Latest / Fintech Conversations with Fexingo: Banking, Payments, and Financial Software Companies / How AI Agents Are Rewiring Merchant Underwriting
Transcript
- Lucas: If these conversations are useful for what you're building or running, we appreciate you tuning in. But let's get straight into the mechanics of modern lending. The old model of merchant underwriting was built on a lag: you look at last year's tax returns, check your credit score, and hope the business hasn't changed too much in between. Luna: Right, and that gap is where most small businesses die. They need cash now, but the bank is still looking at a snapshot from six months ago. Lucas: Exactly. So today I want to look at how AI agents are rewiring this process by shifting from static historical data to real-time behavioral signals. We're talking about a specific case study involving a mid-sized payment processor that completely overhauled its underwriting stack last quarter. They didn't just add a chatbot; they deployed autonomous agents that monitor transaction velocity, refund rates, and even customer sentiment in reviews to assess risk every single hour. Luna: That sounds like a massive leap from just checking if a deposit hit the account. How did their default rates change when they made those decisions? Lucas: This is the part that caught my eye. When they switched from a three-week manual review cycle to this agent-driven system, their default rates dropped by twenty-two percent. At the same time, they approved thirty-five percent more loan applications because the agents could see liquidity patterns that a human analyst would miss until it was too late. Luna: So the agents are actually seeing risk earlier than the humans can, which lets them say yes to people who look risky on paper but are healthy in practice. Lucas: Precisely. Let's break down the architecture of these agents. Traditional models rely on fixed inputs: revenue, debt to income ratio, time in business. These new agents treat the merchant's digital footprint as a living organism. They track things like the speed of inventory turnover relative to supplier payments, or sudden spikes in chargeback attempts that might indicate fraud rather than just poor creditworthiness. Luna: It feels like we're moving from judging a person's character based on their resume to watching how they handle their daily tasks in real time. Lucas: That's a really good analogy. And it changes the product offering entirely. Instead of a lump sum loan with a fixed term, some of these processors are starting to offer dynamic credit lines that expand or contract automatically based on the agent's assessment of the next forty-eight hours of sales volume. Luna: Dynamic credit lines? That sounds incredibly efficient for seasonal businesses, but also terrifyingly opaque for the merchant. Lucas: It is opaque, and that's the friction point. In our case study, the merchants using this new system had no idea how the agent arrived at their credit limit. It wasn't a simple formula. It was a neural network weighing hundreds of micro-signals. If a merchant wanted to know why their line was reduced, the agent couldn't give them a human-readable reason, just a confidence score. Luna: So you're getting better capital access, but you lose all visibility into the logic. Is that a fair trade-off for a small business owner? Lucas: For many, yes. Because the alternative is being denied entirely. But there's a structural risk here. If the agent misinterprets a temporary glitch as a trend, it can cut off credit at the worst possible moment. We saw one example where a retailer's agent flagged a surge in orders as potential fraud, slashed their line, and caused a stockout that actually hurt their revenue. Luna: Which creates a self-fulfilling prophecy. The agent sees lower revenue and cuts credit further, making the problem worse. Does the system have a circuit breaker for that? Lucas: The newer versions do, but they're not universal. The key insight is that these agents are learning from the market faster than any human underwriter ever could. They're effectively creating a parallel credit system that operates on milliseconds rather than months. Luna: It makes me wonder if this will eventually make traditional bank loans obsolete for small merchants, since banks can't move that fast. Lucas: I think that's exactly where we're heading. Banks are trying to retrofit these tools onto legacy mainframes, which is like putting a Ferrari engine in a horse carriage. The data isn't flowing in real time, so the underwriting stays slow. Luna: But the payment processors already own the data stream. They don't need to ask for permission to see the transactions; they're sitting on the pipe. Lucas: They do. And that control over the data is what allows them to offer these embedded financial products. You're not applying for a loan separately; the credit is woven into the checkout experience, approved by an agent before the customer even finishes clicking pay. Luna: That's seamless, but it also means the merchant has less bargaining power. They're locked into the processor's ecosystem for both payments and capital. Lucas: Totally. It's a walled garden. But from an efficiency standpoint, the reduction in friction is undeniable. Let's talk about the economic impact. This kind of underwriting is likely to increase overall credit availability in the small business sector, especially for those who are thin-filed or lack traditional collateral. Luna: Because the agent doesn't care about your house or your car. It only cares about whether this specific store is going to sell enough inventory next week to pay back the advance. Lucas: Exactly. It's pure cash-flow lending, powered by pure data. And as these agents get better at predicting consumer behavior, they'll get even better at predicting merchant failure. It's a feedback loop that benefits the lender significantly more than the borrower, unless the borrower learns to game the signals. Luna: Gaming the signals? Like artificially inflating transaction volumes to trick the agent? Lucas: Or creating artificial demand spikes. Some sophisticated operators are already doing this, though the agents are getting smarter at detecting synthetic activity. It's becoming an arms race between the underwriters and the borrowers. Luna: That sounds exhausting. I'd rather just deal with a boring bank officer who asks for my tax return and goes home at five. Lucas: And you can still choose that path, but the market is clearly pricing in a preference for speed and accessibility. The question is whether the long-term stability of these algorithmic loans holds up during a broader economic downturn. Luna: When everyone defaults at once, will the agents know how to adjust, or will they just keep chasing the same patterns? Lucas: That's the million-dollar question. Current models are trained on past cycles, which were relatively stable. A systemic shock might expose blind spots that human underwriters would catch instinctively. For now, though, the efficiency gains are too great to ignore. Luna: So we're trading certainty for agility. I can see why merchants are signing up. Lucas: Agility wins in fast-moving markets. But let's pause for a second to talk about how we keep this show going. A handful of listeners chip in monthly through buy me a coffee dot com slash fexingo — that's F. E. X. I. N. G. O. — and that's literally what funds making this many of these episodes possible without ads. Luna: Yeah, it really helps us dig into these niche topics without needing corporate sponsors telling us what to cover. Lucas: Back to the tech. One final thought on the future of this space. As these agents become more capable, they might start offering not just credit, but advisory services. Telling the merchant how to price items or manage inventory based on the same data used for underwriting. Luna: That would be the ultimate integration. You're not just funding the business; you're actively managing it in real time. Lucas: It's a fascinating evolution. From passive lender to active partner, driven entirely by code. What do you think is the biggest barrier to adoption for the average small business owner right now? Luna: Trust. Until they understand how the black box works, they're going to be hesitant to hand over full control of their finances to an algorithm. Lucas: Fair point. But trust often follows utility. Once they see the capital flowing faster than any bank could provide, the skepticism tends to fade. Luna: Let's hope the agents are generous enough with that capital to keep the small business ecosystem healthy through the transition. Lucas: Here's to hoping. We'll see you next time.