Key Takeaways

  • Allocators are establishing strict internal lines against using machine learning tools to write investment memos, treating memo writing as cognitive labor that produces the investment edge.
  • Laura Hill deploys software to process private market capital call notices and quarterly statements, converting inconsistent sponsor reports into standardized Excel models for team review.
  • Brian Sugrue treats conversational models as sparring partners for pre-meeting prep, stress-testing internal portfolio assumptions by running automated counterarguments before manager check-ins.
  • The practical objective of automation across institutional desks is expanding the time available for debate, rather than generating automated conclusions or outsourceable theses.
  • These operational lines are defined by Advocate Health's AI Operating Principle.

The Advocate Health's AI Operating Principle

  • Core Mandate: Augment human expertise, accelerate routine work, and improve decision quality.
  • Non-Negotiable Boundaries: Preserve human judgment, data confidentiality, and final accountability.
  • Decision Hurdle: Do not delegate core recommendation generation, position sizing, or forward-looking investment thesis creation to automated models.

When This Works (and When It Doesn't)

This framework operates cleanly for institutional investment offices managing high document volume across hundreds of external private equity, venture, and real estate partnerships. It thrives when applied to ingestion bottlenecks: quarterly valuation updates, capital calls, and manager distribution notices that follow no universal template. Automating that operational drag gives junior analysts hours back each week without risking proprietary data exposure or outsourcing portfolio strategy.

Where the rule meets friction is in speed-sensitive direct deals or secondary co-investments with compressed closing windows. Teams that refuse to run rapid quantitative screening or automated valuation comparisons across comparables will fall behind faster syndicates. The failure mode happens when teams misapply the boundary: treating document extraction as strategic thinking, or conversely, letting a synthetic text generator summarize qualitative GP track records and dilute the diligence committee's skepticism.

Why It Matters

When junior staff use language models to draft investment memos, they eliminate the exact friction where conviction gets tested. As Hill puts it, “A lot of allocators are quick to say, 'Oh, it can draft an investment memo.' What's the purpose of your investment memo? To me, the purpose of an investment memo is to sit down and make yourself think.” The text of a memo is secondary; the discipline of ordering evidence, defending a risk premium, and committing to an underwriting thesis is the real product.

Sophisticated LPs are splitting their technology adoption into two distinct buckets: back-office standardization and pre-mortem sparring. Sugrue notes that software should create room for human evaluation rather than replacing it: “It's not about producing analysis that we could not do before. It's about creating a little more space for us to use judgment in an analysis.” By running automated red team debates against their own theses before meeting GPs, allocators arrive with sharper questions and firmer downside awareness. The divergence in LP performance over the next decade will not come from who wrote memos the fastest, but from who protected the cognitive space required to price risk.