting, screening, and document review. From basic search to deeper research automation (Priority: 5/5): The discussion contrasts using AI as a 'super Google' versus using it for more advanced workflows like cross-company comparisons, memo generation, and offline research agents. Limits of consumer LLMs and hallucinations (Priority: 5/5): Nick explains that generic chat models often hallucinate, making them unreliable for finance unless grounded in trusted sources like SEC filings, transcripts, and verified internal documents. Human judgment in concentrated, qualitative investing (Priority: 5/5): Walker and Nick debate whether AI can replace the nuance needed in concentrated long-only value investing, with both concluding that human-in-the-loop decision-making remains essential. AI's impact on finance roles and skills (Priority: 4/5): They discuss how AI may eliminate routine analyst work and elevate higher-level skills like pattern recognition, judgment, and synthesis across more complex datasets. Using proprietary and internal data to improve outputs (Priority: 4/5): The conversation covers how uploading internal memos, call transcripts, and custom research can improve AI relevance, especially when the user wants the model to reflect their own process and biases correctly. Potential for AI-native investment products (Priority: 4/5): They speculate about future AI-driven qualitative funds, offline agents that monitor markets continuously, and systems that proactively surface opportunities instead of only answering questions.

Key Arguments: The best immediate use of AI for investors is to break workflows into small tasks and delegate the repetitive pieces to the model. Standardized research outputs like quarterly memos and screening reports are especially suitable for AI automation. Generic LLMs are unreliable for finance because hallucinations and source confusion can corrupt answers, especially around company-specific metrics. Finance AI should start from trusted sources of truth such as SEC filings, earnings calls, and verified transcripts. AI will likely not eliminate investing, but it will make the market more efficient and shift alpha toward higher-order judgment. Concentrated, qualitative investing is harder to automate than quantitative trading because the sample size is small and the most important cases are edge cases. AI may become useful for scouting opportunities overnight, summarizing large information sets, and scoring companies on multiple qualitative and quantitative dimensions. Human oversight remains necessary because AI can miss context, footnotes, and non-standard compensation structures that materially affect the thesis.

Data Points: FinTool benchmark accuracy: 98% - Nick cites FinanceBench results, saying FinTool scores about 98% accuracy on equity research questions. ChatGPT benchmark accuracy: 40% - Nick says ChatGPT with search scores around 40% on the same finance benchmark. Perplexity benchmark accuracy: 50% - Nick says Perplexity scores around 50% on the benchmark. AI code contribution: 5-10% to 100% - Nick describes his software engineering team moving from AI writing a small share of code to essentially all of it. Typical investor research time saved: 30 seconds to 5 minutes vs. a full day - Walker says AI helped him pull Caesars acquisition commentary from transcripts in minutes instead of a day. Hedge fund portfolio example: 8 stocks held for 2 years on average - Walker uses this as an example of a low-frequency, concentrated investor with limited training data. Backtest sample size example: ~51 data points - Nick and Walker estimate a small sample from 20 years of investing with an eight-stock portfolio and three-year holding periods. Company coverage example: 50 names - Nick references analysts at banks covering about 50 names and needing rapid report turnaround. Large firm AUM example: $200 billion under management - Nick mentions a large firm like TCW as an example of enterprise-scale customers. Renaissance example: 90 billion; 30% a year - Nick cites Renaissance as a proof point for machine-driven investing success, though the numbers are spoken informally.

Pivotal Quotes: "The best way to improve your usage of AI is to have a clear breakdown of your workflows." — Nick Wustamont: Nick's initial answer on how investors should start using AI more effectively. "The future of AI system where you don't go and ask your question. The AI push you relevant info." — Nick Wustamont: Nick describing a proactive, offline research agent that monitors and surfaces opportunities. "I think you'll have to dedicate that to AI and try to learn on more complex information." — Nick Wustamont: Nick responding to the idea that routine research tasks will be automated, forcing humans up the value chain.

Implications: AI is already valuable for finance, but mostly as a research accelerator. Its biggest impact will be on routine analyst work and source-grounded synthesis, while edge-case judgment, thesis formation, and final investment decisions will still need humans.