Latest / Fintech Conversations with Fexingo: Banking, Payments, and Financial Software Companies / How Fintechs Use AI for Real-Time Lending Decisions
Transcript
- Lucas: If these conversations are useful for what you're building or running, you're in the right place. I'm Lucas. Luna: And I'm Luna. Today we're talking about something that affects millions of small businesses: how fintechs are using AI to make lending decisions in real time. Lucas: Specifically, we're looking at the shift from traditional underwriting — which can take weeks and relies heavily on personal credit scores — to models that analyze live business data and approve loans in minutes. Luna: And a perfect example is Kabbage, which American Express acquired back in 2020. They were one of the first to do this at scale. Lucas: Right. Kabbage built a platform that connects directly to a small business's sales channels, accounting software, and even shipping data. So instead of asking for bank statements from three months ago, they're looking at what's happening today. Luna: That's a huge departure from the old model. A traditional bank might pull a credit report and ask for tax returns. Kabbage is looking at live transaction volume. Lucas: Exactly. And the AI models they use are trained to spot patterns — seasonal dips, rapid growth, payment consistency. They can underwrite a loan in about seven minutes, on average. Luna: Seven minutes versus two weeks. That's a massive efficiency gain. But does it come at a cost? I mean, are these models less accurate because they're faster? Lucas: That's the big question. And the evidence is mixed. During the pandemic, Kabbage processed over a billion dollars in PPP loans, and their default rates were actually lower than the industry average for that program. Luna: Interesting. So in that crisis, the real-time data was actually better at predicting who could repay. Lucas: In part because they could see who was still generating revenue. A business that had shifted to online sales showed up in their data. A traditional bank might have just seen a closed storefront and denied the loan. Luna: That's a powerful example. But I want to push back a little. What about bias? If the AI is trained on historical data, and that data reflects existing inequalities in lending, doesn't real-time just replicate the same problems faster? Lucas: It's a real risk. And fintechs have faced regulatory scrutiny for exactly that. The Consumer Financial Protection Bureau has been looking at whether alternative data — like utility payments or rent — helps or hurts underserved groups. Luna: So the technology is neutral, but the data isn't. Lucas: Exactly. The key is how you design the model. Some fintechs are now using explainable AI — so they can tell you why a loan was denied, which helps with fairness and also with regulatory compliance. Luna: Let's talk about the technical side. How are they actually getting this data in real time? It's not like a bank sends a wire every time a sale happens. Lucas: They use APIs — application programming interfaces — to connect to platforms like Shopify, Stripe, QuickBooks, and even Amazon seller accounts. The business gives permission, and the fintech pulls the data directly. Luna: So it's a permission-based data stream. That's a form of open banking, even if we don't call it that in the US. Lucas: Exactly. And because the data is live, the model can adjust the credit limit dynamically. If a business has a huge sales week, the system might automatically offer a larger line of credit. Luna: That's the opposite of traditional banking, where your credit limit might only change once a year, after a review. Lucas: Right. And this has opened up lending to a lot of businesses that wouldn't qualify for a bank loan — like newer companies without a long credit history, or seasonal businesses that look risky on paper. Luna: But what about the downside? If a business has a bad week, does the AI cut their credit line automatically? That could cause a cash flow crisis. Lucas: That's a real concern. Some fintechs have been criticized for reducing credit lines without warning, based on a temporary dip. The best practices now include giving businesses a notice period or using longer trend data, not just a single day. Luna: So the human element still matters. Even with AI, you need to design the system with empathy. Lucas: Absolutely. And that's where regulation is starting to catch up. The Fed and the OCC have issued guidance on responsible use of alternative data in lending. Luna: Let's talk about another player. There's a company called Fundbox that uses AI to predict cash flow gaps and offer lines of credit. They claim their models can predict defaults better than FICO scores alone. Lucas: Right. And they're not alone. OnDeck, BlueVine, LendingClub — they all have some form of ai driven underwriting. The competitive advantage comes from how fast and how accurately you can assess risk. Luna: And the speed is a feature, not just a convenience. For a small business owner, waiting two weeks for a loan might mean missing an inventory opportunity or not making payroll. Lucas: Exactly. Speed is a form of capital. And that's why the incumbents — the big banks — are playing catch-up. JPMorgan has been investing in real-time lending through their Chase Business Banking platform. Luna: They're trying to combine their massive data with AI models. But they have legacy systems to deal with. Lucas: Right. It's harder to move fast when you have mainframes and decades of processes. That's the fintech advantage — they started with the cloud and APIs from day one. Luna: One thing I want to touch on: the economics. These fintechs often lend their own capital, or they partner with banks. How does the AI affect their loss rates? Lucas: Studies show that fintech lenders using AI have charge-off rates that are roughly comparable to traditional banks, sometimes slightly lower. But they're lending to riskier borrowers, so that's actually a win. Luna: So they're taking on more risk but managing it better with data. That's the promise. Lucas: And they can do it at lower cost because the process is automated. No loan officers reviewing paper applications. No manual data entry. Luna: If today was actually useful to you — and I know Lucas and I love digging into these details — the way these episodes stay ad-free is through listener support. You can help keep the show going at buy me a coffee dot com slash fexingo. Lucas: Yeah, it really does make a difference. We don't run commercials or take sponsor money, so it's a direct way to say this content matters to you. Luna: And it helps us keep bringing you conversations like this one. So, back to real-time lending — where do you see this going in the next couple of years? Lucas: I think we'll see more embedded lending — where the loan offer appears right inside the software a business already uses. Like, you're in QuickBooks, and it says 'you have a $50,000 credit line available, fund in one click.' Luna: That's already happening with Shopify Capital and Amazon Lending. They lend to merchants based on their sales data on the platform. Lucas: And those are incredibly efficient because the platform owns the data. They know exactly how much you sell, your return rate, your customer reviews. That's a rich dataset. Luna: So the future of small business lending is really about data integration. The more seamlessly a lender can plug into a business's operations, the better the lending decision. Lucas: And the faster the funding. Which, for a small business, can be the difference between growing and stalling.