How would you feel to learn that every time you communicate with your lender, AI is also on the line, taking notes?
While that’s not happening with all lenders today, it could become a widespread practice in the future. Several major mortgage lenders have begun touting artificial intelligence as a tool to help make the mortgage process quicker and easier and to help them write more loans.
But how do we define AI? How can it help you in the mortgage process, and are there regulations put in place to protect you? To get the facts straight, we spoke with experts at some tech-focused mortgage companies.
- The difference between generative AI and automation
Today, underwriting for most mortgages is largely automated, with lenders using tools such as Fannie Mae’s Desktop Underwriter. When it comes to talking about AI, it’s important to differentiate between more commonly used automation technologies and the latest craze of generative AI, spurred on by products such as ChatGPT and DALL-E.
“I think a lot of people are using the AI term, but they’re not using actual generative AI,” says Brad Seibel, president of Sage Home Loans. Much of the technology driving things like online lending and fast preapprovals has been around for a while, according to Seibel. (Editor’s note: Sage is owned by Bankrate parent company Red Ventures.)
One commonly used technology in underwriting is OCR, or Optical Character Recognition. OCR allows a loan officer or underwriter to upload an image of text, printed or handwritten, and have that text transferred to a digital format.
“I’d put OCR in with classic machine learning,” says Christopher Jaynes, vice president of Product Forward Home Loans at Sage Home Loans. Machine learning is a branch of AI, but it’s not the same as generative AI. In the mortgage world, OCR can read scanned-in or uploaded documents to help determine if you qualify for a loan and what interest rate you’ll be offered.
This approach is different from generative AI, which is a more recently developed technology that compiles existing information to create new content. Generative AI can help lenders refine the information gathered with OCR, according to Jaynes.
“While we’ve been using AI in some forms and fashions for many years now, it’s really this boom around generative AI where we’re starting to really capitalize in different areas throughout the process,” says Josh Zook, chief technology officer for Rocket Mortgage.
Here’s how some lenders are incorporating generative AI throughout the mortgage process.
- Generative AI as a chatbot on lender websites
Chat features on lender sites are nothing new, but with advances in technology, they’re able to go deeper than before. Consumers can use the chat feature to learn more about different loan products, see what they may qualify for and start the loan process while talking with the AI.
“I’m seeing a lot of companies that are using the chatbot angle of having a consumer come to the website and be able to have a conversation with AI about what they’re looking for,” says Robert Heck, senior vice president of Revenue for Morty, an online mortgage marketplace. “Then also going through the traditional 1003 process through a more dynamic AI-based conversation.”
This means you could visit a lender’s website and use the chat feature to start a loan application. From there, generative AI is also working on the backend, helping lenders move your application from preapproval to underwriting to closing.
- How generative AI is making loan processing more efficient and accurate
For loan officers handling a large pile of loans, generative AI is a useful tool for getting the information needed to process a loan.
“For anyone who has done a mortgage, you know that there’s a lot of documents that are created for it,” says Jaynes. “Like your closing document might be 300 or 400 pages of all your supporting documents and your application.”
Jaynes explains that generative AI can distill this information into points to coach the loan officer into helping you with your mortgage.
“That’s also why we’re seeing a higher velocity of these types of tools being rolled out as co-pilots for existing production team members,” says Heck. “Fannie Mae guidelines may be like 1,400 pages. So, (generative AI) can help teams get to a concrete set of rules faster.”
Generative AI can also help make sense of scanned documents — such as paystubs, account statements, W-2s and more — which can boost the accuracy of document processing.
“We process around one and a half million documents a month,” says Zook. He also shares that Rocket Mortgage has achieved a much higher accuracy rate by using AI to identify the document and extract the data from it.
“We’ve been able to identify correctly the type of document on 70 percent of those documents of the one and a half million coming in, and extract over 90 percent of that information on a document,” says Zook.
Using AI to analyze the context of documents has not only allowed them to reduce the amount of time humans spend compiling and extracting this information but also made the process less error-prone, according to Zook.
But beyond handling documents, generative AI is being used to transcribe and pull information from phone conversations.
