What AI Can and Cannot Do in a Credit Underwriting Process

The conversation about AI in credit underwriting is frequently conducted at two extremes. One extreme treats AI as a replacement for human underwriters — a tool that will automate the credit decision and eliminate the need for experienced credit staff. The other extreme dismisses AI as a compliance risk that institutions should avoid until the regulatory framework is clearer. Neither position is useful.

AI can do several things well in a credit underwriting process. It can accelerate document review — extracting financial data from tax returns, financial statements, and bank statements faster and more consistently than a human reviewer. It can surface patterns in financial data that a human reviewer might miss or take longer to identify. It can flag anomalies, inconsistencies, and missing information earlier in the process, reducing the back-and-forth that slows cycle time.

AI can also support credit spreading, covenant monitoring, and portfolio surveillance — tasks that are time-consuming for human analysts and that benefit from consistent application of defined criteria. These are areas where AI-assisted tools can meaningfully increase the throughput of a credit team without reducing the quality of the analysis.

What AI cannot do is own the credit decision. The institution retains the lender-of-record relationship, the credit authority, and the compliance obligations that attach to both. An AI output that recommends approval or decline is an input to a human decision — not the decision itself. The human reviewer must be able to understand the output, evaluate it against their own judgment, and document the basis for the decision they make.

AI also cannot satisfy a regulatory examination on its own. Examiners will ask who made the credit decision, what information they reviewed, and how they documented their judgment. An institution that cannot answer these questions because the AI made the decision has a governance gap that the AI cannot close.

The institutions that will benefit most from AI in credit underwriting are those that use it to accelerate the work that humans do — not to replace the judgment that humans must retain. That requires a clear definition of which decisions AI will inform, which decisions humans will make, and how the institution will document the boundary between them.