AI lending opens doors for nontraditional borrowers

By early October 2026, the case for credit-box expansion was no longer theoretical: lenders and vendors were tying AI-led underwriting changes directly to lower operating costs and higher throughput when files included nontraditional income. In Analysis, Becca of Oculus said lenders are no longer seeing only “the kind of clean loan file” and instead must evaluate borrowers with “less traditional means of credit and income,” while Penny described building “a little appraisal tool to do appraisal checks of the photos in an appraisal to look for damage in the home,” which “can go through a list of 20 photos much faster than a human” and improve “productivity and speed” as well as review completeness.

The approval-rate side of the equation came from feeding those models real-time, consented signals that conventional underwriting misses, especially for workers without standard pay stubs or bureau depth. In Explainer, a mini truck driver rejected by a standard application became lendable once the model incorporated platform and transaction data showing “96% on time delivery rate and a 4.7 star rating out of five,” “consistently working six days a week,” and “fast tag and fuel spend patterns will show consistent route distances and fuel purchase” behavior, illustrating why alternative income verification was producing measurable gains in both approvals and processing efficiency.

Traditional scoring fails first because both legacy models and many early AI underwriting tools were built around a narrow borrower archetype and optimized for “clean” W2 borrowers, creating an exclusionary credit box. In Chrisman Commentary, Becca says AI processes are “designed for the borrower that we already know how to underwrite, which is that clean W2, two years at the same job, very clean borrower profile,” even as “the pool of borrowers is changing,” while 32 million U.S. adults remain unscorable by traditional credit, including 7 million with no file and 25 million with thin or stale files.

What changes the decision is shifting from the score to the underlying cash-flow evidence: as Becca argues in Chrisman Commentary, reading a borrower’s bank-statement transaction history and income patterns lets lenders underwrite gig and self-employed workers on what they truly earn, not whether they fit a W2 template. That matters because incomplete bureau reporting can distort risk by “skewing 40 points either way,” and, as HousingWire notes, “income stability or instability” is now central, with 50% of consumers experiencing income volatility in 2022 and 62% in 2025.

The clearest early payoff is operational: one lender in an August 2026 case study said it was “underwriting loans… all loans… in 72 hours full underwrites,” and that this had “decreased significantly” from prior capabilities, with approvals now arriving in about 72 hours for most files. That speedup fits the broader October view that AI reduces the back-and-forth of manual validation by automating checks across fragmented data sources, cutting verification effort and compressing turnaround time rather than merely shifting work elsewhere.

Just as important, the early gains do not yet show a clear risk tradeoff: PYMNTS reported Plaid’s LendScore delivered “a 73% approval rate versus 65% for a traditional benchmark at equivalent risk,” while “at the same approval rate” it showed 41% lower risk. PYMNTS also said Personify expanded bank-data collection to more than 90% of applicants “without a meaningful hit to conversion,” with verified bank-transaction data lifting approvals by up to 8% among underserved segments while maintaining credit costs, and that Affirm’s latest model generated 3.4% more completed purchases at comparable risk.