Latest / Fintech Conversations with Fexingo: Banking, Payments, and Financial Software Companies / How AI Agents Are Rewiring Merchant Underwriting
Transcript
- Lucas: The old model of merchant underwriting was built on silence. You submitted your tax returns, maybe three months of bank statements, and then you waited. The lender looked backward, at what had already happened, and decided if you were worthy of capital based on history. Luna: Which makes sense when you think about it, but it also means that if your business is moving fast, you’re being judged by a rear-view mirror. Lucas: Exactly. And today, that rear-view mirror is shattering. We are seeing the rise of AI agents that don’t just review data; they ingest it continuously. These agents are connected directly to payment processors, inventory systems, and even point of sale terminals, creating a live pulse of the business rather than a snapshot. Luna: So instead of asking for a PDF of last quarter’s sales, the system is watching the transactions happen as they occur? Lucas: Precisely. Let’s look at a concrete example: a mid-market apparel retailer preparing for the holiday season. Under the old model, they’d apply for a line of credit in September, hoping the approval came before their peak buying window. If the underwriter was slow or conservative, they might miss out on critical inventory. Luna: And now, with an AI agent monitoring their daily sales velocity, can the credit limit adjust automatically? Lucas: It can. As early October sales tick up above baseline, the agent detects the positive trend. It doesn’t wait for a human analyst to review a report. It pre-approves a temporary increase in the credit line, allowing the merchant to purchase more stock immediately. The risk is mitigated because the agent sees the incoming cash flow in real time. Luna: That sounds incredibly efficient, but it also feels like a massive amount of trust is being placed in an algorithm that never sleeps. Lucas: Trust is the right word, but it’s not blind trust. It’s computational trust. The agent is constantly stress-testing the scenario. If sales drop unexpectedly in November, the agent can freeze additional draws or require immediate repayment of a portion of the line. It’s dynamic, whereas traditional banking is static. Luna: But does this create a feedback loop where only businesses that already look good get better terms? What about the outlier? Lucas: That is the central tension here. In the past, a human underwriter might say, 'This business has no history, but the owner is passionate and the location is prime.' An AI agent might see no data and shut the door. However, newer models are beginning to incorporate alternative data points—foot traffic patterns, social media sentiment, even local economic indicators—to fill those gaps. Luna: So it’s not just about sales data anymore; it’s about building a holistic profile of the business health from multiple sources. Lucas: Right. And this is where the concept of the 'agent' becomes distinct from simple automation. Automation follows rules. An agent reasons. It can negotiate. For instance, if a merchant needs a larger loan than the default limit, the agent might initiate a conversation, asking for specific clarifications or offering tiered pricing based on collateral. Luna: Negotiating? That’s a new word in underwriting. Usually, it’s just take it or leave it. Lucas: It is. And for small business owners who are often intimidated by traditional banks, this changes the power dynamic. They aren’t talking to a loan officer who might be having a bad day. They are interacting with a system designed to find the optimal fit between their needs and the lender’s risk appetite. Luna: I’m curious about the data privacy side of this though. If these agents are pulling from POS systems, inventory software, and bank accounts, who owns that data stream? Lucas: Ownership is messy. Typically, the merchant grants permission through API integrations, but the data often ends up stored by the fintech provider. This creates a valuable asset for them, but it also raises questions about secondary use. Can that data be sold? Can it be used to market other products? Luna: It feels like we are trading privacy for speed. Is that a fair trade-off for most businesses? Lucas: For many, yes. The cost of capital is still the biggest barrier to growth for small businesses. If an AI agent can reduce the time to funding from weeks to minutes, and lower the interest rate by accurately assessing real-time risk, the value proposition is strong. Luna: But what happens when the agent makes a mistake? When it misreads a signal? Lucas: That’s the danger of scale. A human error affects one loan. An algorithmic error could affect thousands simultaneously. We saw glimpses of this during the pandemic when automated fraud filters blocked legitimate transactions en masse. The correction process was slow and painful. Luna: So there needs to be a human-in-the-loop for significant anomalies, even if the day-to-day is automated. Lucas: Ideally, yes. But the economics of fintech drive toward full automation. The margin on a small business loan is thin. Adding human oversight increases the cost structure, which ultimately gets passed on to the borrower in higher rates. Luna: It’s a classic efficiency versus safety trade-off. And I wonder if regulators are keeping pace with this. Lucas: They are trying, but it’s hard to regulate something that evolves every week. Current frameworks are built around periodic reporting and annual audits. Real-time agent decision-making operates on a microsecond timeline. The regulatory gap is widening. Luna: It seems like the definition of 'creditworthiness' is being rewritten in code rather than in policy documents. Lucas: Absolutely. And for the entrepreneur, the skill set is changing too. It’s not just about having a good business plan; it’s about maintaining a clean, transparent digital footprint. Your data hygiene becomes your credit score. Luna: That’s a profound shift. You have to manage your digital transparency as actively as you manage your inventory. Lucas: Let’s pause for a moment to talk about the people making these conversations possible. If these insights are helping you build or run your business, consider supporting this show. It’s ad-free because listeners like you choose to back us directly. Visit buy me a coffee dot com slash fexingo — that's F. E. X. I. N. G. O. Luna: Yeah, and it really helps keep the research deep and the ads out, so thanks for considering it. Lucas: Back to the data hygiene point. It’s interesting to think about how this might change how entrepreneurs view their own operations. You can’t just ignore your backend systems anymore. Luna: It forces a level of discipline that most small businesses haven’t had to adopt until now. Lucas: Exactly. And this brings us to another layer: the interoperability problem. Not all fintech agents talk to each other. A merchant might have one agent for payments, another for lending, and another for payroll. These silos prevent a truly holistic view. Luna: So we’re seeing the next wave of innovation in open APIs that allow these agents to share context securely? Lucas: Yes. Imagine an agent that knows you just paid your payroll, so it understands you have less liquidity for a new order. Or one that sees a supplier delay, so it adjusts your credit limit accordingly. This requires a standard language for financial data exchange. Luna: It’s like building a nervous system for the economy, where every part knows what the others are doing. Lucas: That’s a vivid way to put it. And the benefits are clear: reduced friction, faster access to capital, and more accurate risk pricing. But the complexity is immense. Luna: Complexity always introduces new risks, though. Especially when you’re dealing with someone else’s money. Lucas: True. And we haven’t even touched on the ethical dimension of predictive analytics. If an agent predicts a business will fail in six months, should it start tightening credit immediately? Or should it offer support? Luna: That’s a tough one. Tightening credit when they need it most could accelerate the failure. It’s a self-fulfilling prophecy. Lucas: Right. So the design of these agents matters immensely. Are they designed purely to minimize loss for the lender, or to optimize the health of the ecosystem? Most current models lean heavily toward the former. Luna: It feels like we need a new framework for 'responsible AI' in lending, one that includes empathy as a variable. Lucas: Empathy is hard to code, but fairness is measurable. And that’s where the industry needs to focus its energy next. Luna: As we move deeper into this decade, I wonder if the best lenders will be the ones that learn to collaborate with their AI agents, rather than just letting them run wild. Lucas: Probably. The future isn’t about replacing humans; it’s about augmenting judgment with precision. But getting the balance right is going to take time, and a lot of trial and error. Luna: One thing is sure: the merchants who adapt to this new reality will have a huge advantage over those stuck in the old model. Lucas: Agreed. And for everyone listening, the question is whether you’re ready to let your data do the talking for you.