How would you feel if you knew that every time you communicated with a lender, an AI was also on the phone and taking notes?
Not all lenders are doing it right now, but it could become more widespread in the future: Several large mortgage lenders have begun to promote artificial intelligence as a tool that can help make the mortgage process faster and easier, and therefore make more loans.
But how do we define AI? How can AI help in the mortgage process? And are there regulations in place to protect you? To get the facts straight, we spoke to experts from tech-focused mortgage companies.
The difference between generative AI and automation
Today, most mortgage underwriting is largely automated, with lenders using tools like Fannie Mae's Desktop Underwriter. When talking about AI, it's important to distinguish between the more commonly used automation technologies and the latest generative AI fad fueled by products like ChatGPT and DALL-E.
“I think a lot of people use the term AI but not actual generative AI,” says Brad Seibel, president of Sage Home Loans. Seibel says much of the technology driving things like online lending and quick pre-approvals has been around for a while. (Editor's note: Sage is owned by Bankrate's parent company, Red Ventures.)
One technology commonly used in underwriting is OCR (Optical Character Recognition), which allows loan officers or underwriters to upload an image of printed or handwritten text and convert that text into a digital format.
“I think you'll combine OCR with traditional 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 different from generative AI. In the mortgage world, OCR reads scanned or uploaded documents to help determine whether you qualify for a loan and what interest rate you'll be offered.
This approach differs from generative AI, a more recently developed technique that assembles existing information to create new content, which Jaynes said can help lenders refine the information they collect through OCR.
“We've been using AI in some form for years, but it's the boom in generative AI that's really starting to use it in different areas throughout the process,” says Josh Zook, chief technology officer at Rocket Mortgage.
Here’s how some lenders are incorporating generative AI into the entire mortgage process.
Generative AI as a chatbot on lender websites
Chat features on lender sites aren't new, but advances in technology have allowed them to go much deeper than before. Consumers can use chat features to learn more about different loan products, see which loans they qualify for, and get started on the loan process while having a conversation with an AI.
“We're seeing a lot of companies leverage the chatbot angle, where a consumer can go to a website and have a conversation with an AI about what they're looking for,” says Robert Heck, senior vice president of revenue at Morty, an online mortgage marketplace, “and take the traditional 1003 process and run it through a more dynamic, AI-based conversation.”
This means you can go to a lender's website and use the chat feature to begin your loan application. From there, generative AI also works on the backend to help the lender move your application from pre-approval to underwriting to closing.
How Generative AI Can Make Loan Processing More Efficient and Accurate
For loan officers who process large volumes of loans, generative AI is a useful tool for getting the information they need to process loans.
“Anyone who has ever taken out a mortgage knows there is a lot of paperwork that goes into a mortgage,” Jaynes says. “For example, loan closing documents can be 300 to 400 pages long with all the supporting documents and applications.”
Janes explains that generative AI can distill this information into points and coach loan officers to help with the mortgage process.
“So these tools are being deployed more frequently as co-pilots with existing production team members,” Heck said. “Fannie Mae’s guidelines can be as long as 1,400 pages. [generative AI] It helps the team arrive at a concrete ruleset faster.”
Generative AI can also help make sense of scanned documents like pay stubs, bank statements, and W-2s, improving the accuracy of document processing.
“We process about 1.5 million documents a month,” Zook says, adding that Rocket Mortgage has achieved much higher accuracy by using AI to identify documents and extract data from them.
“We were able to correctly identify the document type for 70 percent of the 1.5 million documents sent to us and extract more than 90 percent of the information from the documents,” says Zook.
According to Zook, using AI to analyze the context of documents not only reduces the time humans would spend gathering and extracting this information, but also makes the process less prone to errors.
But beyond processing documents, generative AI is also being used to transcribe phone conversations and extract information.
“Our banking team works with clients to [our AI tool] “AI is also listening to conversations with customers,” Zuke says. “It pulls out important information. Typically, a banker or loan officer has to sit at a computer and type when they're talking to a customer. One of the benefits that AI gives us is that it allows bankers to focus on the customer's interaction and needs, rather than being bogged down in the administrative responsibility of capturing what the customer is saying.”
