Research
How LPs Use AI: What Works—and What Doesn’t
The technology is useful in operational tasks, but allocators are hard-pressed to find artificial intelligence tools useful for precision tasks, according to the Emerging Allocators Association.
Reported by Matt Toledo
Artificial intelligence use is near ubiquitous among asset allocators, but few are accounting for the tech tools as a line item in their budgets and many cite the tools as best for operational, rather than analytical, tasks, according to data from the Emerging Allocators Association.
Approximately 97% of surveyed limited partners reported using at least one AI tool at work, with the median respondent using two, according to data from EAA’s members.
The EAA—a networking community for institutional investors—also found that Claude is the most-used tool among those surveyed, at 44% of respondents, followed by ChatGPT (27%) and Microsoft Copilot (18%).
Asset allocators working in organizations that are private market limited partners reported finding AI most useful for tasks such as drafting memos and writing up investment committee notes, the most cited workflow of all. Document summarization, meeting note-taking and market research were also cited as proven ways AI can add value to allocator workflows.
LPs also noted using AI for data room analysis, reference call prep and question generation, and financial modeling assistance in Excel, as well as with other tools and for drafting emails and internal communications.
When addressing precision tasks and those that needed proprietary context, such as sourcing and a firm’s voice, the survey found that AI tools were not well suited. Such scenarios include financial modeling and Excel—respondents noted that AI cannot build reliable leveraged buyout models.
Additionally, respondents noted that AI was not a good tool for sourcing general partners. The survey showed that network-driven sourcing, an important route for finding private market managers, cannot be replaced by AI. Additionally, allocators complained that AI cannot create structured data from documents formatted as PDF files and that hallucinations remain a problem with research quality.
These functionalities in which AI tools fail, the survey noted, could also point to potential product opportunities.
While AI use is nearly ubiquitous, only 6% of surveyed LPs said they had a budget line item for AI. Approximately 65% of respondents said their firms pay for all AI tools, while 22% said the firm pays some and individuals pay the rest. Another 8% said they pay everything out of pocket, while 4% reported using no paid tools.
The report suggested that respondents supplementing their organizations’ spending on AI tools signals that firm-provided tools often are not sufficient for the tasks for which users are accessing them.