How AI In Lending Breaks Down When Compliance Is Treated As An Afterthought

AI does not break down in regulated lending because the model is inadequate. It breaks down when compliance, auditability, and governance are bolted on after the workflow is already built. The operations team inherits the gap: manual reviews, additional controls, check-the-checkers layered on top of each other, all because the traceability requirements that regulators demand were not part of the architecture from the start.

Gayle Borst is Chief Operating Officer at Finexus, which builds AI-powered lending solutions for financial institutions. She spent nearly 30 years at a leading financial institution, with the last 18 leading large-scale lending operations teams across the US, India, and the Philippines, spanning both mortgage and commercial small business lending. That background positions her as the conduit between customers and technical teams, translating regulatory and operational reality into what AI workflows actually need to support.

"In lending, it's not enough to say, 'the AI said yes' or 'the AI said no,'" Borst says. "You need to be able to demonstrate what data was used, what confidence thresholds were applied, where human judgment entered the process, and who ultimately had accountability for the decision. When you build that level of transparency into the workflow from the beginning, AI becomes much easier to scale because you're not constantly trying to recreate the decision after the fact."

Compliance belongs in the workflow, not after it

Borst uses a reconciliation tool as the clearest example. The system uses agentic AI to match bank statements against incoming payments: account number, customer name, deposit amount. When everything matches above the confidence threshold, AI approves the reconciliation automatically. When it doesn't, the system routes the exception to a human reviewer.

That routing is only half the story. "The other half is being able to document what happened after the handoff," Borst says. "Did the employee approve it because they had access to additional documentation? Did they override the recommendation because of a policy exception? Did they escalate to a manager?" Every action needs to be captured so that months or years later, the entire decision path can be reconstructed for regulators and auditors.

The alternative, treating compliance as a separate function that reviews work after the fact, produces rework, manual reviews, and controls stacked on controls. "When compliance isn't built in, operations teams are forced to create extra manual steps, additional approval layers, additional controls," Borst says. "It all adds time, complexity, and more opportunities for human error."

The real work is in the exceptions

Borst identifies exception handling, documentation, and data quality as the layers that leaders consistently underestimate. "The majority of the work isn't the transactions where everything matches perfectly," she says. "It's the exceptions. That's where human judgment comes in, and that's where regulators want to know exactly what happened."

The data foundation matters just as much. An AI agent will consistently look at the same field, on the same form, every time. A human underwriter may not. That consistency is an advantage only if the data is clean and the requirements are clearly defined. "Organizations spend a lot of time thinking about how to automate the decision, but not enough time thinking about how to document the decision path," Borst says. "A lot of organizations focus on the wrong efficiency problem. They're looking at the AI itself instead of the operational foundation underneath it."

Speed follows trust

When the operational foundation is right, the competitive advantage becomes tangible. Borst points to preapprovals: a human process that takes three days can run in three minutes with AI-supported workflows, giving the borrower an edge in a competitive offer environment. Underwriting spreading, the process of reviewing all financial documentation to determine whether a loan meets approval criteria, moves from days to seconds. In both cases, the speed is only defensible if the audit trail, exception handling, and compliance documentation are built into the workflow rather than reconstructed afterward.

Borst sees the competitive advantage sitting upstream of the model, in operational systems that earn institutional trust before they ever reach a customer. "The best AI solutions in lending are those that risk teams trust from the beginning, that compliance teams can defend, that auditors can validate, and that empower operations teams," she says. "It all has to start with risk, compliance, and audit having everything in line. Then you produce what you need to produce for your customer."