When it comes to working smarter, Del-One Federal Credit Union President and CEO Dan McCarthy hopes that artificial intelligence, in small doses, can make a difference.
In the last five months, the federal credit union has been slowly rolling out AI tools to aid its 81,000 members in securing loans. Their software vendor has provided them a model to look at loan applications, examine data like credit history and risk assessment before reaching an employee for the final approval.
McCarthy is quick to point out that the model also includes Del-One’s own underwriting criteria, not just standard data points.
“That was important to us because we’re stewards of our members’ money, but also it takes a step to look at additional data points so we can say yes more often,” he said. “It’s still early, but the only data point we have is that our delinquency rate is trending much lower.”
A clearer real-case example of AI at work at Del-One is Breezy, a virtual assistant that can answer members’ simple questions, like about balances. That cuts down the calls customer service representatives handle, and freeing them up to take calls on more complicated matters.
Generative AI has burst onto the scene as a revolutionary tool for financial firms, big and small, as a way to improve productivity by analyzing data, generating content and identifying supply chain issues. Big banks like JPMorgan & Chase Co. have already used AI for simple tasks like answering emails to risk assessment, fraud detection and brainstorming for marketing materials.
“It’s an unbelievable thing, and we already have hundreds of use cases for it. It will change and enhance jobs, and eliminate some jobs,” JPMorgan CEO Jamie Dimon told reporters in February. “But when we talk about technology, you have to consider it has the potential to make lives better, like cars got rid of the horse and buggies.”
But for smaller companies here in Delaware like Del-One, it’s a chance to explore more productions and solutions and bring them to market faster in time.
Hypothetically, an AI model could study a member’s bank statements, notice a recurring pattern of spending money at a mechanic. That could prompt Del-One to offer an auto loan to that customer.
“It really is about being able to dive deeper into trends and giving our members a solution for something before,” Del-One Chief Technology Officer Marc Kidwell said. “But it’s also about testing things and seeing where the model is touching our systems and where the results are coming from. We’ll be doing our due diligence, and we need total transparency from our vendors.”
Big Data, Big Challenges
Banks have been using AI in the last three decades, namely in data classification and finding patterns in large chunks of data. Automation has helped reduce costs, while relying on machines to handle manual tasks and reducing errors. Roughly 40% of financial services companies rely on machine learning for fraud detection and forecasting, according to a report from Standards & Poor.
That same report found that machine learning in banking, financial services and insurance accounted for about 18% of the total market as of 2022.
But banks and financial services may have the biggest potential to tap into generative AI— models that can create new content like programing scripts, text, images and videos.
Global analytics software leader FICO has been on the forefront of the AI revolution for three decades, starting with complex machine learning models in the 1990s, bringing its early fraud detection program online and exploring language learning models. Today, FICO builds industry-based models to give clients new insights on available data – and it holds hundreds of patents in AI and machine learning models to create its own algorithms.
A real life example is FICO’s fraud detection model. If you swipe your credit card, your transaction history is summarized by a model and compared to your past purchases.
“One of the things human beings have a difficult time is understanding how to combine lots of complicated, disparate information,” FICO Chief Analytics Officer Scott Zoldi said. “The more data we have, we can improve decisions and turn it into more loans and open access to more credit. What we want to do is essentially treat AI as a tool where the human is always in control of what it’s learning.”
But Zoldi also sees big risks in companies not using AI ethically. There’s already concerns that automating claims processing and loans have discriminated against people of color. He recommends that corporations develop a step-by-step protocol on what algorithms are used and how to test its flaws – and that employees still spend time with the data while in the process of building the models.
There could be 100 versions of the model a developer tests before the final one makes it to market.
“You need a human in the loop to basically spend whatever time it takes to really understand the data and see its biases,” Zoldi said. “We want to treat AI like a tool where the human is in control of what it’s learning – and if we see what it’s learned because it’s dangerous or unstable, it can be removed.”
That’s why he’s leading the charge to help educate the next generation of programmers and analysis with the 12- week FICO Educational Analytics Challenge.
Right now, Delaware State University students are working with real-world data from FICO to find gaps in the information and develop a program to eliminate bias as much as possible.
“With so many people leveraging AI, not everyone knows how to see where it could be failing. You have to take the time to examine the model you’re building and make sure it works,” Zoldi added. “The generation we have today wants to be right at the front of technology. Everyone is thrilled with it, but one of my missions is to make sure that generation that’s going to help us figure this out will really understand the implications of it.”
Prepare for the revolution
Many big banks and financial firms are still testing the waters when it comes to generative AI, since the risks and regulations have yet to be clearly determined. But in the meantime, Delaware’s workforce will still have to contend with a seismic shift in financial technology skills.
At the University of Delaware’s Lerner College of Business and Economics, Oliver Yao is already thinking on how to face the future. The university has doctorate programs in data analytics to position students for data scientists roles at banks.
“The magnitude of this revolution is the same level as when the internet came 30 years ago. It’s going to change our lives forever,” Yao told the Delaware Business Times. “Where I see the future is making sure our students understand not only the domain knowledge of these companies, but also data techniques and make the right decisions.”
This semester, UD launched a master’s degree program in analytics and a certificate in generative AI in business. Yao said that with data-rich sectors like banks, there’s an opportunity to use it for forecasts, based on bank statements, stock information and finding liquidity issues.
“Something that could take a group of analysts days to do can be generated within minutes – and will still need to be checked,” Yao added. “Some jobs will be replaced, but the ones who have an understanding of how it works or be trained in it will be at an advantage.”
Meanwhile, training programs like Zipcode Wilmington are starting to consider how AI can be integrated in day-to-day work. Kristofer Younger, director of education, said it will push the need for programmers and software developers to work with the high-powered tool.
“It’s like a laser cutter. It’s extraordinarily powerful, but you still have to work your way through the problem to produce what you need,” Younger said. “With intermediate and advanced programers, it definitely can help create code faster, but whether it’s better, I’m a little skeptical.”
Younger also pointed out that major financial firms could be slower to adopt generative AI, since so much of it is open-source licensed program – or free to use for all. If a bank uses a proprietary software system, using open source code could open up potential lawsuits.
From his perspective, there will still be a need for programmers with strong math and statistics skills to get a handle on algorithms – and understand how to scale the models. In essence, there needs to be a human in the loop to break down complicated problems and solve them.
“There’s going to be plenty of cases of cookie-cutter codes, but there still will have to be humans to break it down and see how it needs to fit together,” Younger said. “It’s like calling an expert and asking them to do your homework. Why wouldn’t you just hire the expert instead?”
Editor’s note: a previous version of this article incorrectly stated the title of FICO Chief Analytics Officer Scott Zoldi. We regret the error.
