How AI Turns Distance into a Testable Lending Question

In U.S. bank small-business lending, AI matters for proximity only if it changes the cost of producing usable evidence about a borrower. The specific question is whether systems that collect, organize, and interpret hard information can compensate for some of the soft information a bank normally gathers through local relationships. That mechanism can be explained clearly—and, unlike a vague claim that banking is becoming digital, it can be tested against lending distance and loan outcomes.

The distinction is important. A communication tool can move a document from one place to another. An AI system, at least in the account examined by the Census working paper, could change what the bank is able to learn from that document. The former reduces friction in a workflow. The latter could alter the informational reason a bank has traditionally needed to stay close to a borrower.

AI’s job is to make hard information usable

Small businesses often have more information than a bank can efficiently process. Imagine a restaurant with sales records, invoices, bank transactions, payroll data, tax documents, and notes about a seasonal disruption. The records may arrive in different formats. Some may be scanned, handwritten, or organized according to the owner’s habits rather than a lender’s template. The difficulty is not necessarily that no evidence exists. It is that collecting, cleaning, matching, and interpreting it may take more staff time than a small loan can economically support.

The paper identifies three ways AI could change that calculation.

The first is cheaper data acquisition. Image recognition and language-processing tools can extract information from raw financial records rather than requiring a staff member to re-enter every figure. If a bank can gather more of the available evidence within the same budget, it may no longer need to rely on a narrow set of easily portable indicators. The owner’s personal credit score and a debt-service measure may remain useful, but they need not carry the entire burden of describing the business.

The second is cheaper integration. A borrower’s information may be scattered across statements, internal bank records, transaction data, and documents supplied during the application. An AI-enabled workflow can help standardize information from disparate sources and place it in a common view. That matters because evidence that is technically available but isolated in different systems is not yet decision-useful. The bank must be able to connect the facts, compare them, and notice when they do not fit together.

The third is more effective pattern recognition. Traditional statistical models often depend on a limited set of assumptions and variables that can be expressed in a standardized form. Machine-learning methods can search for nonlinear relationships in data that a simple rule or a time-constrained human review might overlook. A pattern is not automatically a reason to approve a loan. It is a signal that may help a reviewer distinguish a temporary irregularity from a persistent weakness, or a promising firm from one whose numbers are misleading.

Taken together, these mechanisms could make hard information a partial substitute for soft information.