Why Morgan Stanley is turning advisors loose to build their own AI agents
Processing Content
- Key insight: Unlike many firms that simply provide AI agents for wealth management, Morgan Stanley lets advisors try building their own. The 50,000 attempts made so far have yielded mixed results.
- Expert quote: “If you want to build your own agent, and if you have a level of knowledge and complexity, the world is your oyster.” — Koren Maranca, AI head for Morgan Stanley’s wealth management division
At Morgan Stanley, advisors aren’t restricted to using just AI agents and other systems devised for them by the firm. They can also go out and make their own.
So far, Morgan Stanley advisors have made nearly 50,000 attempts using natural-language prompts to produce AI agents designed in large part to give them more time with clients by automating routine tasks.
“If you want to build your own agent, and if you have a level of knowledge and complexity, the world is your oyster,” said Koren Maranca, head of artificial intelligence for Morgan Stanley’s wealth management division. “And you can curate any agent that’s going to be impactful to you and your team.”
Maranca said advisors have built agents specifically to track portfolio holdings that are of particular interest to their clients.
“Or it could be something more complex of looking across their book and pulling in all of the outgoing flows over a certain amount every day and delivering that information to them at a certain period of time on a daily basis,” she said.
Why most of the new AI agents don’t pass the test
Of course, that’s not to say anywhere near the majority of those 50,000 agents are in day-to-day use. As is the case with most technological experiments, most don’t make it past early testing.
Maranca said agents that prove useful enough to be embraced by many people on an advisor’s team are subjected to careful reviews to make sure the answers they provide are both accurate and complete. Agents that seem promising enough for possible adoption throughout the firm are put through even more extensive testing.
But doing all that for 50,000 agents, most of which never going anywhere, would be impossible.
“Sometimes people build an agent, they use it twice, and then they never touch it again,” Maranca said. “So we’re not going through comprehensive reviews for an orphaned agent. But those agents that we see are actually being incorporated into an advisor’s practice, those go through a review and approval post-use.”
Morgan Stanley was one of the earliest large wealth managers to enter into a contract with one of the big AI companies. Within months of OpenAI’s release of its ChatGPT large language model in 2022,
How Morgan Stanley uses AI to cut down on call center queries
Beyond giving advisors leeway to tinker with AI independently, the firm has devised its AI @ Morgan Stanley, or AI@MS, to develop a set of agents designed to help in various common financial-planning situations. Some of the most popular agents available to all advisors at Morgan Stanley offer assistance in everything from providing answers to financial-planning questions that previously would have gone to an internal call center to organizing an advisor’s day to projecting income from a high-yielding savings account.
Morgan Stanley
Maranca said many of the uses Morgan Stanley initially found for AI@MS four years ago “are now commonplace” among wealth management firms. But Morgan Stanley’s systems have since been through at least two iterations, and Maranca believes they maintain a place on the industry’s most advanced frontier.
One of the earliest uses Morgan Stanley found for AI was to put it between advisors and the subject-matter experts sitting in the firms’ internal call centers. Maranca said roughly half of all advisor inquiries are now first taken up by an AI agent.
An advisor, for instance, might have a technical question about how to close a client’s trust account when it still has a balance. It’s the sort of inquiry that a call-center expert might have to research independently, possibly by consulting AI, before providing an answer.
So why not give an agent the first crack at it? Maranca said advisors can still go to the call center if they find the AI assistant’s responses unsatisfactory.
“And if the model doesn’t know the answer, we’ve created an experience that’s smart enough to know that and say, ‘You know, I can’t answer that question, but I know you’re asking a very detailed, complicated matter and you have an issue related to a live order or a maturing security or whatever,” Maranca said. “And then the person is being intelligently connected to the right part of the service desk to answer that question well.”
Maranca said Morgan has grown confident through repeated testing that the answers provided by its AI agents are accurate. What it’s more concerned about now is making sure they are also complete.
“What we did is we pulled in experts from across the entire firm, over 1,000 people reviewing the model outputs to help us define how complete the answer is, until we got to a threshold where we were at a productionalization,” she said.
AI agents can lay out advisors’ daily plans, project savings income
Meanwhile also popular, Maranca said, is an agent that simply helps advisors plan their days. By opting in with a single button click, Morgan Stanely advisors can receive a daily “morning briefing” containing a list of suggested tasks for the day.
A typical briefing, Maranca said, will cover topics such as, “Here’s what you have coming up. Here’s how you should prioritize your time. Here are some of the items that I’ve already developed for you related to what you have coming up.”
Advisors can be provided with dossiers running to many pages of information and recommendations useful for a scheduled meeting.
“This is a good example of something that an advisor could do themselves, but also as part of that morning brief, something that could be proactively delivered to them if they decided to opt into that experience,” Maranca said.
Unlike many wealth managers, Morgan Stanley hasn’t put an exact number on how many hours a week it thinks AI is saving its advisors. Morgan Stanley executives have instead said they expect AI will allow current advisors to take on more clients without automating away aspects of financial planning most requiring human judgement and experience.
Maranca said another commonly used agent helps advisors project investors’ income from money invested in what the firms calls its signature-savings accounts. These high-yielding accounts, offered through Morgan Stanley, are designed to give clients an alternative to money market funds and other options for investing cash in an easily retrievable way.
Maranca said Morgan Stanley’s AI agent allows advisors to project clients’ likely income from putting certain amounts of money into a signature savings account.
“So if they have $5 million, they can see what that projection would be for the client, and use it related to their then follow-up conversations with the client related to potentially using a signature savings account,” Maranca said.
Other firms place restrictions on employees’, clients’ uses of AI
Many of the uses Morgan Stanley has found for AI fit in with what industry surveys suggest advisors most want out of technological innovation. In its
Even as Morgan Stanley finds ever more uses for AI agents internally, some of its industry rivals are setting stricter limits on what employees can do with burgeoning technology. Bloomberg reported on Monday that Goldman Sachs had banned bankers and traders from using Anthropic Claude Code and Microsoft’s GitHub to write code to automate tasks such as monitoring client trading flows and modeling market changes.
“We make certain AI tools available to our people to determine the best use cases across different parts of the firm,” a Goldman spokesperson told Bloomberg. “After a period of experimentation where we assess a variety of factors including productivity and risk management, we put them in the hands of those who are best suited to leverage them for the long term.”
Meanwhile, much of the debate about AI in the industry revolves around the question of whether firms should allow outside agents like Meta’s Muse or OpenAI’s Dots to have access to their sites and possibly client accounts. Morgan Stanley, like JPMorgan and Charles Schwab, now blocks outside bots.
The next step for AI at Morgan Stanley
Maranca said Morgan Stanley’s priority is to make sure any client data exposed to AI remains private and secured. As for the phase of AI@MS, Maranca said she envisions a day when advisors will no longer have to go to an online dashboard and choose an AI agent they think is best suited to their purposes.
The firm is now working on what it calls AI@MS FA Assist, which will eventually be able to pull in an agent or agents automatically in response to natural-language prompts. The goal, Maranca said, is to eliminate “a lot of the point and click.”
“So, conversationally, an advisor will be able to analyze their clients’ portfolios, simulate risk issues or recession events … find additional models that might be interesting for the portfolio or for the client based on a whole number of attributes, and then ultimately curate that into a proposal,” Maranca said. “And that becomes a fully conversational engagement, just back and forth with the advisor, but, again, with underlying agents.”
