Jack Woodruff on AI and the Human Work of Stock Picking

Jack Woodruff introduces himself as the founder and former CIO of Candlestick Capital Management, a long–short equity fund, and a former portfolio manager at Citadel’s Surveyor Capital. Kepler CEO Vinoo Ganesh opens the conversation by asking what fundamental investing could look like five years from now.

Woodruff says long-term stock picking will still involve assembling a mosaic of information and finding a view that differs from the market. He expects the way people synthesize information and the baseline of readily available research to change dramatically, while acknowledging it is too early to know precisely how analyst and portfolio-manager roles will evolve.

Ganesh asks whether a single strong analyst might cover 400 stocks. Woodruff says people’s attention remains a bottleneck: investment management involves deciding where to spend limited time. The more names an investor covers, the less time is available for each, and additional research eventually has diminishing returns.

AI could help an investor cover more names or prepare a general background faster. Woodruff gives a hypothetical example of work that used to take six to eight hours being prepared in minutes. That could free time for proprietary research, channel checks and analysis of an industry, rather than guarantee a proportionate reduction in staff.

Competition changes the calculation. If one team finishes the old workload sooner, another may use its additional time to look for a further piece of information. Woodruff expects analyst work to change because the same preparatory tasks may no longer be what firms most value.

Asked what junior analysts may stop doing, Woodruff rejects the idea that serious investors will never read filings or transcripts. Summaries can save time, but someone close to a company still needs to absorb the information. He expects some large-scale data processing to become easier to delegate to tools.

For Woodruff, a promising use is helping an analyst identify which companies deserve attention. The goal is not to automate every decision but to direct limited research time toward a possible divergent view. Ganesh asks whether time saved on data work will make analysts smarter or simply busier.

Woodruff expects the baseline amount of work to rise. If everyone has access to the same analysis, it becomes a requirement for competing, and investors look for the next layer of information. He describes an investment mosaic as several supporting data points from different sources rather than one decisive fact.

Ganesh asks whether Woodruff, who began as a CPA, would sign off on a model-produced number without knowing its origin. He says no. Even human data entry can cause costly errors; before relying on automated outputs or trading decisions, he wants numbers to be accurate and individually auditable.

He notes that some users already ask models to help build financial models but then have to audit the figures manually. That reduces the time saved. The ability to trace and check each number, he argues, is a crucial condition for trust, rather than an assumption that errors will simply disappear.

Ganesh asks what information Woodruff would refuse to let an AI vendor see. Woodruff says security is a matter for his technology leadership, but he would be nervous about internal data leaving the firm and wants strong isolation. He does not present a detailed technical security design.

Woodruff describes an initial automation phase in which tools and data are assembled, followed by a more interesting phase in which they help analysts decide where to focus. Earnings previews and discounted cash-flow models are means to an end, he says; a fund still needs to find views that differ from competitors.

When Ganesh asks whether an edge will come from better data, questions or judgment, Woodruff answers that all matter. Common information can point researchers toward a special project, but people must investigate, build conviction and act before everyone else sees the same opportunity. He treats this as an opinion about a competitive process, not a guarantee of returns.

Asked how he would build a new fund, Woodruff jokes that he would buy Kepler, then says he is unsure of the full staffing mix. He expects to need senior stock pickers and may need fewer traditional analysts, alongside skills such as investigative research and data engineering.

He contrasts the traditional pyramid—many analysts preparing information for senior stock pickers and portfolio managers—with a potentially more vertical structure. Even so, he may still want people gathering proprietary information to feed the research process. He presents this as an evolving possibility, not a settled staffing plan.

Ganesh asks whether AI helps a small fund that can act larger or a large fund that can act smaller. Woodruff says funds have different mandates, ranging from rapid turnover to long holding periods. He expects each to be able to do more, but says the effects depend on its strategy.

He illustrates idea selection with a portfolio of 20 names: if the candidate pool grows from 100 to 200 or 400, the fraction selected falls from 20% to 10% or 5%. More choice could improve selection only if research on the larger pool remains as solid. The arithmetic is an illustration, not evidence of a higher realized hit rate.

Woodruff says a very large firm may span many asset classes and strategies, making direct comparison difficult. In his view, the flow of good ideas is central to an investment firm: as an idea plays out, the firm needs another researched opportunity to replace it.

Finally, Ganesh asks for an overhyped AI idea. Woodruff is skeptical of a fundamental hedge fund with no analysts: he thinks an automated system might replicate an earlier fund, while competitors continue to improve. He notes that quantitative funds have long automated decisions, but doubts humans will disappear from fundamental investing in the way enthusiasts suggest.