How Machine Learning Decides Your Business Loan’s Fate in Milliseconds
As someone who’s spent years in the credit risk space, I’ve watched the transformation of small business lending from a relationship-driven process to one where machines make split-second decisions about creditworthiness. But here’s what most business owners don’t realize: if you received a quick rejection, chances are your application never made it to the credit decision process.
See Lenders or alternate lenders in this case have to pick and choose resources to deploy. There’s a metric we refer to specifically – cost to serve (CtS) – which determines the total costs from the lead generation page, marketing and all the human touch points involved to process an application. Lenders look to boost profit margins by optimizing the cost it takes to serve a loan. Any spare change goes directly into the company’s bottom line.
Traditionally, banks were the only option for small businesses to access capital. And it made sense – in a way. Assuming you have a business or personal account with them, banks have your cash flow information, they know the owners, and they have the most data to make smart credit decisions. However, there’s another side to the story there. See, we live in an era where
There’s a delicate balance at play here. And that’s where automated scoring and ML-driven declines come in. To streamline and serve a targeted customer base, lending institutions employ ‘hard cuts’ – the industry’s equivalent of a nightclub’s velvet rope. When you apply for a loan as a fresh applicant, the lender will typically collect information about you, your business, cash flow etc. A lot of these early evaluations rely on self-reported data. The system then generates a list of data points to see if you’re application should even make it to the credit decisioning engine. And that’s where the dreaded hard cuts come in and I’m here to tell you how they work.
For reference, I led the decision science product team at Ondeck Capital and have some background determining and implementing these thresholds – and I want to explain why they’re in palace and how they work. While modeling a typical portfolio, we’d dig into early signals for creditworthiness based on characteristics of bad loans we’ve given out in the past to determine what our absolute minimum thresholds ought to be. It makes sense, both in practice and in theory.
This data-driven approach serves two crucial purposes. First, it helps protect the lender from high-risk loans that are statistically more likely to default. Second, it streamlines the lending process by quickly filtering out applications that don’t meet basic risk management criteria, allowing more resources to be devoted to promising candidates.
Processing a small business loan application is expensive. Traditional banks spend thousands of dollars evaluating each application, conducting due diligence, and making credit decisions. With margins already tight in small business lending, banks need to ensure they’re spending these resources on applications they’re likely to approve.
Too low a credit score? Rejected. Business less than a year old? Rejected. Operating in a high-risk industry? Rejected. All this happens before anyone even looks at your financial statements or business plan. And it’s all done through modular ML modules for each characteristic. It’s not like you need multiple hard cuts to go against you. Your application gets rejected even if miss one criterion. This might sound harsh, but it’s transforming the economics of small business lending.
Lenders therefore use this mechanism to determine who makes it through. Sometimes though, it’s not just a hard and fast rule. In many cases sophisticated machines determine thresholds dynamicially through classification models and weighted average scores. The level and nuance of a hard cut relies on the lender and their risk appetite.
For business owners, understanding this new reality is crucial. The days of walking into your local bank branch and persuading the manager to take a chance on your business are largely over. Instead, success in securing business credit now requires meeting specific, predetermined criteria before a human ever sees your application.
There’s an opportunity here though that I should mention. For instance, one lender’s hard cuts may not apply to another lender’s credit policy. If Lending Company A has a hard requirement for 2 years in business but Lending Company B only requires 1, there’s an opportunity to leverage this difference to attract more applicants. Businesses that may struggle to meet the stricter criteria of Company A could find a welcoming alternative with Company B. This type of cross-selling and lead generation benefits all in this case. Lender A might charge a referral fee, Lender B gets a potential customer, and the applicant gets an opportunity to access capital.
The next time you apply for a business loan, remember: your first challenge isn’t convincing a banker to believe in your business. It’s getting past the automated gatekeepers that decide whether your application deserves a deeper look. In today’s lending landscape, that might be the most important hurdle of all.
