How Will Chat GPT Affect Banking?

Hi, and welcome to another episode of What's Going On in Banking. I'm Ron Shevlin, chief research officer at Cornerstone Advisors and a senior contributor to Forbes, where I write the Fintech Snark Tank.

 

When we set out to start this podcast, we decided we were really going to focus on breaking news in banking and get to the "so what" behind it. Today's episode is about something that broke a couple of months ago, so I hope you'll give me a pass. The podcast hadn't even started when it happened.

 

I want to talk about how ChatGPT, the new AI tool developed by OpenAI, could affect banking. The technology has taken the world by storm. It reached roughly 100 million users within its first few months, making it one of the fastest-growing applications ever.

 

Before we get into ChatGPT itself, though, I want to put it in a broader context. The technology is really a form of conversational AI, and I've been writing a lot about conversational AI recently. I've put a stake in the ground that conversational AI is becoming a foundational technology in banking. To me, that means it is becoming a must-have technology alongside digital banking, mobile banking, and the other core pieces of the technology stack.

 

Why is it so important? First, banks have to do a much better job of providing digital service. When people hear "conversational AI," they often equate it with chatbots. Chatbots are a form of conversational AI, but then the whole category gets dismissed because people say, "Come on, Ron. You've used chatbots. The user experience is horrible."

 

In some cases, that's true. But chatbots are really an entry point. They're an early stage in the evolution of conversational AI. I wrote about this a couple of months ago in a report called The Chatbot Journey, which looked at how chatbots are evolving from simple customer-service tools that answer one question at a time into what a lot of people call intelligent digital assistants. I'm not particularly crazy about that term either, but the idea is important.

 

The difference is that these tools are beginning to use machine learning to create more conversational and advisory capabilities, rather than simply resolving one customer-service question at a time. That's one reason I think conversational AI is moving toward becoming a foundational technology.

 

Another reason, and one I think bankers often overlook, is internal employee support. It's just as important for an employee to be able to turn to a conversational AI tool and ask questions. Banks are struggling to recruit and retain people, and I often joke that if you walk into a bank branch today, you may end up talking to someone who has five minutes of experience.

 

My wife handles most of the banking in our household, and it's frustrating when she walks into a branch and realizes she knows more about the institution's products than the employee she's talking to. I've spoken with financial institutions that are trying to treat chatbots, or more advanced digital assistants, almost as members of the team by using them to support the employee experience.

 

The third thing driving the move toward conversational AI is the introduction of ChatGPT itself. It launched only a few months ago and has already become a major phenomenon.

 

To help me make sense of ChatGPT, where it's going, and how it may affect banking, I've asked Chris Nichols, director of capital markets at SouthState Bank, to join me. I chose Chris for a couple of reasons. First, I read just about everything he writes. He's brilliant, and I agree with almost everything he says, which makes him one of the only people in the world I agree with other than my wife, of course.

 

The other reason is that he recently published a LinkedIn post outlining 15 use cases for ChatGPT in banking.

 

Chris, thanks a lot for joining me. I want to get into those use cases, but I don't want you to go through all 15. Which of them do you think will have the biggest impact on banking?

 

Thanks for having me, Ron. I'm a big fan of your work as well, so the feeling is mutual.

 

As far as ChatGPT goes, we've been experimenting with it and its predecessors for a little less than a year. We're starting to roll out training more broadly across the bank. We think working with this technology is going to become a necessary skill, so we're trying to familiarize bankers with it and help them learn how to use it effectively.

 

The first step is productivity. Any banker can use it right now for things like summarizing an email thread, summarizing other work, or helping proofread a document. You can use tools like Grammarly and work through a document step by step, or you can put it into ChatGPT and have it make corrections all at once. The accuracy isn't perfect, but in many cases it's comparable and much faster.

 

One of the biggest productivity benefits is what we call solving blank-page syndrome. If you need to start a policy, a marketing blurb, website copy, or almost any other written document, ChatGPT gives you a starting point. And it does that pretty intelligently.

 

We've tried to measure the quality of what it produces, and I often describe it as a B-level student. It's not perfect, but it's not bad. More importantly, it gets you started. It can create paragraphs, pages, or even full drafts, whether you're working on legal language, marketing copy, or something else.

 

It also does a good job of adopting personas. If I need to get inside the head of a small-business owner, a doctor, or a government employee, I can ask it to write a paragraph explaining treasury management to that particular audience. In this era of hyper-personalization, we're constantly creating multiple ads, headlines, taglines, and opening paragraphs. ChatGPT can speed that work considerably.

 

So that's step one: productivity. Step two is using it with employees. Our intranet is a good example. I think we have one of the better intranets in the industry, but it is still extremely hard to find information. The bigger the bank gets, the harder that becomes. When we were a $400 million bank, everybody knew everyone and information was easy to find. As you grow across state lines and across different offices, being able to use natural-language generative AI to find internal information becomes incredibly valuable.

 

The third step is what you mentioned earlier: putting it in front of the customer. That could mean using our own data and the ChatGPT API to productize information, whether it's through a chatbot or a tool that can help answer customer comments on Google, Yelp, or other channels more efficiently.

 

I loved the use cases you shared. Three of them stood out to me. One was the ability to create code. You can tell it to write C++, Java, Python, or another language. Have you actually done that, and have you found the code to be reliable or bug-free? What kinds of things have you asked it to build?

