Getting real about AI

“Daddy, why do we need artificial intelligence when we have the real kind?”

Jeffrey McMillan’s young son recently posed that question to his father, who is chief analytics and data officer at Morgan Stanley. That’s a heavy question for someone who has headed up the Wall Street firm’s efforts to bring AI to 19,000 financial advisors, comprising most of Morgan Stanley’s investment advisors.

“I’m a poster child for AI, but it is really hard to do,” said McMillan during the recent DataDisrupt Financial Services 2018. McMillan says there is much that AI can do, but there is much that it cannot do.

“Sometimes we need artificial intelligence,” said McMillan, “and sometimes we don’t. It’s not about AI, it’s about clients and our business model. Part of it is having AI in place to move our business forward and some of it’s a work of fiction.”

Elaborating, McMillan said that the real state of AI in financial services today is quite mixed.

To make his point, McMillan described a conversation he had with an AI expert who told him that, if asked, he could produce a basic bot in ten minutes. And it would be a terrible bot. McMillan said that producing a robust bot can take months, even years, and requires the involvement of many experts throughout a financial institution.

“We don’t have a technical problem here,” he said. “We have a business problem. The hype has passed the reality.”

Ultimately, McMillan’s presentation answered his son’s question with this conclusion: For what banks and financial advisors need to deliver to customers today, ultimately a partnership of human augmented by AI works best.

What does AI do today?

AI has five chief functions today, according to McMillan:

1. Detection—AI is really good at this, finding anomalies in data sets for detection of fraud and other factors, for example.

2. Robotic processes—AI handles repetitive tasks well. “Maybe not the sexiest application,” said McMillan, but it’s really powerful.”

3. Recommendations—This represents the partnership between human and AI now, and McMillan detailed his firm’s experience with that later.

4. Providing answers—Humans still do this task better than does AI, according to McMillan.

5. Reasoning. “These technologies don’t reason yet,” said McMillan. “They are not as smart as humans are.”

What are you really going for?

McMillan said financial services executives should stop using terminology like “solving for AI.” Instead, he explained, they should be speaking of the problems they are attempting to solve.

AI represents a tool, he said, which is capable of handling some tasks, but not others. He said that asking what a bank’s AI strategy was made as much sense as asking “What’s your fax strategy?” when that was a fresh technology.

“There is no solution that we can plug into that will solve all of our problems,” said McMillan. In fact, at the current state of AI technology as applied to Morgan Stanley’s advisory business, that’s not even expected.

McMillan pointed to several stressful life events, including going through a divorce and having a child diagnosed with a life-changing illness. Anyone wanting an advisor’s help planning for the financial effects of such matters does not want to talk to a bot.

“In the world of digital, the most powerful driver of revenue for our firm, the most powerful driver of client satisfaction, the thing that reduces attrition the most, is a human-to-human conversation,” said McMillan. “Human contact is the most powerful driver of our business model today.”

“Experience, intuition, and empathy” remain key components of the customer-advisor interaction, in McMillan’s words.

So, what Morgan Stanley has been using AI to analyze customers’ preferences and additional factors in order to enable the technology to perform as a tireless, constantly up-to-date assistant to the human financial advisor.

“I want to make the humans having conversations supersmart,” McMillan explained. Everything possible to customize is customized and personalized, he said. Some AI applications are engineered to place customers into groupings that generate recommendations based on the behavior of those cohorts. McMillan resists that categorization.

“I want to sell things to you based on you and your behaviors,” he explained.

Joining AI and human expertise can deliver, said McMillan. He expressed this as an equation: “Experts + Algorithms + Intuition=Differentiated Insights.”

Understanding a tool and its applications

Putting aside science fiction films, McMillan said, application of logic and reasoning in the same sense that humans do so isn’t here yet.

“This is what AI can’t do yet,” said McMillan. “We are very, very far away from this.”

Part of this is fundamental. In the absence of data, informed decisions are impossible for both human and bot.

“The algorithms are easy. You have to have the data,” said McMillan.

Even if applicable data exists, it may exist in a format that isn’t AI friendly, according to McMillan. “We live,” he said, “in a society where the data was never designed for artificial intelligence.”

“We are very, very far away from human cognition,” said McMillan. “I don’t think we’ll see it in my lifetime.”