Is artificial intelligence software mature enough for the investment industry?

Jeff McMillan, chief analytics and data officer at Morgan Stanley, has some doubts.

Wealth managers need to provide accurate, specific advice about clients' portfolios, he pointed out. And artificial intelligence isn’t reliable enough to provide consistent insights right now, he said at the InVest conference in New York this week.

"If you ask Alexa for a song and she gives you the wrong one, it's not a big deal,” McMillan said. “If you ask an AI engine a question about a customer holding and its answer is about the wrong asset class, it is a big deal. The level of accuracy is important."

The idea that artificial intelligence software could help financial advisers assist their clients more efficiently has been floated since 2013, when ANZ Bank in Australia became the first bank to announce it was working with IBM's Watson technology. At the time, the bank said it was planning to use Watson to assess new customers’ financial situations more quickly and comprehensively than a financial adviser could and help create financial plans. The bank spent more than a year feeding the system documents and questions and answers, and reportedly put it into production in one location last year.

A few other investment management companies have gone public with artificial intelligence for financial advisers. BlackRock has built a risk management platform called Aladdin that is used in-house and offered to institutional investors. It also offers Aladdin Risk for Wealth Management, a subset of the Aladdin platform, to wealth management clients. Goldman Sachs uses the AI-based financial research platform Kensho. UBS, Deutsche Bank and others are using an AI engine called Sqreem.

McMillan said that in a firm like Morgan Stanley, much of the detailed knowledge is in people’s heads. Chatbots are not equipped today to handle queries like "what the IRA policy is for a 65-year-old client who lives in the state of Utah," he said.

People with questions are more likely to chase down the experienced person who knows the answers.

"That's how human knowledge is largely transferred in an organization, even today," McMillan said. "Search really doesn't work. But that information exists in the four walls of our organization. Imagine a world in which all that knowledge could be transferred into a system and people could contribute that knowledge, and then using natural language processing you'd be able to access that instantly." (Morgan Stanley is building such a system; this will be covered in a future column.)

Jeff McMillan, chief analytics and data officer at Morgan Stanley.
Getting it right
"If you ask Alexa for a song and she gives you the wrong one, it's not a big deal,” says Jeff McMillan, chief analytics and data officer at Morgan Stanley. “If you ask an AI engine a question about a customer holding and its answer is about the wrong asset class, it is a big deal. The level of accuracy is important."

The AI systems out there today can't really do this, he said. And AI might not even be the right idea.

"I'd rather have curated intelligence," McMillan said. "I'd rather not live in a world where I'm trying to get a neural net to figure out what you're trying to say to me. I'd rather have a world where I ask you what you're trying to say, I tell you to say it in a structured way and I enable my financial advisers to access that knowledge in a structured way."

McMillan acknowledged that natural language processing, a component of AI, has gotten good at understanding spoken questions and getting the intent.

"Where things tend to fall down is curating answers out of unstructured text in a way that's going to help you manage people's wealth," he said.

The critical missing piece, McMillan said, is curated content that is tagged and searchable.

"Which is why in our experience you actually need human beings to manage this today in combination with really good technology," he said.

Jody Kochansky, managing director at BlackRock, also said imperfectly managed data is an obstacle to using artificial intelligence in managing wealth.

"Our experience is that big data is still today a big problem," he said, referring to client, message, operational and other types of relevant data.

It would help if firms had a data-centric culture, Kochansky said.

“At the end of the day, you need everybody to be participating in a data consortium. If I identify bad data, I'm empowered to go clean it,” he said, adding that in that situation when one person fixes a data set, everyone else using that same data should benefit from it.

Drew Sievers, CEO of Trizic, a company that provides wealth management software for large firms, also sees limitations to AI in this field.

“AI is emerging technology,” he said. “It’s not as sophisticated as everybody thinks. In this wealth space as we talk about new fintech, there's a lot of emerging technology that's being deployed; in some cases either the technology is not quite there yet, or the technology is there but the implementation of that technology isn't quite yet. In the area of AI, it's the former.”

Sievers agrees with McMillan that natural language processing has gotten better. But he also agreed that content needs to be structured in a way that the processing can read and retrieve the right information.

“You're effectively tagging content, because people don’t write in the way that NLP is coded,” Sievers said. “NLP doesn't work perfectly. NLP doesn't get sarcasm, for instance. That's improving and it will get there, probably.”

Jason Mars, CEO of Clinc, which has developed an AI personal assistant called Finie, said AI engines can extract specific tidbits of information. He objected to the idea of making people ask questions in a structured way.

"Demanding that I learn a structured way to articulate that so the system understands is just more burden, more complexity," Mars said. "The power of AI is we have an opportunity to remove that complexity so I can say it any way I want and the system will be able to give me that information in a clean, insightful way."

After meeting with 10 different vendors in the space, McMillan said he has not seen any that fulfill his firm’s requirements.

"I think we'll get there, we're just not there yet,” McMillan said.

Kochansky said the use of AI in wealth management is inevitable.

"It's going to happen and it will be highly embedded in everything you do," he said.

Kochansky said AI can help with tasks today in areas like automatically rebalancing clients' portfolios.

Where it doesn't work yet is "when your client's having one of those life moments — maybe they're struggling with a decision about how they're going to fund a retirement portfolio or maybe there's a death in the family," he said. "Those are the areas where the adviser wants to spend more time building that relationship and being there when their client needs them and the computer is not really going to help."

It’s safe to say that AI technology is still immature. Even IBM refers to Watson as a toddler. It’s also clear that AI requires clean, accurate, well-organized data to work with. For many firms, this is a journey that is just beginning and it will take several years to provide accomplished virtual assistants to financial advisers.

Editor at Large Penny Crosman welcomes feedback at penny.crosman@sourcemedia.com.