Can AI Help Asset Owners Solve Their Data Problems?

Clean data may be the biggest hurdle for allocators hoping to use AI to help improve their investment process, according to Purdue Research.

Art_AdvisorsAI_0116 (1).jpg

Illustration by II

Allocators are buried in data.

A growing trove of quarterly manager performance reports, due diligence documents, outlooks, and market research has made allocators’ jobs — essentially finding the best investments — far more challenging.

So when ChatGPT launched in November 2022, some tech-savvy asset owners got excited that large language models could help them understand and use this mountain of data more efficiently.

In early 2023, Purdue Research Foundation’s Blair Webb embarked on a project with his team to understand how asset owners could do just that.

The director of investment research and analysis for the foundation conducted a survey in 2023 of allocators with between $1 billion and $40 billion in assets about how they were thinking about implementing AI into their investment processes.

The team identified nine key use cases for AI. Seven of those cases were focused on improving productivity. They included drafting informational portions of investment memos, creating call and meeting transcripts, reviewing pitches, creating a searchable report database, standardizing reporting across information sources, uploading information to databases, and assisting in prioritizing potential investments.

“The main use cases are just on this big problem of all this data we receive on existing investments and being able to get insights from it,” Webb said.

His team also identified two ways to use AI that could drive investment alpha: Flagging LinkedIn connections who may help with due diligence, and identifying and filtering information like news on investments in a portfolio.

Respondents ranked the automatic data collection use case as the one they were most interested in. A searchable document repository that creates insights was ranked second. Pipeline prioritization ranked last.

“I would say that everyone has AI front of mind, [but] it’s still unclear to us which of those categories has the most efficient tools that we want to use,” Webb said by phone.

Webb and his team saw a divergence among survey participants based on size. Those that were larger, or that had more resources, were focused on how AI could improve alpha, while the more resource-constrained organizations were looking to improve productivity with the nascent tools.

“Ultimately everyone is trying to improve their returns,” Webb said. “The big organizations are less concerned about productivity gains.”

AI’s rapid evolution has presented challenges for allocators when it comes to implementation. “Spending resources, time, and money on tools where you’re not sure if things will change substantially,” is a hurdle for asset owners, according to Webb.

Along with the external survey, Webb and his team explored the tools available now for allocators. They found that legacy research management systems and customer relationship management tools are limited in their ability to adopt AI. Meanwhile, AI-first products fail to solve all of the use cases for allocators, instead just focusing on one or two problems.

Purdue also hired an outside development firm to test whether they could build a searchable, “insight-able” repository of documents. The development firm tried to build a chatbot like ChatGPT that could answer internal data questions.

According to Purdue’s analysis, AI could answer questions, such as ‘Who are the partners of XYZ Capital Partners?’ or ‘Summarize this macro information from Fiscal Year 2023.’ But AI struggled with the more quantitative data insights. According to the analysis, for a question like ‘What is the best-performing company owned by XYZ Capital Partners?’ the system output would sometimes return not just incomplete insights but also inaccurate answers.”

Purdue found that the chatbot had trouble taking in tables and graphs and outputting accurate information. In other words, data quality must be a focus for systems being built for allocators moving forward.

Purdue’s foundation, for its part, will continue searching for the answers to this AI conundrum.

“It’s a new vertical in our organization rather than a project that will start and end,” Webb said. “We also don’t claim to have all the answers. The main reason for us sending out the white paper was to try to expand the network and folks that want to share ideas and want to collaborate on the topic.”