AI credit scoring is widening microloan access at CDFIs

12 November 2026

The Community Development Financial Institutions Fund awarded $1.73 billion in grants to 604 CDFIs in fiscal 2024, according to its annual report. Yet the Government Accountability Office found that many of the smallest lenders in that network still lack the technology to take an application online or underwrite a loan efficiently. That gap is where a new generation of AI-assisted underwriting tools is being tested, with mission-based lenders trying to approve more small-business borrowers without loosening the standards that keep their loan funds intact.

Why CDFIs are turning to AI and alternative credit data

The GAO's review of CDFI Fund programs drew on the Fund's FY 2023 survey of certified institutions. It found that many small CDFIs and minority depository institutions, or MDIs, did not have the technology needed to provide online services or to underwrite loans efficiently. When the agency asked why, the most common answers were cost and staffing. Technology is expensive, and a lender with a handful of employees has little spare capacity to evaluate vendors, migrate data and train people on a new system.

Federal support has grown, and the GAO's highlights show it. The CDFI Fund's technical assistance grants rose from $3.6 million to $25.2 million per year over the last 10 years. Technical assistance money can help institutions pay for the kind of capacity that a core award does not always cover. Even so, the GAO's findings suggest that the need has outrun the funding for many of the smallest institutions.

The pressure is practical. A CDFI that still processes applications on paper, or by re-keying documents into spreadsheets, spends staff hours on work that a software system could handle. Those hours come out of the time loan officers could spend with borrowers who have thin credit files, irregular income or no collateral. That is the typical CDFI client, and it is the reason these lenders exist.

Traditional credit scores often fail the same borrowers. A business owner who has been paid in cash, has little revolving credit or has recently immigrated may look like a poor risk on paper while running a profitable shop or food truck. Alternative credit data and machine-learning models promise a fuller picture. For lenders, the question is whether the tools can deliver that picture at a cost they can afford and in a way that keeps people accountable for the decisions.

SmartLend and other AI underwriting pilots in mission-based lending

One of the most visible examples is SmartLend, built by Accessity in partnership with Mastercard Strive USA. According to Mastercard Strive, the AI-powered tool predicts customer experience outcomes and the likelihood that a borrower will repay or default. The pairing of those two predictions matters. A lender that only scores default risk may approve loans that are technically safe but leave borrowers frustrated or poorly served. A lender that also anticipates how the process will feel to the applicant can adjust before problems arise.

Mastercard Strive reports results from the program. The figures below are the organization's own claims rather than independent evaluations.

• Processing time reduced by 44%, according to Mastercard Strive
• Customer satisfaction reached 92%, according to Mastercard Strive
• Underwriting time could be cut in half, according to Accessity
• Every transaction still requires a human approval or declination, according to Accessity

That last item is the safeguard that regulators and community groups tend to look for first. Accessity told ImpactAlpha that the models it built could cut underwriting time in half while still requiring a person to approve or decline every transaction. In other words, the model recommends and the staff member decides. For a mission-driven lender, that division matters. A score can flag a loan as risky, but a loan officer who knows the neighborhood, the industry and the applicant's circumstances can weigh context that no model has seen.

The efficiency gain also matters because of how CDFIs are staffed. If a model halves the time spent on routine underwriting, the same team can review more applications or spend longer on the difficult ones. Faster turnaround helps borrowers directly. A small-business owner waiting on a decision for equipment or inventory often cannot wait weeks, and some turn to high-cost online lenders when a mission lender is too slow.

Why alternative scoring can widen access for minority-owned businesses

The lending logic behind alternative scoring starts with a data problem. Accessity's historical transaction dataset surpassed 11,000 transactions, as the company described to ImpactAlpha. After cleaning, it was reduced to more than 3,000 final transactions for model building. That is a modest sample by the standards of large banks, which train models on millions of records. It is a realistic picture of what a community lender has to work with, and it shows how much of a lender's history can be lost to inconsistent record-keeping before any modeling begins.

The second piece of context is performance. Accessity says its historical gross loss rate was 4.5% as of 2023. A low loss rate cuts two ways. It shows that a lender serving borrowers outside the mainstream can manage credit risk. It also means a new model has to beat a record that is already reasonable, and it has to do so without shutting out the people the lender was created to reach. A model that improved the loss rate by quietly turning away applicants with unconventional profiles would defeat the purpose.

This is why the design of the inputs matters for minority-owned businesses and other borrowers who have historically had difficulty with conventional credit. Standard scores lean heavily on credit-card history, mortgage payments and other formal records. Alternative approaches can look at the business's own transaction history, repayment patterns on earlier loans from the same lender, and other signs of stability. Whether a given model does that fairly depends on the data and on testing, which is one reason human review remains in the process.

Federal policymakers are paying attention. The Treasury Department's small-business financing policy paper said the working group could discuss and harmonize supervisory policy on artificial intelligence and novel underwriting mechanisms. That is a signal of intent rather than a rule, and it does not set standards for any particular lender. Still, it suggests that regulators expect these tools to spread and want consistent expectations across agencies. For CDFIs, which answer to several overseers and funders at once, consistent supervision would reduce the uncertainty that can make a small institution hesitate to adopt new technology.

Navigating online application portals and digital verification tools

For borrowers, the change shows up first in the application itself. A lender with a working online portal lets a business owner start an application from a phone, upload documents instead of dropping them off, and see where the file stands. That is exactly the capability the GAO found many small CDFIs and MDIs lacked. Where a lender has not built it, the borrower may still be asked to print, sign and mail forms, which is slower and easier to get wrong.

Preparation still makes the biggest difference to how quickly a file moves. Applicants should gather recent bank statements, tax returns, business registration records and a clear list of what the loan will pay for before they begin. Scanned documents that are cropped, blurry or missing pages are among the most common reasons a digital application stalls. A clean upload gives any automated review, and the staff member who follows it, a complete file to work from.

Digital verification tools check identity and income documents electronically rather than by hand, and some systems return a preliminary indication before full underwriting. Borrowers should treat a preliminary result as a first step. As the SmartLend design illustrates, a person approves or declines each transaction, so an early positive signal is not a commitment to lend. Applicants should ask what happens next and what documents remain outstanding.

Certification and pricing are worth checking before applying. To verify regional certification status and review specific microloan interest rate caps, small business owners can visit the website of their local community development financial institution. Rates and terms differ by lender and by program, and the institution's own published materials are the most reliable place to confirm them. A borrower comparing two CDFIs should check the total cost of each loan, including fees, and not look only at the headline rate.

Some applicants will not be comfortable with an entirely online process, and a good lender should not require it. The GAO's findings are a reminder that digital capacity varies widely across the sector, and a CDFI that has invested in technology is not automatically a better partner than one that has not. The better question is whether a lender explains its decisions, offers a human contact and moves at a pace that fits the borrower's needs. Business owners who ask those questions before they apply will be in the best position to use whichever tools their lender has.