AI in the Investment Office: 8 CIOs on What’s Working

AI is reducing time spent on specific investment-office tasks: one CIO says Claude summarized a legal document with 185 tracked changes in about a minute, while AI-assisted analysis of global borrowing costs cut work estimated at two to three weeks for two analysts to a couple of days. Another speaker says AI-generated board memos save about a day.

AI is expanding teams’ capacity for research and underwriting without yet establishing better investment outcomes: an active-equities team reduced portfolio-wide forensic screening from multiple days to under an hour, while a secondaries team says it can evaluate more deals involving portfolios of tens to more than 100 companies. A CIO says AI has not yet helped with investment decisions or position sizing.

Speakers describe AI as supporting, rather than replacing, human investment judgment: it can help prepare manager-meeting questions and examine investment theses, but one CIO says AI has not yet improved investment decisions, and another speaker cautions that AI-generated memos can feel empty. A thesis-coach tool is intended to surface assumptions and evidence gaps for teams to assess.

Data protection and reliability are constraints on deploying AI in investment work. One office excludes identifying family information and uses enterprise services that do not train on submitted data; separately, an AI pacing model double-counted a commitment, while another firm is holding back on agents because email access could expose extensive data.

Useful AI analysis depends on organized, high-quality data: speakers point to consistent naming, tags, and searchable records as foundations, while one says preparing data is time-consuming and unstructured prompts reduce accuracy. A team is centralizing and inventorying data in a lakehouse before using LLMs to identify patterns.

Data autonomy influences whether organizations build internally or rely on vendors, and vendor performance remains uncertain. One organization wants to retain ownership of its data and AI outputs, while Laura Hill says her team has limited confidence that existing vendors have meaningfully embedded AI and is treating vendor-based AI as a lower priority.

AI is being applied to standardize investment operations, but human review and practical workflow limits remain. General Gist converts inconsistently formatted manager statements and capital calls into a uniform Excel template for analyst review, while document retrieval remains difficult across eight to ten separate data rooms and vendor routing tools have disappointed one team.