Hard-won wisdom from AI's early adopters

At Wells Fargo, Chuck Monroe (pictured) set up AI Enterprise Solutions in March 2017 in an effort to ensure all AI projects and AI-related tech vendor relationships were overseen by a single office. U.S. Bank, BBVA Compass and Banco Popular similarly have groups that help deploy robotics and AI throughout their organizations.

But Monroe didn't get immediate cooperation from other departments.

"We spent several months going around the organization trying to understand who is doing what," said Monroe, who is executive vice president and head of enterprise AI solutions. "As you can imagine, in a large organization like Wells you come in and you're now a new center of excellence at the company, nobody wants to take your phone call."

The fix, he found, was to get funding for all AI activities at the company.

"The first call they don't take, the second call is a voice mail saying, 'I have money,' and 10 minutes later they're calling me back," Monroe said.

Wellington Holbrook (pictured), the chief transformation officer of ATB Financial, now has a vision for how AI will pervade the organization.

The firm has been using technology called Eva from Finn.ai, and Holbrook expects to ramp up its use rapidly.

“We see a world within about 18 months where almost all work will be done by, assisted by or augmented by AI,” he said. “Eva is the interface to do that.”

Holbrook expects 25% of the organization will be completely automated by AI in 18 months.

“That's the work of 1,250 people and people know that,” he said. “Some teams will be affected more quickly than others. We need those people more than ever.”

Not everybody has been on board with the changes, however.

“We lost a quarter of our technology teams in the first six months of this transformation,” Holbrook said. “That was hard. Some people voluntarily left.”

For instance, as part of the transformation, the bank has migrated from Microsoft Office to G Suite.

“That was a hard change for a lot of people in IT, to the point where people were going through massive efforts to sabotage for various reasons, sometimes just because they have been using Microsoft Word since 1995 and can't change,” Holbrook said.

Others didn’t like the bank’s shift from waterfall to agile development, where software is being created in two-week sprints.

All bankers fear that regulators will object to their use of AI and a perceived “black box” in which decisions are made without human oversight. But that's no reason to resist using AI, bankers said.

"We've been working closely with our regulators for that very reason,” Monroe of Wells Fargo said. “It is a real concern.”

The bank has a model governance process through which every model has to go through independent review, in part for explainability.

“If we decline a loan, we have to tell you why,” Monroe said. “And if there's some variable that's sitting out here that puts you over the edge for the decline, and I can't explain that, that's never going to make it into production."

Moreover, banks can't just automate existing processes or layer AI on top of them. It has to be more fundamental than that.

"Automating bad processes won't do anybody any good," Monroe said.

Avoiding excessive regulatory scrutiny can also be helped by working with auditors and information security experts ahead of time.

"Especially if you're going to build these on the cloud, you open up part of your internal system to the cloud, the infosec team is going to have something to say about it," said Federico Montanana, program director, artificial intelligence and RPA at Banco Popular.