AI Moves Beyond Hype to Reshape Lending for Underserved Borrowers

AI is shifting from hype to practice in the workflows of lenders serving thin-file and subprime borrowers. For years, these applicants have struggled to access affordable loans because older scoring systems were built with mortgage lending in mind.

Now, new models and compliance frameworks signal a more inclusive path. The change matters most for nonprime households, where even modest gains in approvals can greatly affect paying bills, avoiding delinquency, and finding affordable credit.

For subprime families in particular, access to installment loans can mean steering clear of payday products or high-cost alternatives that quickly spiral into debt cycles.

Kareem Saleh, founder and CEO of Fairplay, framed the opportunity in terms of both precision and trust. As he told us, sharper predictions from payroll APIs and AI models must be paired with transparency and compliance.

“Accuracy without explainability erodes trust — and in lending, trust is everything,” Saleh said. That balance between performance and fairness is where the current wave of innovation is heading.

For lenders, one of the toughest hurdles to AI adoption has been regulatory acceptance. Dr. Andrada Pacheco, chief data scientist at VantageScore, underscored the need for explainability. She argued that innovation is only sustainable when models remain auditable.

VantageScore 4.0, now accepted by Fannie Mae and Freddie Mac, illustrates that point. The model leverages machine learning at the variable-building stage but avoids opaque decisioning. That distinction allows it to extend coverage to thin-file borrowers without raising red flags with regulators.

SAS echoed that concern. Naeem Siddiqi, a senior risk advisor at the firm, said lenders must ensure their AI algorithms remain explainable, interpretable, and compliant with laws like the Equal Credit Opportunity Act.

He noted that adverse action requirements — mandating clear explanations for denied credit — apply to AI just as they do to legacy scorecards. Transparency, in other words, is not optional.

Another sign of maturity is the arrival of purpose-built standards for AI in finance. Kaustubh Joshi, chief business officer at iBusiness Funding, pointed to the creation of BAICS (Banking AI Controls Standard), the first framework designed specifically for secure AI adoption in banking.

He described BAICS as more than technical guidelines — it is a signal that leading institutions are aligning around compliance and security as foundations for innovation. For lenders wary of regulatory backlash, such frameworks offer a road map for responsible deployment.

The SAS team also stressed the need for simple oversight. Banks already use reports to check whether models remain stable, predict risk well, and avoid unfair bias. Applying those same checks to AI helps lenders adjust as data and risk conditions change. 

Besides compliance, the actual value of AI is that it can handle richer and varied data sources. Scores constructed using mortgages and revolving accounts rarely mimic the full financial profile of underserved borrowers.

That mismatch leads lenders to pad for uncertainty, often with less desirable rates or even rejections as the fallback. In subprime segments, extra padding equals higher APRs and smaller loan sizes — outcomes that limit borrower mobility and lender profitability.

S.P. “Wije” Wijegoonaratna, co-founder and CEO of Aliya Financial Technologies, said, “AI changes this by shifting the focus from historical credit files to actual financial behavior.” By analyzing transaction data, AI can create individualized virtual cashflow statements that measure income, expenses, and volatility.

This mirrors corporate underwriting practices but applies them at the consumer scale. The result: more precise risk assessments, fairer pricing, and expanded access to affordable loans.

Saleh echoed that point, saying, “Payroll data can fill gaps in thin files if paired with privacy safeguards and compliance by design.” The potential is especially strong for nonprime consumers who may have steady incomes but lack the credit history to prove it under older systems.

“Accuracy without explainability erodes trust — and in lending, trust is everything.” — Kareem Saleh, founder and CEO of Fairplay

Terisa Roberts, Global Solution Lead for risk modeling and decisioning at SAS, said, “Automation provides a low-cost option for lenders to assess borrowers who might otherwise be excluded.”

By automating processes that would take loan officers hours, AI can cut costs and lower the barrier to entry for consumers who fall outside traditional scoring models.

The complexity of small business lending makes it another proving ground for AI. Joshi explained, “Small business lending is perfect for agentic AI implementation, where the agents will act like experts for each process variation.”

Joshi added that, unlike consumer finance, which often revolves around single metrics like FICO, small business underwriting involves countless process variations — from collateral types to guarantor structures.

AI agents, he suggested, can act as digital experts for each variation, handling tasks with speed and accuracy. That could give small firms, long underserved by digital innovation, access to faster and more consistent financing without sacrificing compliance.

Roberts reinforced the idea, saying, “AI tools like OCR and large language models can speed up the work of preparing credit memos and reviewing financial statements.” AI systems can even identify missing data and automatically request it from borrowers, acting as a digital assistant to human loan officers.

Together, these tools cut down on the manual drudgery of due diligence and free loan officers to focus on judgment calls.

Looking ahead, the consensus among experts is that AI will continue to redefine how lenders serve subprime and credit-invisible markets. Roberts told us, “automation lowers barriers for applicants who might otherwise be excluded because they lack a credit score.”

Siddiqi added, “Open banking, where applicants allow lenders access to their accounts, will make it easier to prove cash flows and repayment ability in the absence of bureau data.”

As more fintechs experiment with digital footprints and unstructured data, AI systems can identify repayment capacity beyond traditional score inputs. That evolution may get millions of unbanked or underbanked individuals into small loans.

Done responsibly, it may also rebuild trust in an industry where opaque practices have long fueled skepticism.

For lenders, efficiency improvements translate into faster approval and reduced subprime portfolio losses — in the higher-risk segments, better targeting may be the difference between sustainable growth and unmanageable delinquency.

Notably absent from the discussion were Upstart, which declined to comment, and Zest, which did not respond to a request for comment as of our publication date.

AI in lending has already moved beyond buzzwords into practice. Tools like payroll APIs, cash-flow modeling, and agentic AI platforms are opening doors for borrowers long shut out of the system. These advances could make credit fairer and more affordable for thin-file and subprime applicants.

But realizing that promise requires transparency and compliance to stay front and center — so innovation builds trust rather than erodes it.