Can Bank AI Replace Proximity?
U.S. bank small-business lending is the subject here, and the question is precise: can artificial intelligence reduce the need for the local branches, face-to-face relationships, and community knowledge that have traditionally helped banks decide which small firms are creditworthy? This series supports a practical judgment rather than a slogan. It will help readers distinguish a bank reaching a distant borrower from a bank no longer needing proximity at all.
That distinction matters because small businesses are economically large but individually difficult to evaluate. The FDIC’s 2024 report notes that the United States had more than 33 million small businesses in 2023—99.9% of all firms, 46.4% of private-sector employment, and 43.5% of gross domestic product. Their access to credit affects more than a bank’s loan book. It affects whether a young firm can hire, whether an established firm can survive a shock, and whether a community has businesses capable of growing within it.
Yet a small business is not always a clean data object. Its financial records may be incomplete, its owner’s experience may matter as much as its balance sheet, and its prospects may depend on local customers, suppliers, and economic conditions. A bank may need to understand not just what the numbers say, but why they look that way. That is where proximity enters the credit decision.
“Proximity” does not mean only a nearby building
A branch is the most visible form of proximity, but it is not the whole idea. Physical closeness can make it easier for a borrower to meet a banker, bring in documents, or ask questions. It can make it easier for bank staff to visit a business and observe the setting in which a loan will be repaid. A local presence can also connect a bank to accountants, business owners, civic organizations, and other sources of referrals and context.
The less visible form is informational proximity: the ability to gather and interpret evidence about a borrower. Banks often call the quantitative, standardized part of that evidence “hard information.” Credit scores, financial statements, debt-service measures, and collateral can be recorded and transmitted without carrying every detail of the original conversation. “Soft information” is different. It includes a loan officer’s assessment of the owner or management team, knowledge of local market conditions, and context learned through repeated interaction. It is more difficult to reduce to a portable score.
A third form is relational proximity. A small-business borrower may value continuity with staff, the willingness to discuss an unusual situation, or the flexibility to structure a loan around the firm’s actual cash flow. These are not merely sentimental additions to underwriting. They can change what information a borrower is willing to disclose and how a bank responds when a case does not fit a standard template.
The three forms overlap, but they are not interchangeable. A video call can replace some travel while leaving the bank with less local knowledge. A better data system can make records easier to analyze while doing nothing to create trust. An online application can widen the top of the funnel while leaving the difficult work of verification and judgment untouched. Asking whether AI “replaces proximity” therefore requires asking which function is supposed to disappear.
The evidence begins with a contradiction
The FDIC’s Small Business Lending Survey provides a useful baseline. The 2022 survey, conducted between June 2022 and January 2023, received about 1,300 responses from a sample of 2,000 banks, a 68% response rate. It asked banks about their typical practices rather than about a single technology pilot or a handful of high-profile lenders.
The result is unmistakably local. About four in five banks define their small-business lending market by their branch footprint. Across all banks, 36% report that their small-business borrowers are generally within 20 miles of a branch, while two-thirds place them within 60 miles. Among banks that use a branch footprint to define their market, the average distance is about 40 miles. The FDIC also finds that this average does not differ significantly according to whether a bank uses FinTech, uses automated underwriting, meets applicants, or is small or large.
That does not mean every bank refuses distant business. Twenty percent of banks say their market is not based on a branch footprint, and some institutions lend far beyond their local area. But the normal bank model has not become geographically weightless simply because digital communication exists. When banks explain why their market does not extend farther, the leading reasons include borrowers’ preference for a closer bank, difficulty forming a deposit relationship, difficulty obtaining referrals, the practical challenge of visiting distant borrowers, and the difficulty of understanding local economic conditions or verifying information.
At the same time, another source points in the opposite direction. A February 2025 working paper using confidential Census technology-survey data finds that banks with greater AI use lend relatively more to distant small-business borrowers. The paper reports that AI use among the banks in its technology sample rose from 14% in 2017 to 43% in 2019. Its main result is not that AI banks lend more to everyone. Rather, the usual decline in lending as borrower distance increases becomes less pronounced for banks with greater AI use, with the strongest discrete effect appearing for borrowers more than 1,000 miles away.
The two findings are not logically inconsistent. The FDIC survey describes the dominant operating model across banks in 2022 practice. The Census study identifies a distance-related change in the composition of lending associated with earlier AI adoption. One describes the center of the market; the other detects movement at a particular margin. A bank can continue to rely on branches and relationships for most borrowers while using AI to make a limited set of distant cases more assessable.
What would count as replacement?
There are at least three different claims hidden inside the word “replace.” The weakest claim is that technology can reduce some travel, paperwork, or manual review. The FDIC evidence already shows this kind of change: banks use FinTech in compliance, servicing, portfolio analytics, application processing, and other stages. But ordinary process automation is not the same as AI-driven underwriting, and remote communication is not the same as a fully remote lending relationship.
A stronger claim is that AI can replace some locally gathered information. If a system can extract usable facts from messy financial records, combine them with other data, and identify patterns that a conventional review would miss, a bank may be able to assess a borrower without knowing that borrower through years of local interaction. This is the specific possibility the Census study investigates indirectly through lending distance.
The strongest claim is that branches, human judgment, and local relationships no longer matter to bank small-business lending. The available evidence does not support that conclusion. Only 6% of banks in the FDIC survey allow all four identified application steps to be completed online. Branches and on-site visits remain widely available, and remote communication often supplements rather than substitutes for in-person interaction. Nearly all banks encourage staff to develop relationships, share expertise, maintain continuity with customers, and participate in their communities.
Even technology adoption itself is more restrained than the word “AI” can imply. In the FDIC survey, 31% of banks were using FinTech in at least one lending step and another 22% were discussing or developing it. But 69% used it in no step, and the largest uses were not borrower identification or underwriting. Regulatory compliance was the most common use. Only 8% reported using FinTech in underwriting and decision-making, and 4% used it to identify potential borrowers. Large banks adopted technology more broadly, yet the report still found that most FinTech banks typically met applicants.
This is why the central question is not whether banks are becoming digital. They are. The question is whether a particular digital capability changes the information economics of a particular loan. AI could make distance less costly when the main obstacle is collecting and interpreting hard information. It may do much less when the obstacle is building trust, understanding a local industry, monitoring an unusual business, or deciding how much flexibility a thin-file borrower deserves.
The rest of this series will examine that distinction closely. First, it will reconstruct why small-business lending became local and how small and large banks use different mixtures of hard and soft information. It will then map where banks actually use technology, before examining how AI could change the cost of collecting and interpreting evidence. The empirical findings can then be read at the right scale: as evidence about distant lending, borrower quality, pricing, and underserved areas—not as a declaration that every branch has become unnecessary.
The provisional answer is therefore deliberately narrow. Bank AI may stretch proximity by allowing some credit decisions to travel farther than a bank’s traditional information network. That is a meaningful change, especially where local credit is scarce. But it is not the same as making proximity disappear. To see why the distinction matters, the next step is to understand what local relationships have been doing inside the lending decision all along.