Managing Change Across Modern Financial Operations - with Nathaniel Bell of Wells Fargo

Daniel Faggella (0:12)

Welcome everyone to the Emerge AI in Business Podcast. Today's guest is Nathaniel Bell, Data Management Director at Wells Fargo. In this episode, Nathaniel unpacks why fragmented systems, not AI models, are what stop regulated institutions from running workflows consistently at scale. He explains why agendic AI risks industrializing bad data quality, how tying every tool to a measurable business outcome gives auditors the evidence that they need, and why no process should be automated before it's standardized. Just a quick note for our audience, that the views expressed by Nathaniel on today's program do not reflect that of Wells Fargo or its leadership.

According to Nielsen, 91 percent of podcast listening happens alone, indicating deep, undistracted attention, ideal for complex B2B messaging. To learn how leading brands and AI startups connect with enterprise AI buyer audiences at scale, download our media kit at emerge.com/ad1. That's emerj.com/ad1.

Now, the conversation with Nathaniel.

Yolandi de Weerdt (1:32)

Nathaniel, welcome back to the show.

Nathaniel Bell (1:34)

Thank you for having me. Glad to be back.

Yolandi de Weerdt (1:36)

I'm excited to see what is different from the last two times that we have spoken to you, and your mindset change on certain things, and maybe even new friction points that we didn't really know existed previously. But before we get into all of that, we need to start with something that I found quite interesting in preparation for this. I found that Deloitte's latest banking article says that AI adoption across banks is still throttled by brittle and fragmented data foundations.

And everyone's saying the same cliche thing. It's not because the models aren't good enough, it's because nobody's dealing with the plumbing underneath it, something like that.

When we talk to people running these types of workflows, it's very rare that they say, we've got one big broken problem. It's normally four or five different tools, each dividing or making its own decisions about document validations or policy checks, and nobody's really owning their entire picture. So I feel like that's where we should start today. What makes it so hard to run these workflows consistently at scale?

Nathaniel Bell (2:38)

Yeah, that's a great question. And I think when you think about the challenges, I think especially in my perspective and from my experience in banking or any really heavily regulated industry, everyone first jumps to technology. But in my experience, the technology I think is rarely the problem, right? If you look at the history, a lot of the challenges, consistent outcomes across the board, but that has been after thousands of processes, dozens of platforms, multiple lines of business have come together as there have been mergers, as there have been acquisitions, right? As products have been created. So it isn't really a question, I think sometimes of how do we automate, it's how do we scale trust while we automate? And I think those are different questions, different problems, and that's where I like to kind of frame it. So I know maybe to answer your question directly, you said how, why is it hard to scale? And I think you hit on one of those first reasons, which is system fragmentation. Like I said, a lot of these massive institutions evolve from acquisitions, product expansion, regulatory requirements, right, is a big one, and then compounded by years of technology change. Even outside of finance, you know, in the nuclear industry that I was in for a while too, I mean, you were talking, running on very old code bases, but they're baked in to the risk and safety analysis that have gone through the regulators. So it's very difficult to change. So if you put all that together and you have all these multiple stores and multiple teams, operationally, everyone thinks they're doing right, you know, in their own system and in their own process. But we continue to struggle to execute consistently because there's not a single source of right context. I'm not necessarily truth. I don't want to go that far, but that context is what matters, especially, right, when we talk about AI and agentic AI. So I think that's the first major reason that fragmentation. It's not that just like you said, a system touches, or a workflow, I should say, it touches 10 systems, 20 teams. That consistency becomes a problem for data, I think, before it becomes an operational problem. And I actually roll on into the second one, just because I mentioned it. I think regulatory requirements or policy interpretation is one of the second pillars, right? We don't, I mean, it would be awesome if we did, but we don't normally get requirements or regulations shipped to us as executable code, right? We need our guidance and we have to interpret that. Anyone listening who has played the telephone game knows how a small difference in interpretation can have a big difference in the outcome. And I think, you know, the next piece would be the regulatory burden. Again, a lot of the industries that I've had experience in are heavily regulated, and that becomes, I think, a scale problem, right? Every new regulation, every new change to that regulation adds complexity, a control requirement, not an expectation, some layer of complexity. The reality is that most institutions, I don't think, can afford to keep adding manual oversight.