Artificial intelligence adoption in mortgage lending is not hitting a single wall of resistance. Instead, it is fracturing along organizational lines. Benjamin Gerochi, AI Engineer and AI Solutions Architect at Fintor, argues that executives and front-line workers are broadly enthusiastic about AI, while middle managers remain the most skeptical layer.
That split, he says, reflects a genuine structural problem: AI systems are probabilistic, and the mortgage industry demands deterministic accuracy. The people caught in the middle are the ones who bear that accountability.
The Probability Accuracy Problem
The technical issue at the center of middle management’s resistance is not that AI is unproven. It is that AI is inherently probabilistic. Every output from a large language model carries a degree of uncertainty. In consumer applications, that uncertainty is tolerable. In mortgage lending, a processing error can derail a borrower’s home purchase or create regulatory exposure for the lender. That changes the calculus entirely.
“AI still makes mistakes,” Gerochi says. “It’s still a stochastic system, and what that means is there’s still probability involved. You can’t avoid the mistakes that get made.”
He draws a distinction between deterministic systems, where a defined input produces a defined output, and stochastic systems, where probability is always present. Current AI models fall into the stochastic category. That is not a criticism of the technology. It is simply a description of how it functions. But in a regulated industry, compliance staff cannot accept a system that is right most of the time.
Why Middle Managers Resist
The organizational split Gerochi describes follows a clear logic. Executives operate at a level of abstraction where competitive strategy matters more than individual loan accuracy. Front-line workers, such as processors and data entry staff, experience the tedium of repetitive manual work and are often eager for relief. Middle managers and senior processors, however, sign off on accuracy, field compliance questions, and absorb the consequences when something goes wrong.
“At the highest level, there is a lot of excitement, because there is no doubt that AI will change the industry,” Gerochi says. “But at the middle level, there is a lot of skepticism.”
This pattern is not unique to mortgage lending, but the stakes here amplify it. A mistake in a loan file affects a borrower’s ability to close on a home, potentially on a deadline. Middle managers who have spent their careers building processes to prevent those mistakes are not going to abandon their skepticism because an executive attended a conference on AI efficiency.
The broader industry implication is that AI vendors targeting mortgage lenders cannot sell to the C-suite and expect adoption to follow. The resistance is structural and lies in the layer of the organization that controls day-to-day workflow decisions.
What Lenders Really Want
Beyond organizational dynamics, the specific questions lenders raise reveal what adoption will require. Gerochi says the most common concern he encounters is not about AI capability. It is about accuracy and what happens when the AI is wrong. “A lot of the skepticism is around the accuracy of the data and making sure that the AI is right,” he says.
That concern shapes what lenders are willing to adopt. Systems that remove humans from the loop entirely face the steepest resistance. Systems that use AI for specific, bounded tasks, such as document checks, data extraction, and data entry, while preserving human review at key decision points, encounter less friction. The distinction reframes the question. It is no longer “can we trust the AI?” but “can we verify what the AI did?”
For the broader mortgage technology market, this suggests that the path to adoption runs through accountability structures rather than accuracy claims alone. Lenders are not asking whether AI can process documents faster than humans. They are asking who is responsible when it makes a mistake, and whether the system is designed to catch errors before they reach borrowers or regulators.
An Illustrative Case Study
Fintor’s approach illustrates one way vendors are attempting to close this accountability gap: pairing automated processing with a human checkpoint before outputs move forward in the pipeline. Rather than positioning AI as a replacement for judgment, the model treats it as a first pass that still requires sign-off. “We have AI that extracts the data, but then we also have a production team that goes and ensures that those outputs are correct,” Gerochi says.
This is not the only way to structure that checkpoint. Gerochi’s account does not address how the model holds up under volume, across different loan types, or when errors slip past both the AI and the reviewing team. Those are the questions a skeptical middle manager would likely ask next, and they point to a broader test facing any AI-plus-human review model: whether the human layer remains a meaningful check as transaction volume scales, or becomes a rubber stamp under production pressure.
About the Expert: Benjamin Gerochi is an AI engineer and solutions architect at Fintor, a company building an AI-native mortgage operating system for lenders that integrates with existing loan origination platforms.
This article is intended for informational purposes only and does not constitute legal, financial, or investment advice. The views and opinions expressed herein reflect those of the individuals quoted and do not represent an endorsement of any company, product, or service mentioned. Readers should conduct their own due diligence and consult qualified professionals before making any investment decisions.
