Latest / Fintech Conversations with Fexingo: Banking, Payments, and Financial Software Companies / How Fintech Is Using Behavioral Data for Lending Decisions
Transcript
- Lucas: So there's this question that's been hovering around fintech lending for years: what do you do with someone who doesn't have a credit score? Maybe they're young, maybe they're an immigrant, maybe they just never borrowed before. Luna: Right — the thin-file problem. About 26 million Americans are credit invisible, and another 19 million have scores but not enough history to generate a reliable rating. Lucas: Exactly. And the traditional answer has been: either you deny them, or you charge a high rate to compensate for uncertainty. But a growing number of fintech lenders are trying something different — they're looking at behavioral data. Luna: Behavioral data meaning… how you interact with the application itself? Not your credit history? Lucas: Exactly that. How fast you fill out the form, do you pause at the terms and conditions, do you scroll back and forth, how you type your name — all of that becomes a signal. The logic is that these micro-behaviors correlate with conscientiousness and financial responsibility. Luna: It sounds a bit like a digital version of a loan officer reading body language across the desk. Lucas: That's actually a perfect analogy. And the most prominent company doing this at scale is Upstart. They went public in 2020, and their model now processes billions of dollars in loans each year using over 1,500 data points — including education, field of study, employment history, even the time of day you apply. Luna: And they claim lower default rates than traditional models? Lucas: They do. Upstart says their model reduces losses by about 75% compared to traditional credit scoring, while approving more borrowers. But there's an even more extreme example: a company called LenddoEFL, which started in emerging markets. Luna: LenddoEFL — I've read about them. They used to analyze your Facebook friends and smartphone usage patterns to determine creditworthiness. Lucas: They still do, though they've shifted more toward psychometric testing now. But the core idea is the same: if you don't have a financial footprint, maybe your digital footprint is just as predictive. In the Philippines and Kenya, they've been able to approve loans for people who'd otherwise be completely shut out. Luna: Let's talk about the numbers. Is there any independent validation that this actually works better? Lucas: There is, and it's surprisingly recent. In May 2026 — literally this month — the Federal Reserve published a study comparing alternative data models against traditional fico based underwriting. Luna: What did they find? Lucas: The headline: alternative data models approved 27% more borrowers than traditional models, and those borrowers defaulted at roughly the same rate — not higher. So you're expanding access without taking on more risk. That's a pretty powerful finding. Luna: But that's across all alternative data, not specifically behavioral. Did the Fed look at behavioral data alone? Lucas: They didn't isolate it, but behavioral data is a subset within those models. And there's another study from the Consumer Financial Protection Bureau, from 2024, that specifically looked at cash-flow data — which is related — and found that it predicted defaults better than credit scores for subprime borrowers. Luna: So the promise is real. But I've got to ask the obvious question: bias. If you're training a model on behavioral data, are you encoding racial or socioeconomic bias into the algorithm? Lucas: That's the big risk, and it's not hypothetical. A 2023 study from the University of California found that certain behavioral signals — like typing speed or correction frequency — correlate with language background and education level. If your training data is biased, your model will be too. Luna: And it's even harder to audit because these models are often black boxes. You can't just look at a coefficient and say 'this variable is discriminatory'. Lucas: Right. The algorithm might latch onto a seemingly neutral proxy — like the device you use — that actually correlates with race or income. Upstart has been investigated by the CFPB and the DOJ over fair lending concerns. They settled with the DOJ in 2024, agreeing to pay $3.9 million and adjust their model. Luna: So the regulators are paying attention. What's the current state of play? Are banks adopting this, or is it still mostly fintechs? Lucas: It's moving mainstream. JPMorgan filed a patent in 2025 for a behavioral scoring system that uses mouse movements and keystroke dynamics. Wells Fargo is piloting a similar tool for small business loans. And the big credit bureaus are getting involved — Experian now offers a 'consumer attributes' product that includes behavioral signals. Luna: So it's not just fringe fintechs anymore. This could reshape underwriting entirely. Lucas: That's the trajectory. And the interesting thing is, it's not just about thin-file borrowers. Even people with good credit scores might be scored differently based on behavior. Imagine your credit card company adjusting your limit in real time based on how you interacted with their app. Luna: That sounds both brilliant and terrifying. Lucas: It's a trade-off. More access, fairer pricing potentially, but also more surveillance. And there's a privacy dimension we haven't touched — do people know their typing patterns are being evaluated for a loan? Luna: Under current regulations, probably not. Most disclosures are still focused on traditional credit data. Behavioral data usually falls into a gray area. Lucas: That's the regulatory frontier. The CFPB is expected to issue new guidance later this year on alternative data and algorithmic fairness. It could require lenders to explain which behavioral signals they used and how they were weighted. Luna: That would force a lot of transparency. And maybe that's a good thing — if the models are truly better, they should be able to withstand scrutiny. Lucas: I think that's the right framing. The data is powerful, but it needs guardrails. And that's actually a good segue to something I've been meaning to mention. Luna: Oh? Lucas: We talk a lot on this show about how fintech is reshaping access and fairness — and that's only possible because we keep the show ad-free. No sponsors, no pressure to sell you anything. Luna: Right. And that means listener support genuinely makes a difference. Even a couple of dollars a month helps cover hosting, research, and production — it's what keeps these conversations going. Lucas: If you've gotten something out of this episode or any of the others, you can chip in at buy me a coffee dot com slash fexingo. Seriously, every bit helps, and there's no pressure — we're just grateful you listen. Luna: And it keeps us independent. So, whether you do or not, thanks for being part of this community. Lucas: Alright, back to behavioral data. So where does this go next? I think the real frontier is combining behavioral data with cash-flow data — looking at your actual spending patterns alongside how you interact with the application. Luna: So a holistic view of the borrower, not just one snapshot. Lucas: Exactly. The startup Nova Credit is already doing something like that for immigrants — they pull credit bureau data from your home country and combine it with U.S. cash-flow data. The approval rates for that group are dramatically higher than traditional models. Luna: And that's a huge market. There are about 45 million immigrants in the U.S., many of whom have excellent credit history in their home country but zero in the U.S. Lucas: Right. And if behavioral data can help bridge that gap, you're not just expanding credit — you're enabling economic mobility. That's the bigger story here. Luna: It's a good reminder that underwriting isn't just about risk — it's about access. Who gets to participate in the financial system. Lucas: Exactly. And that's the conversation we'll keep having. For now, I think the key takeaway is: behavioral data is real, it's being used, and it's going to be regulated. How that balance plays out will define the next five years of consumer lending. Luna: A good note to end on. Thanks for listening.