 

It works very well in limited use cases. I think the coding capability gets dismissed sometimes as something that's only useful for IT, but there is a gap emerging in banking where business lines increasingly control pieces of technology and don't rely entirely on the IT department.

 

You've heard a lot about low-code and no-code environments. If it's truly no-code, that's great. You can connect visual boxes and build workflows. But as soon as you want to do something more advanced, you often need at least a little bit of code.

 

Whether that's Java, Python, C#, or something else, ChatGPT becomes very useful for the business-line user who has a basic understanding but isn't a professional developer. I have enough coding background to be dangerous, but I'm not a developer. If I need to write Python, I can run the problem through ChatGPT, get reasonably good code, and then have our developers review it. That gets me much farther much faster.

 

I think this will become increasingly important because more technology responsibility is moving into the business lines. A lot of the younger bankers we hire already have some coding experience, often in Python. This becomes another tool that helps them extend those skills.

 

Another use case that caught my eye was product design. You alluded to this in your opening comments when you talked about asking ChatGPT how to explain treasury management to a doctor. But I'm interested in the broader product-design capability and the idea of taking on a specific customer persona.

 

I spend a lot of time telling midsize community financial institutions that they need to find their niches and develop products and services for specific segments, such as young physicians. Can you actually ask ChatGPT, "What are the unique banking needs of a doctor, retiree, or engineer?"

 

You can, and it does a pretty good job. Again, I wouldn't say it's perfect, but it solves that blank-page problem and gets you thinking. There is almost never a time when I use it and don't get at least one insight I hadn't thought about immediately, even if I might have gotten there myself three weeks later.

 

It can also look at product structures, suggest features, and even help create rough wireframes. You can ask it what attributes a product should have for a particular segment or how an existing product could be adapted for doctors, for example.

 

There are two skills bankers need to pay attention to here. The first is prompt writing, which is really the ability to ask the right question. You're going to get it wrong at first, but you'll get better. The second is the ability to drill down and refine the result. You can start with a broad product idea, then ask it to create pieces of a webpage, a workflow, a wireframe, or a list of the elements that should appear on a product page.

 

For example, you could ask it to help design a good user interface for sending a wire or making a request-for-payment transaction over a real-time payments network. It can get you surprisingly far down the road.

 

I'm glad you brought that up because I want to ask specifically about training. One of the things I struggle with is simply knowing what I can ask and how I should phrase it. How are you training people at the bank to use the tool?

 

The first step is putting it in everyone's hands and giving them the basics. We want people to feel comfortable logging in, writing a prompt, and figuring out the use cases that make sense for their jobs. Everyone is going to use it a little differently.

 

For me, summarization and getting started on paragraphs and pages are instrumental. I use it multiple times a day. It really is a game changer.

 

One thing that turned the light bulb on for me a few months ago was realizing that I had shifted some of my behavior away from Google searches and toward ChatGPT searches. In certain situations, it's simply more efficient because it synthesizes information quickly. You have to know when to use it and when not to, but that ability to synthesize information is a major change.

 

So the progression is: get bankers comfortable using the tool, teach them to write better prompts, then talk about the professional version, APIs, and use cases where the bank can connect the technology to internal systems or data. That's when you can begin building things like an internal chatbot that crawls the intranet and answers questions from bank documents.

 

It still doesn't sound like an easy training task to me. I can't imagine HR giving one generic ChatGPT training session to software engineers, product managers, marketers, and lenders. It seems like it would have to become pretty domain-specific.

 

I think it's probably easier than you think. Learning to use ChatGPT reminds me of learning to use Google effectively. Search engines weren't completely intuitive at first either. Over time, people learned how to search better, how to use snippets, and when to keep digging.

 

ChatGPT may actually be easier because it works in natural language. People can improve very quickly, even in a single session, just by using it.

 

All right, last question. Do you have any compliance concerns about banks using ChatGPT?

 

Absolutely. Like anything else, garbage in means garbage out. Whatever you get from the system needs to be checked and reviewed with a skeptical eye. You have to ask whether the information is accurate.

 

If you're looking up well-established information, such as details about Harry Truman for a speech, the system is probably going to be fairly accurate because there is a lot of reliable information available. If you're asking about a new or emerging area where the information online is full of hype or uncertainty, the system can fall down pretty quickly.

 

So the first responsibility is checking the output. The second is understanding the data and the process behind the answer. If you're using a professional or enterprise version and connecting ChatGPT to your own data, you may have more control over that. But you still need to understand how the information was synthesized and how much confidence you should have in the result.

 

Sometimes the use case is relatively simple. You may ask it to create a visualization or identify trends in a financial statement. Over time, as people begin using it for basic credit work, it could become a precursor to more automated underwriting. But that's exactly why banks need to learn how these systems work before rolling them out too aggressively.

 

I think that's where compliance has legitimate concerns. If the industry rushes to deploy the technology without understanding how the application produces its answers, that's a problem. That's why we're spending time now learning the system before using it more broadly.

 

My sense, Chris, is that a lot of the people raising compliance concerns are still thinking primarily about customer-facing use cases. And yes, there are serious questions if ChatGPT produces advice that gets transmitted directly to a client. But a lot of the use cases you described are internal and employee-facing, where the compliance concerns are more limited and manageable.

 

Chris Nichols, director of capital markets at SouthState Bank, thanks a lot for being on What's Going On in Banking. And everybody, thanks for joining us. Look for us on your favorite podcast platform, please subscribe, and of course, give us a five-star rating. Thanks a lot. We'll see you on the next episode.