Not only does this give them more time to focus on their clients, but Zook says having AI record the information also reduces the room for human error.
AI Concerns in Mortgage Lending
While generative AI may reduce human error, the technology itself is not immune to error. These errors, called “hallucinations,” can manifest in a variety of ways.
Many text-based generative AIs like ChatGPT are not known to be good at math because of the way they are trained. Jaynes explains that when generative AI is used to create text, it uses a large database of text to guess what words will come next. It can learn about the features of language and build sentences accordingly.
But the rules of math are much more precise, and you don't want to guess numbers when talking about mortgages. That being said, the way generative AI handles math is a major area that needs improvement.
“OpenAI is [up] “In a hybrid approach, you use generative AI, but you use it to generate code that does the calculations,” Jarnes says.
The problem, according to Jaynes, is that this approach doesn't always produce the right results, and it's not enough, so it's best to leave things like calculating interest rates and monthly payments to tried-and-true calculators.
Another worrying area where AI can hallucinate is around racial bias. US housing has a long history of racism with discriminatory practices like redlining and disparate home assessments in black neighborhoods. When AI is trained on these biases, it will display them.
According to a study published by MIT in 2024, the latest version of OpenAI's chat-based generative AI, ChatGPT-4, guided potential home buyers to buy in specific neighborhoods based on their race. Black home buyers were recommended majority-black neighborhoods, while white home buyers were recommended majority-white neighborhoods. These findings were even more pronounced in more racially segregated cities, such as New York City and Chicago.
Before this technology is widely adopted, lenders and borrowers will need to have trust in the companies developing it and the results it produces.
Many lenders will be slow to adopt until stronger regulations come into place
“The mortgage industry is a highly regulated industry at this point, so I think it's normal for things to move over years rather than months,” Heck said.
Government agencies have begun to issue some guidance on how generative AI can be used in the housing sector. In September 2023, the Consumer Financial Protection Bureau (CFPB) issued a statement clarifying that lenders who reject borrowers based on credit must explain why.
“Technology marketed as artificial intelligence expands the data used to make lending decisions and the list of reasons a loan may be denied,” CFPB Director Rohit Chopra said in a statement. “Creditors must be able to provide specific reasons for denial, and there is no special exception for artificial intelligence.”
Additionally, the Department of Housing and Urban Development (HUD) released guidelines in May 2024 stipulating that lenders must comply with the Fair Housing Act when using AI or algorithms in advertising.
Generative AI is still in its infancy.
More government leadership and regulation is needed for AI to be more widely adopted, primarily because most loans must meet certain standards to enter the mortgage market.
“The flip side is that every loan gets sold and securitized, and even if a bank buys the securities, there are minimum requirements that they have to meet,” Seibel says. “So even if the AI says, 'I don't need a pay stub; I know this person has a job,' the bank still needs that piece of paper to buy the loan as part of the securities. Until some of this decision-making is accepted at the end of the loan destination cycle, I think adoption of applicable locations will be limited.”
When taking out a mortgage, you still need a human touch
While technology has become a big part of the mortgage process, many people still want to speak to a real person when taking out a mortgage.
“The reality is, this is probably the biggest transaction you'll ever make, and you want to talk to people, especially if you're a first-time home buyer,” Seibel says.
“I still think that people generally want and trust the human element,” Heck adds, in large part because buying a home and getting a mortgage is an emotional process, he says.
For Rocket Mortgage's Zook, AI can be used to free up the human element.
“We've found that the best use cases are helping humans do what they're good at and helping computers do what they're good at,” Zook says.
While AI can handle data entry and find patterns within the data, human loan officers are better at guiding borrowers through the process.
“In all the use cases that I mentioned, we always have humans involved. We're not using AI or automation in any kind of lending decision that goes into it,” Zook said.
At the end of the day, loan approval still needs to be done by humans, and that doesn’t seem like it’s going to change anytime soon.
Visit Bankrate online at bankrate.com.