UBS Chief Data and Analytics Officer Joe Cordeira conservatively estimates the firm's U.S. advisors are saving 10,000 hours a month by turning to AI to prepare for client meetings.
And that's just
"We can now see: When are there opportunities that we can find a win-win situation for our client to bring over assets?" Cordeira said in a recent interview. "Yes, that benefits the client. They can get better returns. And obviously, UBS gets assets."
AI has also proved useful in digging up new client leads for UBS.
"There's public data out there about board memberships, affiliations and things like that," Cordeira said. "We can tell that the client and the prospect were there at the same time and likely have a relationship. So connecting the relationship network. There are a whole bunch of different ways we find these leads."
Other applications include providing insights culled from the firm's reams of market and investment data. AI can also alert advisors when clients might have an overconcentration of stocks in their portfolios or need help arranging a complex loan.
Overseeing it all for UBS is Cordeira, who joined the firm on its investment banking side in 2014 following a data research role at Morgan Stanley.
With many advisors
This conversation has been lightly edited for brevity and clarity.
Financial Planning: What's the main use UBS advisors have found for AI so far?
Joe Cordeira: Probably the flagship offering, what advisors are most using and are excited about, is something called STAAT [Smart Technologies and Advanced Analytics Team] Insights. It's an AI-driven engine that serves up highly relevant insights and opportunities to advisors.
It both provides tangible growth opportunities to advisors and helps them streamline service to their clients so that they can ultimately delight them — like when an advisor anticipates a client's needs and delivers something that the client didn't expect.
Obviously, when clients are happy, they refer their friends, they bring more money to UBS.
FP: What types of suggestions does the engine make?
JC: There are actually 400 different scenarios, some of which are investment-related. But a lot of the advice and services that our financial advisors offer are not. Our financial advisors are like the CFOs of their clients' lives, and they're looking at everything.
One of the things we do, before a client meeting, is automatically send a cheat sheet to our advisors — a client briefing that tells them what they should be aware of, what's going on with this client and where the opportunities are.
We've found that there are situations where our advisors can help our clients save on their borrowing costs. Because we have their mortgages here at UBS, we have capabilities in the investment bank for sophisticated clients that can reduce their borrowing cost effectively using a cross-currency swap. That's very specific to UBS.
Other examples: Many of our clients are business owners. And there's regulation throughout the United States around businesses of a certain size needing to provide retirement plans for their employees. We can help our clients do that. We can connect those dots and help the client get ahead of that regulation.
And then things more investment-related — there could be a surge in a stock. Something goes up and, all of a sudden, the client has a concentrated equity position. Well, of course, we alert the advisor to that, and we can offer them solutions. We rebalance the portfolio, but also things like a prepaid equity hedge could help.
But it's a range of things from life events to product opportunities to just being able to say happy birthday.
The way we did this was we worked backward. We looked at the best advisors, and we saw: How are they serving their clients? And we then codified that process in a set of AI models. Now all of our advisors are learning from the best advisors and how they're serving their clients.
FP: What are some other innovations you've found a place for, following the release of OpenAI's ChatGPT and similar generative AI systems?
JC: The STAAT Insights engine — that's actually pre-generative AI. Now we are taking these capabilities even further for advisors.
There are two solutions I would call out, although there's a lot more going on. No. 1, as is everyone, we're investing in AI chatbots that will help our advisors. We're using Microsoft Copilot [Microsoft's rival product to ChatGPT]. All employees of UBS globally have Copilot, and that's kind of our basic Swiss Army knife day-to-day productivity-savings tool.
Where I think it gets really interesting is when we then take the same capabilities and train models that are purpose-built specifically for our FAs. So if you look at the types of questions FAs are trying to answer for their clients day in and day out, Copilot is not going to really help with that, because it doesn't know our clients, doesn't know our business. So we're training these models.
Then a lot of people have taken their internal knowledge base and built a chatbot on top of that to basically replace search. So there's something called our branch guidelines and procedures. And it's thousands of articles on: How do I get this done for my client? Can I do this? Can I do that?
It used to be very labor intensive for advisor teams to find the answers to questions. They were calling an internal call center. Now they just ask a question and they get an answer right away. That chatbot alone is getting 20,000 or so prompts a month from our FAs.
The second strategy is our own internal research. We have a chief investment officer, and we have an investment bank and a research organization that I actually came from. And they are producing tons of content around what's going on in the markets. It's just impossible for an advisor to keep up with everything UBS is producing and saying. And then a client asks a question, the advisor has to go away, read an article, figure out UBS' view and then go back and talk to the client.
You can imagine AI is really good at this.
FP: One concern for any firm using AI to conduct research is the systems' tendency to 'hallucinate' — or make up information to produce a seemingly convincing but ultimately fallacious answer. What are you doing to mitigate this?
JC: All models will hallucinate. But I would say, as new versions of open AI models are produced, hallucinations become smaller and smaller, such that they're really not something we're very concerned about anymore.
We're constantly measuring the performance of the models, and it's very small percentages. We also enable our users to provide feedback around the quality of the output and are very consistently receiving high marks.
Hallucination is still a problem, and something we need to worry about. There is definitely a need to have a human in the loop and advisors need to cite where information came from. But two years ago it was a big concern, and it hasn't been a major challenge for us in this space lately.
FP: Many firms are
JC: An example of that might be just a simple meeting schedule. There's so much back-and-forth, so much time is spent scheduling a meeting. Well, we have a meeting-scheduling agent that detects that a client is looking to schedule a meeting. And then in one click, an advisor can distribute their availability to a client. You can imagine that, in the future, that can be even more automated. It's kind of a low-risk scenario where the advisor is fully supported by an agent in automatically scheduling a meeting.
Think about all the basic tasks that clients are asking for advisors and their teams to do: "Please send me this report or that investment research." We can anticipate that need without the advisor having to go find it. We can surface that in a conversation that our advisor is having with their client, and answer that right there. Whereas the insights I talked about previously were great before a meeting, but now we're talking about during a meeting.
FP: You figure that
JC: If you just look at how advisors would have done some things before, like research — by calling the call center — and now they're just getting answers to their questions, it's around 24,000 different prompts per month.
Where it's a little harder to nail down is Copilot. But if you think about a meeting summary versus someone taking notes and then keying them into the CRM [customer relationship management system], that could be 20 to 30 minutes per client meeting.
FP: Do you think AI will ever replace human advisors?
JC: We get that one a lot. Unequivocally, we think the answer is no, especially in our business. We specialize in high net worth, ultrahigh net worth clients. With highly complex clients, the human touch is absolutely critical. With things like trust, nuance, judgment, AI is not there yet and probably could never be there.
I think if you're thinking about something like your financial life, it's really important to have a human at the other end of it. So we're not planning on replacing advisors with AI. We're planning on empowering advisors with AI.
On a per-advisor basis, they will be more productive because of AI, absolutely. But that wouldn't be a reason to have fewer advisors.
FP: Would you say most of your advisors are embracing AI?
JC: Let me give you some stats here: 80% of them are actively using that STAAT Insights engine.
There are always going to be some advisors who are comfortable in their ways. They don't want to learn new things, and their book of business requires it less.
I would say overall, though, they're really embracing it. Copilot is at over 85% usage currently. And if you pull it all together, easily 95% of advisors are using some kind of AI in some shape or form.
Not everyone's using it as intensely, but it's very interesting. There's clearly a segment that are leveraging the tools to their full capabilities. I think it's really interesting to see how they're doing, and how that's growing with other advisors sharing best practices.
