Morgan Stanley Rolls Out OpenAI GPT-Powered Assistant for Financial Advisors

Financial services are moving forward with generative AI despite risks

Source: OpenAI illustrative mock-up of Morgan Stanley Assistant Using GPT-4

Earlier this year, OpenAI and Morgan Stanley announced they were collaborating on an effort using GPT-4 to organize its knowledge base. Jeff McMillan, chief analytics and data officer, wealth management, at Morgan Stanley, told CNBC in an interview this week that the new AI Morgan Stanley Assistant is now ready to launch. McMillan told CNBC:

 We have a knowledge base of over 100,000 documents and the idea is to bring that knowledge base—that differentiated intellectual capital—and put it together on top of the models of GPT-4; and essentially create an ecosystem that's private to Morgan Stanley that's controlled by Morgan Stanley and then making that content available to our financial advisors.

The idea is that we we take the most knowledgeable person at this firm and make that person available to every single one of our employees our financial advisors 24 hours a day 7 days a week virtually instantaneously. That’s the goal.

Assistants / Copilots Rise to the Top

Whether you want to call them assistants or copilots, these generative AI use cases are proliferating quickly. The copilot term is generally adopted by software companies that embed an assistant with generative AI powered functionality. Enterprises that create their own copilot are more likely to call it an assistant.

Morgan Stanley’s assistant is capturing the knowledge of its in-depth document repository. It is unclear how the company prioritizes that information based on recency or applicability, but that certainly will be a key curation decision for any assistant for wealth management advisors. “The bank spent months curating documents and using human experts to test responses,” according to the interview.

Testing is another consideration in order to minimize the incidence of hallucinations, which occur when the generative models provide incorrect information in response to queries. Using knowledge bases like Morgan Stanely’s approach helps reduce hallucinations, but it will not eliminate them.

McMillan also said other solutions are in the works. According to CNBC:

It’s just the first in a series of solutions based on generative AI planned by the bank, according to McMillan. The firm is piloting a tool called Debrief that automatically summarizes the content of client meetings and generates follow-up emails.

Unlike an assistant, Debrief is positioned as a tool that will save time in recording information and executing follow-up tasks. The assistant may be a productivity enhancer as well, but it is also designed to help the wealth advisors provide better service and advice.

The scope of functionality in terms of investing advice or market analysis is unclear. However, this is not a new idea. I was working on a similar concept related to voice assistant for wealth advisors five years ago for a global bank. The idea is not just to help wealth advisors become more efficient. A key goal is generally to level up the quality of insight and service across the workforce.

YMYL is in Scope

The Morgan Stanley announcement also highlights another interesting development in generative AI adoption. There was widespread speculation that heavily regulated industries such as finance and healthcare would be among the slowest adopters of the new technology due to concerns about accuracy. “Your Money or Your Life” (YMYL) domains are generally more cautious because of regulation and the often significant implications of errors.

However, we are not seeing that hesitation. Financial services and healthcare organizations are at least early experimenters and, in many cases, early adopters of generative AI. Why? Both industries work with a lot of unstructured data in natural language form of text or voice. Generative AI solutions based on large language models (LLM) are poised to provide outsized benefits for these industries. Thus, the more aggressive adoption.

Adopters in these industries are weighing the risk and identifying use cases that can provide a reasonable risk-reward tradeoff. Let me know what you think!