Morgan Stanley's wealth division has put an AI agent next to the staff who handle client requests. It can look up information and carry out actions such as moving money, opening accounts and changing beneficiaries, and the division says it is already seeing productivity gains. The interesting part is why it works at all.
Each assistant moved closer to the work
The bank's first generative AI tool, in 2023, was an assistant that answered advisers' questions from a library of about 100,000 research documents. The next one, in 2024, was Debrief, which sat in on client video calls and wrote the notes advisers used to type up by hand. One adviser in the pilot put the saving at 30 minutes a meeting.
So the tools went from reading, to writing, to doing. Doing is a different kind of job. A tool that reads can be wrong and an adviser will catch it. A tool that moves money has to be allowed to touch only certain things, in certain ways, and to stop and call a human at the right moment. That is why the bank's technology chief for wealth stresses boundaries around what an agent can see, what it can do and when a person steps in. It is the same order of work other banks are choosing, governance first and agents second.
The agent needs to know who the client is
The boundaries are only half of it. The other half is knowing which client, which accounts and which systems the request touches.
Imagine a client who has a retirement account, a joint account with her husband and a trust. She calls to change a beneficiary. If the bank's records are organized by account, she appears three times, perhaps under three slightly different profiles, and a person has to decide which entries belong together. People do that by experience. An agent can only do it if the records already say it.
Morgan Stanley's wealth division made that change around eight years before the agent. It reorganized its data around the whole client relationship, and the first use was a program that suggested next steps for advisers. Nobody was thinking about agents then. The records were simply built the right way, and they turned out to be what the agent needed.
What the hours turn into
Here is a rough sum. If one of these requests takes ten minutes of switching between applications and the agent cuts that to two, a service associate who handles thirty a day gets about four hours back. The bank's own view of those hours is clear. A senior executive said in June that agents will let the firm grow customer support and plan administration without hiring thousands more people. The saved time becomes capacity for more clients, not shorter days. For the associate, the job shifts toward the calls that need judgment and away from retyping.
The same test works for a small business
A clinic keeps patients in a booking system, a billing system and an email tool. A distributor keeps the same customer in an order system, an invoicing tool and a spreadsheet of special prices. Ask an agent to rebook, refund or reprice and it will hit the same problem as the bank: three records, and no clear sign that they are the same person.
The cheapest preparation for agents is dull. Give each customer one ID that every system uses, and decide which system holds the truth when two disagree. Morgan Stanley spent years on that kind of work before anyone could show an agent off. A business that does it now will find agents easier to adopt than the one that buys the clever tool first.