According to HousingWire, mortgage lenders are pouring money into artificial intelligence, yet the financial results tell a different story. In the first quarter of this year, the average cost to originate a loan climbed to $11,898 while lenders netted just $727 in pre-tax production profit per loan. That gap between investment and return reveals something important about how the industry actually works.
The real issue is structural, not technological. Most AI tools that lenders have deployed so far improve individual tasks within the existing workflow, but they do not fundamentally change how mortgage companies operate. The mortgage business remains deeply reliant on people. From origination through underwriting, closing, and servicing, growth still requires hiring more staff to collect information, resolve exceptions, and manage handoffs between different systems and institutions. Whether a firm is large or small, adding capacity tends to mean adding people and management layers without necessarily improving profitability.
This heavy reliance on people stems from the fragmentation built into mortgage lending itself. Each loan involves countless variables—the borrower's income and credit, the property type, the loan product, investor requirements, regulations—that create enormous operational complexity. A single loan often moves between institutions that do not share systems, data, or processes, which means people end up carrying information and context from one place to the next. Retail lenders internalized more of this workflow to create consistency, but that approach comes with higher human capital costs passed on to borrowers. Brokerages preserved lender and product variety, but that optionality came with operational complexity and variation across guidelines and portals.
Technology over the past twenty years has tried to address fragmentation, but it has mostly optimized individual pieces of the process rather than solving the underlying problem. The industry built separate systems for origination, pricing, underwriting, documents, and compliance, with people still required to connect those systems and handle exceptions. The real opportunity with AI is different. Instead of relying on employees to learn and navigate every lender's unique process, AI can translate lender-specific requirements into a common workflow, interpret documents, standardize information, and route exceptions to human experts while keeping controls and decisions under human oversight.
The challenge is organizational, not technological. Established mortgage companies have built their roles, reporting structures, compensation models, and controls around human coordination. They are layering AI onto their existing fragmented software stacks, accelerating current workflows without changing the underlying operating model. That approach produces only incremental gains. Real competitive advantage will come when companies use AI to redesign their operations so they can handle more volume, more lender choices, and more personalization without increasing human coordination at the same rate.
What I am seeing locally is that Bay Area and East Bay lenders and brokers who can move beyond incremental AI improvements will gain real advantage. The firms that figure out how to streamline handoffs and reduce the manual translation work between different lender systems will be able to offer borrowers faster service and potentially better pricing. For Fremont and the broader region, this shift matters because it could eventually bring down origination costs and make lending more efficient, though we are still in the early stages of that transformation.
