US banks that advertised more AI-related jobs also held a smaller share of loans to small and medium-sized businesses, according to new research from the Federal Reserve Bank of San Francisco. The result is an early association, not evidence that AI is causing individual firms to be refused credit.
The researchers used AI-related skills in online job advertisements as a proxy for how intensively a bank was adopting the technology. They found that banks above the sample average on that measure had small-business and farm loans making up 12% of lending, compared with 21% at banks below the average.
The gap was also present within each size group. Among small banks, the equivalent shares were 13.3% and 21.4%.
Why the type of information matters
Credit decisions draw on more than a score or a set of accounts. A bank may assess financial statements, tax records and repayment history. This is often called hard information because it can be standardised and processed at scale.
Smaller firms may also be assessed through softer information: a lender’s understanding of the owner, local market and trading history. Relationship managers can build that picture over time, but it is harder to reduce to a uniform data field.
The San Francisco Fed authors say current AI tools may be especially useful for processing the first category. That could make lending based on formal records cheaper or quicker to assess, while leaving relationship-heavy lending less attractive. It is a possible explanation for the pattern, rather than a demonstrated mechanism.
The distinction matters because a small business is not always small in an everyday sense. It is a legal classification that varies by industry and programme, as our guide to the SBA’s proposed small-business size standards explains.
Adoption is concentrated at large banks
The study matched job-posting information with regulatory Call Report data for 1,006 banks representing more than 87% of US banking-system assets. By the end of 2025, AI-related vacancies accounted for 8.86% of job postings at large banks, 4.48% at medium-sized banks and 1.15% at small banks.
The authors classify a bank with assets below $10 billion as small and one above $100 billion as large. They report that 84% of small banks in their sample had never posted an AI-related vacancy.
That measure has an important limitation. A smaller lender may use an outside software provider without advertising for its own AI specialists. The job-posting data mainly show in-house activity, so they may overstate the gap in technology use between large and small institutions.
A signal to watch, not a lending forecast
High-AI banks in the sample also had average returns on assets about 0.38 percentage points higher than lower-AI banks and slightly higher shares of problem loans. Neither pattern establishes that AI created the difference. Banks with more complex loan books may have stronger reasons to invest in technology in the first place.
The same caution applies to small-business lending. The paper does not show that a particular applicant was declined because an algorithm replaced a relationship manager. It shows a bank-level relationship from 2021 through 2025 between the proxy for AI adoption and the share of lending directed at smaller firms.
That leaves several possible outcomes. AI could eventually help lenders handle unstructured information such as business documents and customer histories. It could also help smaller banks deliver faster service if vendor tools become more accessible. Or it could widen the advantage of large institutions with more data, capital and specialist staff.
For owners, the immediate practical lesson is not that bank finance is disappearing. It is that the information a business can document clearly may become more important as credit systems change. Clean accounts, current cash-flow forecasts and a clear explanation of how a loan will be repaid remain useful whether a lender relies mainly on people, models or both.
The evidence comes from the Federal Reserve Bank of San Francisco’s September 2026 Economic Letter, by Greeshma Avaradi, Naomi Halbersleben, Zheng Liu and Mark Spiegel.
