AI and Fraud in Fintech: Expert Q&A from May 2025 Market Pulse

Both before and during each Market Pulse webinar, our audience submits their burning questions to our expert panelists, some of which we run out of time to cover in the live webinar and which we then answer in this blog. For our fintech-focused May Market Pulse webinar, our panel included industry experts Rich Franks, Global Head of Credit Strategy & Data Science at PayPal; Brett Manning, Product Manager for Personal Loan Machine Learning at Upstart; and Alex Johnson, Founder of Fintech Takes. Below are their answers on questions around artificial intelligence (AI) and fraud. 

Q: What do you think about the latest innovations in the fintech industry occurring around AI and how do you think they are starting to impact the market?

Rich Franks, PayPal: There's a lot of discussion right now about AI and three different use cases. The first use case is the customer facing use case. I think there's a lot of really cool stuff happening there, and while much of it benefits credit, much of it is not necessarily led by credit. Also, the question continues to get asked: What about credit decisions? What about underwriting? I actually just had a pitch from an AI company that was talking about how they could make business loan underwriting better with AIm which is interesting because business loan underwriting is fundamentally more manual. So, the pitch makes a lot of sense. I think it's probably a while before generative AI is the way that we do consumer underwriting, just because there's a lot of transparency and observability that is required for many organizations. It's not quite there yet.

The industry has been using AI for a long time, just not Chat GPT. So, in some ways, it's actually not new at all. The thing that I'm the most excited about right now, and especially at Paypal where we have many customers, many interactions with those customers, and massive amounts of data that we can use at scale to make better decisions using AI behind the scenes, so we can understand, interpret, and utilize signals, both within our decision and frameworks, as well as monitoring output frameworks.I spend a lot of time on that right now, and it's really cool.

Brett Manning, Upstart: Rich described Upstart as an AI Company, but we never set out to be an AI company. Yet, that's how we talk about ourselves, and Dave [Girouard, CEO of Upstart] told this story yesterday at an event that it kind of evolved naturally as we were trying to underwrite better, that we had to use more and more sophisticated techniques. So, we completely agree that investing in servicing and making sure that those repeatable standardized transactions all happen seamlessly, and that we don't need unnecessary agent time spent dealing with stuff that we do a thousand times a day. 

But at Upstart we do, or my passion at least because I'm the Project Manager for Machine Learning use the underwriting model to drive a lot of our growth. A significant portion of the growth we saw last year did come from underwriting improvements and that ability to differentiate risk between borrowers. And while we're not asking Chat GPT, “Is this a good borrower or bad borrower,” we are using techniques that were developed for LRM for the machine learning underwriting process. So taking the credit file, which is semi-structured, taking transaction data and cash flow data, which is even less structured, looking beyond that at what other unstructured data we can transform into model signals that we can then use to get useful information about how credit worthy the individual is. Then, we take the next step and turn the result into an explainable result that we can put into an adverse action notice, so that we can actually tell the borrower what the decision is, why it was made, and the way it was made in order to comply with all of the laws. 

Franks: And machine learning is AI, right? There's some very sophisticated machine learning techniques that are AI, and, sometimes, we get wrapped up in thinking of generative AI as AI. But there are lots of different flavors of AI. And by that definition, I've been using AI and underwriting since 2006. This is nothing new here. We've gotten better at it, faster at it, and we’ve gotten the regulators more comfortable with it, but, in some ways, it’s just an evolution of underwriting. 

Q: As it relates to fintechs, what are the ways that you can improve fraud detection or what are some of the trends you might be seeing in that space without compromising user experience? 

Franks: Well, fraud is really difficult, right? Because it's not a predictable symmetrical distribution. It's highly asymmetric. There may be years of nothing in this one fraud vector and then all of a sudden, thousands of hits in an hour. So I think the key with fraud is to recognize that you don't have it right even if you think you do. And then, you have to be able to react quickly when that happens in terms of being able to build models faster, get third party data sources in faster, and being able to monitor in the right way so that you can catch it. 

Fraud is so much about good process, recognizing that the third party data sources have gotten amazingly better over time. And then, if you have a lot of consumer interactions like we do, you have the benefit of that on top of all that third party data. But the name of the game is just cycle times.

Manning: There are some outstanding products out there that we can tap into to minimize that user friction. AI is clever in creating fraud and it’s even more clever in detecting it. So staying ahead of that is always going to be a moving target. But as long as you're not complacent, it should be doable.

Alex Johnson, Fintech Takes: I do think we are entering an era where bad guys using AI, especially generative AI, is going to be a big deal. I was talking to someone who was saying, with this latest generation of image generation models, that they are now able to produce digital identity documents, such as pictures of driver's licenses, that are able to pass through some, though not all, of the vendor products that perform ID verification. That's the first time that's ever happened. They can also replicate effective liveness if you don't have great liveness detection. Many of the fraud controls that are in place are built on the assumption that it would be incredibly expensive for a fraudster to be able to do this, so we don’t expect that to happen. Generative AI knocks a lot of those costs down to levels where it suddenly becomes mathematically viable for fraudsters to make money in ways that they weren't able to before or to circumvent controls that they couldn't circumvent before. 

Responding to that is going to be really important, and I think one of the ways to do that is going back to on-the-ground truth that is really difficult to get around. Bank account data is one that’s really hard to fake, because a fraudster would have to be able to set up a bank account with transactions and make it look like a plausible bank account for three years, which is a lot of work and expensive. You can’t really replicate that or cheat with generative AI. It's much, much easier to cheat in other ways using generative AI. So I think we'll get back to that and will see a big surge in behavioral biometrics where you are looking at the individual patterns of people if you know these customers and you've interacted with them before. That just gives you a much richer data set to work off of. So, while I think we are going to see a surge in AI-assisted fraud, there are ways to combat that.

Keep Your Business Goals Within Sight

Need your business to run better and more efficiently? Reach out at riskadvisors@equifax.com.  

We hope you will join us for our June 2025 Market Pulse webinar taking place on Thursday, June 26, 2025, where our Equifax Advisory panel, joined by economist Dr. Amy Crews Cutts, will discuss how you can cut through the information overload to find some certainty amidst economic uncertainty. To ask questions in real time and gain deeper insights before anyone else, you must be there. Don’t miss it!

Interested in learning how personalized insights can drive your business forward? Sign up for a complimentary 60-minute Market Pulse Advisory with one of our experts. 

Find our monthly Small Business Insights, National Consumer Credit Trends reports, the Market Pulse podcast, and more at our Market Pulse hub. 

Broaden your perspective with insights that inspire bold innovation, confident adaptation, and decisive leadership through our Trends and Insights resources. 

Finally, connect with us on YouTube and LinkedIn for even deeper insights.