AI Exposure Is Entering the Business Credit File

Artificial intelligence is beginning to affect business credit before its long-run winners and losers are settled. Banks are not simply rewarding borrowers that say they use AI. They are forming views about whether AI will strengthen or weaken the economics of the borrower’s sector during the life of a loan.

That shifts the burden of proof. A company seeking credit needs to connect AI adoption, competitive defense, and implementation cost to cash flow that can repay the lender. A technology label is not an underwriting case.

What the evidence shows

The Federal Reserve’s January 2026 Senior Loan Officer Opinion Survey, released February 2, asked domestic banks how their willingness to approve commercial and industrial loans had changed since January 2025 for sectors with different levels of AI exposure. The question assumed that the borrower’s other characteristics were typical for its sector, isolating the bank’s view of the sector effect.

For sectors expected to benefit from high AI exposure, 22.2% of respondents said they were somewhat more likely to approve a loan, 72.2% were about as likely, and 5.6% were somewhat less likely. Large banks were more positive: 42.1% said they were somewhat more likely to approve.

The reaction to sectors expected to be harmed was much stronger. In the detailed response table, 52.8% of banks said they were somewhat less likely to approve and 5.7% said much less likely. Among large banks, the figures were 63.2% and 5.3%, respectively. Every respondent said approval likelihood was unchanged for sectors with little AI exposure.

The asymmetry is the finding: lenders were more united about possible downside than upside. A bank may see many ways for a borrower to spend on AI without producing durable profit, but it can react quickly when it believes automation will compress price, weaken demand, reduce switching costs, or impair collateral.

The survey captures judgment, not realized defaults. Fifty-four domestic banks answered the question about beneficial exposure; 53 answered the question about adverse exposure. It does not show whether individual applications were approved, how loan terms changed, or whether the sector assessments will prove correct.

The operating consequence

Credit underwriting is forward-looking. A lender does not need to wait for a sector’s revenue to decline if it expects the borrower’s economics to weaken during the loan term. That view can affect approval, line size, maturity, covenants, collateral, and pricing well before public financial results show a clear break.

The effect may be sharper for smaller companies. They have fewer financing alternatives, thinner management capacity, and less ability to document a transformation program. In the same survey, 24.5% of banks expected the quality of small-firm C&I loans to deteriorate somewhat during 2026, compared with 13.2% for large and middle-market firms. The Fed did not attribute that difference solely to AI, but it raises the cost of arriving with an untested story.

Sector labels can also produce mistakes. Two firms exposed to the same technology can have opposite credit outcomes. One may automate a costly workflow while protecting distribution and customer relationships. Another may reduce labor expense but lose pricing power because competitors can reproduce its service. A lender that stops at the sector label can miss both the strongest borrower and the weakest one.

For operators, the practical implication is that AI strategy now travels beyond product and investor conversations. It can enter working-capital discussions, refinancing, acquisition facilities, and covenant negotiations. Claims that once sounded promotional may be tested against debt-service capacity.

What operators should do now

Build an underwriting bridge from the workflow to cash flow. Identify the process being changed, the baseline cost or revenue, the adoption rate, implementation expense, timing, and control environment. Show which effects have been observed and which remain forecasts. Lenders need to see when the benefit reaches free cash flow, not merely when a tool is deployed.

Map displacement risk with equal care. If AI lowers barriers to entry, compresses prices, or lets customers do more work themselves, quantify the exposure and explain the response. Relevant defenses may include proprietary distribution, embedded workflow, contractual revenue, regulated permissions, unique data rights, or a service component that remains scarce. Name the mechanism rather than declaring the company protected.

Stress-test borrowing capacity under a downside case. Assume slower renewal, lower gross margin, higher technology spending, or a longer payback period. Calculate covenant headroom and identify which investments can be slowed without damaging the core business. Preparing this case internally is better than allowing the lender to supply a harsher one with no operating context.

Keep evidence at the level of the borrower. Sector research is useful for setting questions, but it should not determine the answer. Track productivity, error rates, customer retention, price realization, implementation cost, and security incidents in the exact workflows affected.

Lenders should apply the same discipline. AI exposure belongs in underwriting when it changes cash-flow durability, not as a blanket premium or penalty. The right unit of analysis is the borrower’s mechanism, management response, and financial capacity.

The decision

AI has entered credit allocation before the market has reached consensus on its economic effects. The Federal Reserve survey does not prove that banks are correctly identifying winners and losers, but it shows that the judgment is already influencing approval posture.

Borrowers facing disruption need more than reassurance, and expected beneficiaries need more than enthusiasm. Both need a quantified account of what changes, what it costs, and how the result reaches cash available for debt service.