The loan officer’s desk in the lobby of nearly every community bank in the United States today is essentially the same as it was in 2005. The same forms, the same beige folders, the same wary grin. In less than nine minutes, a fintech is approving a half-million-dollar working capital line for a Texas HVAC company in a converted warehouse with too many monitors and not enough chairs. The contrast is almost dramatic. Additionally, the numbers supporting it are beginning to take a serious turn. Like many of its AI-native competitors, the company in question is simultaneously undercutting traditional business lenders on price, speed, and approval rates while growing by nearly 70% annually.
Anyone who followed the consumer-fintech wave of the 2010s will notice a pattern here. Square reduced the amount of money paid to merchants. Klarna gradually reduced point-of-sale credit. The banks woke up after shrugging for too long. The current state of small and mid-sized business lending is reminiscent of that same film, but with more sophisticated instruments. Unlike traditional FICO-based engines, which rely on about twelve variables, modern machine-learning underwriting models are capable of evaluating thousands of variables within a credit file. Often referred to as the canary, Upstart’s SEC filings reveal 27% more approvals and 75% fewer defaults. It’s really hard to argue with that math.
It’s difficult to ignore the legacy system’s stubbornness. When the American credit score was created in the 1950s, it was a very clever short cut. However, it speeds up aging. Years of good behavior can be overshadowed by a single late payment. A Tuesday’s cash flow doesn’t reveal anything about a Thursday. Lenders have been using a tool designed for a black-and-white television era to make million-dollar decisions, and small businesses—those with seasonal revenue, erratic deposits, and the occasional difficult quarter—paid the highest price. The score isn’t being fixed by the AI-native lenders. They’re avoiding it.
Here, the economics are important, and they are important in a subtle way. According to Accenture’s 2024 research, AI-enabled fintechs will generate 30% more revenue per customer and have about 40% lower customer acquisition costs than businesses that continue to operate manually. The cost of operating these models has skyrocketed due to cloud infrastructure. In just a few weeks, pre-trained foundation models from Anthropic, OpenAI, and the open-source Llama and Mistral can be optimized for credit applications. What needed a department five years ago can now be completed by a small team. During earnings calls, bank executives consistently fail to mention that.
But beneath all of this, there’s a tension. In a report released this week, the Financial Stability Board cautioned that over one-third of private credit transactions now involve AI-related lending, an increase from 17% over the preceding five years. The FSB warned that significant losses might result from a severe correction. It’s possible that these fintechs’ speed advantage also contributes to their vulnerability. If the model drifts and no one notices, less expensive underwriting may turn into reckless underwriting. The report included a warning about the failure of First Brands and Tricolor, both of which were supported by private credit, last year.
We seem to be in the middle innings rather than the end as we watch this play out. Within three years, the percentage of fintechs that invest in AI will rise from 70% to 100%, and the differentiator will shift from whether or not a company uses AI to how cleanly it does so. Banks still exist. They have customer inertia, regulatory protection, and deposits on their side. However, small business owners remember the difference between a nine-minute decision and a nine-day one. Additionally, keeping in mind is how clients move in the lending industry.
