What a diligence process spends its hours on.

Before anyone commits capital, someone reads. Filings, transcripts, the data room, a stack of expert-call notes — the hours before a judgment is possible go mostly to taking in what is already written down. It is the part of the work least visible in the final memo and the part that consumes the most time.

This is where generative AI has quietly become standard. In AIMA’s 2025 survey of alternative managers, 95% reported using it in their work, and 58% expect to widen its use in the investment process itself — up from 20% two years earlier. Allocators have noticed: 60% said they are more likely to back a manager that invests seriously in the technology. The front office has adopted; whether is no longer the question.

What adoption conceals is that most of it is ungoverned. The same survey found half of managers under a billion in assets running these tools with no restrictions at all — which is to say improvising, not building.

Our reading of where the value actually sits: current models are genuinely strong at the reading — ingesting a corpus and surfacing what is in it — and genuinely weak at the judgment that follows. Handed the judgment, ungrounded, they are confidently wrong; a peer-reviewed study of verifiable legal questions found leading models inventing facts in most of their answers. The useful system does not make the call. It compresses the reading, so the team’s hours go to the call.

That is a different thing from a subscription or a chatbot. It is built around a firm’s own sources and its own questions — and deciding what to hand it is the first piece of the work.