How would you react to discovering that every time you interact with your lender, AI is also listening in and taking notes?

While this isn’t the case with all lenders currently, it could become a common practice in the future. Several major mortgage lenders are starting to promote artificial intelligence as a way to streamline the mortgage process and increase loan origination.

But what exactly is AI? How can it assist you during the mortgage process, and what regulations exist to protect you? To clarify, we consulted experts from tech-focused mortgage companies.

The difference between generative AI and automation

Today, the underwriting process for most mortgages is largely automated, with lenders utilizing tools like Fannie Mae’s Desktop Underwriter. When discussing AI, it’s essential to distinguish between traditional automation technologies and the newer generative AI, inspired by innovations like ChatGPT and DALL-E.

“As people use the term AI, they often aren’t referring to true generative AI,” says Brad Seibel, president of Sage Home Loans. Many technologies that facilitate online lending and quick preapprovals have been in use for years, according to Seibel. (For context, Sage is a subsidiary of Bankrate’s parent company, Red Ventures.)

One common technology in underwriting is Optical Character Recognition (OCR). This allows loan officers or underwriters to upload images of printed or handwritten text, converting it into a digital format.

“OCR fits into the category of classic machine learning,” explains Christopher Jaynes, vice president of Product Forward Home Loans at Sage Home Loans. While machine learning is a subset of AI, it differs from generative AI. In the mortgage industry, OCR helps read scanned documents to assess loan eligibility and determine interest rates.

This contrasts with generative AI, which synthesizes existing data to create new content. Generative AI can assist lenders in refining the information obtained through OCR, according to Jaynes.

“While we’ve been utilizing AI in various forms for years, the recent surge in generative AI is allowing us to leverage it across different areas of the mortgage process,” notes Josh Zook, chief technology officer for Rocket Mortgage.

Here’s how some lenders are integrating generative AI throughout the mortgage journey.

Generative AI as a chatbot on lender websites

Chat features on lender websites have been around for a while, but advancements in technology are enhancing their capabilities. Consumers can now use these chat functions to explore various loan products, check their eligibility, and initiate the loan process through AI interactions.

“I’m seeing many companies implementing chatbots that enable consumers to engage with AI about their needs directly on the website,” says Robert Heck, senior vice president of Revenue for Morty, an online mortgage marketplace. “This includes navigating the traditional 1003 application process through a more interactive AI conversation.”

As a result, you can visit a lender’s site, utilize the chat feature to kick off a loan application, and behind the scenes, generative AI aids in progressing your application from preapproval to underwriting and finally to closing.

How generative AI is making loan processing more efficient and accurate

For loan officers managing a large volume of loans, generative AI is a valuable tool for gathering essential information needed for processing.

“As anyone who’s gone through the mortgage process knows, it generates a significant amount of documentation,” says Jaynes. “Closing documents can span 300 to 400 pages, filled with supporting documents and the application itself.”

Jaynes notes that generative AI can summarize this information, guiding loan officers in assisting borrowers effectively.

“This is why we’re seeing a rise in these tools being implemented as co-pilots for existing production teams,” adds Heck. “Fannie Mae guidelines, for instance, can be as lengthy as 1,400 pages, so generative AI helps teams quickly access relevant rules.”

Moreover, generative AI aids in interpreting scanned documents, such as pay stubs and W-2s, enhancing document processing accuracy.

“We process about one and a half million documents each month,” says Zook, highlighting that Rocket Mortgage has significantly improved accuracy through AI in identifying and extracting data from documents.

“We’ve accurately identified the type of document for 70 percent of those million and a half, extracting over 90 percent of the information contained within,” Zook explains.

Using AI for document analysis not only reduces the time humans spend on compiling and extracting information but also minimizes errors.

Beyond document management, generative AI is also used to transcribe and extract information from phone conversations.

“As our banking team interacts with clients, our AI tool listens to the conversations,” shares Zook. “It captures key information that typically requires the banker or loan officer to type notes during the call. This allows them to focus more on building the client relationship rather than on administrative tasks.”

This approach not only enhances client interaction but also reduces the likelihood of human error in capturing important details, according to Zook.

The concerns with AI in mortgage lending

While generative AI can help minimize human error, the technology itself is not without flaws, often resulting in “hallucinations” that manifest in various ways.

For instance, text-based generative AI like ChatGPT struggles with mathematical accuracy. Jaynes explains that these models predict the next word based on a vast database of text, effectively learning language patterns. However, the precise nature of math doesn’t lend itself to guesswork, especially when it comes to mortgage calculations. As a result, improvements in how generative AI handles mathematical tasks are essential.

“OpenAI has developed a hybrid approach that uses generative AI to produce code for mathematical calculations,” Jaynes notes. However, this method doesn’t always yield accurate results, making traditional computing the better choice for tasks like calculating interest rates and monthly payments.

Another significant concern involves biases inherent in AI training data, particularly those related to race. The history of housing discrimination in the U.S., such as redlining and appraisal gaps, can be reflected in AI outputs. A 2024 study by MIT revealed that ChatGPT-4 recommended homebuyers purchase in neighborhoods based on their race, suggesting majority-Black neighborhoods to Black buyers and majority-white neighborhoods to white buyers. These trends were more pronounced in cities with higher segregation, such as New York City and Chicago.

For widespread adoption of this technology, lenders and borrowers must have confidence in both the developers and the accuracy of the AI-generated outcomes.

Many lenders will be slow to adopt until there’s more regulation

“The mortgage industry is so heavily regulated that changes tend to happen over years rather than months,” says Heck.

Recently, government agencies have started to provide guidance on the use of generative AI in housing. In September 2023, the Consumer Financial Protection Bureau (CFPB) issued a statement requiring lenders to clearly explain why they deny borrowers based on credit.

“Technology marketed as artificial intelligence is broadening the data used in lending decisions and increasing the reasons for credit denial,” said CFPB Director Rohit Chopra. “Creditors must specify their reasons for denial; there’s no special exemption for AI.”

Additionally, the Department of Housing and Urban Development (HUD) released guidelines in May 2024 detailing how lenders must comply with the Fair Housing Act when employing AI and algorithms in advertising.

More government oversight and regulation are necessary before AI can be more widely adopted in the mortgage sector. This is largely due to the fact that most loans must meet specific criteria to qualify for the mortgage market.

“On the back end, where loans are sold and securitized, there are minimum requirements that must be met,” explains Seibel. “Even if AI suggests, ‘We don’t need a paystub; we can verify employment,’ banks still require that documentation to purchase the loan as part of a security. Until the entire loan lifecycle accepts some of this AI decision-making, its adoption will be limited.”

The human touch is still needed when getting a mortgage

While technology is playing an increasing role in the mortgage process, many people still prefer to speak with a person at some point.

“The reality is that for many, especially first-time homebuyers, this is likely the biggest financial transaction they’ll make, which drives the desire for human interaction,” says Seibel.

Heck adds, “People generally value and trust the human touch.” He notes that the emotional nature of buying a home and securing a mortgage contributes to this preference.

For Zook at Rocket Mortgage, AI is about enhancing the human element.

“We’ve found that the best applications of AI involve helping humans do what they do best while allowing computers to handle data entry and pattern recognition,” says Zook.

This way, human loan officers can focus on guiding borrowers through the process.

“We ensure that there’s always a human involved. We don’t rely on AI or automation for any lending decisions,” Zook emphasizes.

Ultimately, a human must approve the loan, and that’s unlikely to change anytime soon.