AI Can Stretch Proximity—It Cannot Make It Disappear
U.S. bank small-business lending is where this series asks its final question: can artificial intelligence replace proximity? The evidence supports a two-part answer. AI can replace some of the information work that makes distance costly, and it appears to help banks extend selected loans to distant borrowers with better observed outcomes. But if replacement means that branches, human relationships, local knowledge, monitoring, and flexible judgment no longer matter, the answer is no. Readers can judge future claims about AI lending by asking which function of proximity has changed, what evidence now performs that function, and who bears the risk when the substitute fails.
The apparent contradiction disappears once proximity is treated as a bundle of capabilities rather than a synonym for a nearby building. A branch can be a place to submit an application, but it can also connect a bank to local knowledge and referrals. A meeting can be a social interaction, but it can also reveal information missing from a credit file. A relationship can create trust, but it can also support monitoring and flexibility after the loan is made. AI’s demonstrated contribution concerns one part of this bundle more directly than the others.
The answer depends on what proximity is doing
The strongest case for replacement concerns informational proximity. The Census working paper proposes that AI lowers the cost of extracting facts from messy records, integrating data from disparate sources, and recognizing nonlinear patterns that conventional review may miss. If those functions produce sufficiently reliable hard information, a bank may need less locally gathered soft information to assess a distant borrower. The bank’s decision can travel farther because more of the evidence can travel with it.
This is a real change, not a semantic one. The paper finds that banks with greater AI use lend relatively more to distant small-business borrowers. The result is strongest at the far edge of the measured relationship: when distance is grouped into under 50 miles, 50–250 miles, 250–1,000 miles, and more than 1,000 miles, the AI interaction is statistically significant only in the group beyond 1,000 miles. A one-standard-deviation increase in AI use attenuates the usual negative relationship between distance and loan growth by about 46% for loans below $100,000 and 64% for loans from $100,000 to $1 million.
That finding does not mean that AI removes every mile between a bank and a borrower. It means that the bank’s lending becomes less sensitive to distance in the observed data. It also does not mean that total lending rises by those percentages. The study finds no comparable increase in overall small-business credit when distance is not considered. AI appears to change the composition or geography of lending more clearly than it changes the total amount.
The broader FDIC survey establishes the other half of the answer. About 80% of banks define their small-business lending market by their branch footprint, and the average reported market among footprint-based banks extends about 40 miles. Only 6% of banks allow all four surveyed application steps to be completed online. The survey finds no statistically significant difference in average market distance according to whether a bank uses FinTech or automated underwriting. Even when communication is remote, banks commonly retain branch, on-site, and staff-centered practices.
The difference between the two sources is not a contradiction in measurement. The FDIC survey describes the dominant bank operating model in reported 2022 practice. The Census study identifies a specific distance-related effect associated with AI adoption in earlier data. One describes the center of the market; the other detects movement at an outer margin. A bank can remain relationship-oriented for most borrowers while using AI to make a small number of distant cases more assessable.