Strle spent nearly 13 years at JPMorgan Chase and close to six years at Wells Fargo. He’s now a little over a year into his tenure as Discover Financial Services’ chief information officer. “Essentially, all the transactions or money movement in the entire country will have one of those three companies on either end of that transaction,” he tells Fortune.

He also likes that the financial sector has a lot of responsibility to ensure that technology works properly. “You’ve got this area of banking where it’s really, really important to people when they swipe the card at the checkout or at the restaurant,” says Strle. “They’re counting on you, right?”

Discover and others in financial companies are also counting on big benefits from generative artificial intelligence. The technology could add between $200 billion to $340 billion in value annually, mostly due to productivity gains, according to McKinsey Global Institute’s estimates. But the sector has been fairly cautious when putting gen AI into production due to high regulatory constraints, fears over protecting customer data, and questions about high costs with hazy details concerning what the return on investment should be.

“A lot of the tools that are out there, which have a flat cost to them, puts a lot of pressure on us to understand the value,” says Strle. “There needs to be a better connection between the expense and being able to understand the value.”

This interview has been edited and condensed for clarity.

Fortune: What led you to join Discover in July 2023?

What really drew me to Discover was this unique arrangement where it’s direct to the consumer. When you don’t have the branch footprint, the dynamics of how you roll things out is dramatically different because we have to have consistency in how our products work on digital. There’s a dynamic across the industry for the players that have been around for a long time; trying to figure out how to be more direct to the consumer, more digital enabled, and drive great customer experiences. Discover started there. By nature of how we’re set up, we’re going to be technology leaning all the time.

When CIOs join a new company, they often talk about changes they made to the org chart or reevaluate vendor relationships. Have you made any of those bigger changes and, if so, why?

I generally take a very selective approach when it comes to making those reorganization changes. The major change that we made was creating a customer success organization. We wanted to put way more of our focus on what the customer was experiencing from their perspective when using our products and services, which spans multiple systems backed by multiple teams.

Financial institutions are using generative AI in a lot of different ways. What’s been your focus thus far with that technology?

There is the autonomous interaction with the customer, which is the highest risk element of what we do. We have to be able to explain very clearly through our policies and our procedures what those models are going to do, and they are going to do them consistently in a way that’s fair to the customer. [Then] there’s human-in-the-loop, where generative AI can help you do things. Summarizing calls [with generative AI] is in production now and helping us make sure that the agents who are human and doing the best that they can are getting backed up with this additional capability, which can help digest how the conversation went and can be used for coaching and feedback and understanding customer sentiment.

Why is it so important to keep humans in the loop when deploying generative AI? 

This is an emerging area of understanding of how humans interact with AI. It’s so good and so powerful at what it does that it’s almost training you to be less diligent. That’s a real dilemma. The better these tools get, even if we’re talking about human-in-the-loop, there is the risk that people start to shut their brain off because it does seem so good at what it does. And then the machine is operating the human at that point. That can cause a lot of unintended consequences and risks.

Financial companies tend to lean toward “build” versus “buy” when deploying technology. With generative AI, what is your thinking?

As we sit right now, I think it is difficult for us to fully take advantage of the commercially available products. We are super protective about our customer data and if that data is leaving our ecosystem, it is done with a healthy—borderline unhealthy—level of paranoia about where it’s going and how it’s going to be used. Then, you have to ask the question: Is this benefiting this commercial product and potentially leveraging intellectual property that belongs to us as a company? And we’re helping them develop a product that they can sell to more people.

How would you grade the progress the financial sector has made with generative AI when compared to other sectors?

I would probably describe it as being in the early phases of what will eventually be a very robust enabler. When you look at the chat capabilities, there is so much risk in potentially giving advice that can be harmful or might not be uniformly available to all of your customers. The other element is around really making sure you can maintain tight controls over your data and your data governance, while still being able to leverage these tools.