“As our banking team is working with clients, (our AI tool) is also listening to the conversations with the clients,” says Zook. “It’s pulling out key information that normally when a banker or loan officer is talking to a client, they’re sitting at a computer, and they have to type this in. So, one of the benefits that you see is that AI allows the banker to really focus and work with that relationship and the needs of the client, rather than the administrative responsibilities that go into capturing what the client said.”
Not only does this free up attention to focus on the client but also having the AI record the information lowers the margin of human error, says Zook.
- The concerns with AI in mortgage lending
While generative AI may be reducing human error, the technology itself isn’t immune to errors. Known as “hallucinations,” these errors can pop up in many ways.
Because of how a lot of text-based generative AI — like ChatGPT — is trained, math isn’t its strong suit. Jaynes explains that when you use generative AI to create text, it uses a large database of text to guess what word comes next. It can learn how language functions and build sentences accordingly.
But the rules of math are much more precise, and you don’t want to be guessing at numbers when talking about a mortgage. That said, how generative AI handles math is a major area where it needs to improve.
“OpenAI has come (up) with a hybrid approach where you’re using generative AI, but you’re using it to generate the code to do the math,” says Jarnes.
The problem is that, according to Jaynes, this approach doesn’t consistently get it right, and that doesn’t cut it. So, calculating interest rates and monthly payments, for instance, are best left to tried-and-true computing.
Another concerning area where AI hallucinates is in regards to racial bias. Housing in the U.S. has had a long history of racism, with discriminatory practices such as redlining, and the Black neighborhood home appraisal gap. When AI is trained on these biases, it displays them itself.
ChatGPT-4, OpenAI’s latest iteration of its chat-based generative AI, steered prospective homebuyers to buy in certain neighborhoods based on their race, according to a 2024 study published by MIT. Black homebuyers were recommended majority-Black neighborhoods, and white homebuyers were recommended majority-white neighborhoods. These findings were amplified in more segregated cities, such as New York City and Chicago.
Before this technology is widely adopted, lenders and borrowers are going to need to trust the companies developing it and the results it generates.
- Many lenders will be slow to adopt until there’s more regulation
“The mortgage industry is such a heavily regulated industry at this point that I think you typically see things move over years as opposed to months,” says Heck.
Government agencies have begun releasing some guidance on how generative AI can be used in housing. In September 2023, the Consumer Financial Protection Bureau released a statement clarifying that lenders denying borrowers on the basis of credit must explain why.
“Technology marketed as artificial intelligence is expanding the data used for lending decisions, and also growing the list of potential reasons for why credit is denied,” said CFPB Director Rohit Chopra in the statement. “Creditors must be able to specifically explain their reasons for denial. There is no special exemption for artificial intelligence.”
In addition, the Department of Housing and Urban Development released guidelines in May 2024 stating how lenders must adhere to the Fair Housing Act when using AI and algorithms in advertising.
- We’re still in the very early days of generative AI.
More government leadership and regulation are needed before AI will be more widely adopted. That’s largely because most loans have to meet specific criteria to participate in the mortgage market.
“On the backside, where all the loans are sold and securitized, and even where the banks buy the securities, there are minimum requirements that have to be met,” says Seibel. “So even if AI says ‘We don’t need a paystub. We can see this guy’s got a job,’ in order for a bank to buy that loan as part of a security, that piece of paper has to be there. Until the very end of the cycle of where loans go accepts some of this decision-making, I think adoption is going to be restricted on where it can be applied.”
- The human touch is still needed when getting a mortgage
While technology is becoming a bigger part of the mortgage process, many people still want to talk to a person at some point in getting a mortgage.
“The reality is that this is probably your single biggest transaction, certainly if you’re a first-time homebuyer, and that makes you want to talk to a person,” says Seibel.
“I still think that people generally just want and trust the human component,” adds Heck. He says that a large part of that is because buying a home and getting a mortgage can be an emotional process.
Where AI can do the data entry and find patterns in the data, the human loan officers can excel at coaching borrowers through the process.
