The mortgage application process is a bit like assembling a pie: Both require several ingredients and steps, and take longer to prepare than anyone cares to think about.

Mortgages come with reams of paperwork, and strict industry standards determine how chunks of information are to be treated. Sleep on the rules for either, and that new home or a sumptuous dessert will seem unattainable.

While some pieces of the mortgage information pie, like appraisals, are procured on demand, others — like data on income levels, credit scores, and existing debt — may already be in the data mix, but require retrieval.

This information could be siloed in storage by credit reporting agencies, for example, but can later be retrieved in the back office: This is where automation is pitching to lenders.

“Though information is at the end of it, if we can really manage the information and data more clearly, we can control processes, be more consistent, and we can automate,” Nate Longfellow, head of digital, home lending at Wells Fargo, told Bank Automation News.

Although the industry has been increasingly interested in leveraging data, legacy systems are built around people managing decisions in the process, Longfellow said, and therefore require systemic changes to support the shift toward automation.

While the pandemic shuttered most back offices last year, demand for mortgages continued to grow. Amid interest rates as low as 2.17% for a 15-year, fixed-rate mortgage in December 2020, new home applications saw a 28% year-over-year uptick, and refinancing applications jumped by 102% on an annual basis, according to December 2020 data from the Mortgage Brokers Association.

“The pandemic really highlighted a lot of inefficiencies in companies,” said Craig Le Clair, principal analyst at research firm Forrester. “In a somewhat negative way, it exposed humans as being kind of friction in the process.”

And although interest rates have risen in the past few weeks, they have not yet dampened demand among buyers as loan applications have continued to rise. This rising demand has further accentuated the need to implement robotic process automation (RPA) and other automations to deal with application volumes.

Volumes depend on external factors, said Matthew Lautz, chief executive of Neostella, an RPA solutions and consulting service that partners with bot vendor UiPath. The surge in demand for refinancing inspired by low interest rates has “created a really big visibility within the mortgage sector on the need to have automation to handle these varying flows without having to change their headcount,” Lautz told BAN.

Efficiency and scale

Automated mortgage procedures can be broadly split into two buckets: document processing and data gathering — which can be carried out by RPA bots — and decisioning, which involves underwriting and utilizes more complex techniques like machine learning (ML) and artificial intelligence (AI).

RPA can automate the various pieces of information that must be ordered with each application received, Lautz said. Even something as simple as ordering flood insurance can be automated, he added, and noted that “every lender will have a story” of forgetting to order flood coverage and thereby delaying a mortgage closing.

Some processes only save 15% or 20% efficiency with automation, said Lautz, but when that’s a process that is repeated 10,000 15,000 times a month, “that can still be a very nice return on investment.”

“Across the board, the average mid- to large-complexity mortgage process will have between a six and eight-month return on investment [horizon].”

While RPA can help lenders scale basic rules-based tasks, the more complex processes, like automated underwriting, involve ML and AI. Compared to its enthusiasm for RPA, the mortgage sector has adopted a more cautious approach toward automating such tasks. But, both the Federal National Mortgage Association (Fannie Mae) and Federal Home Loan Mortgage Corp. (Freddie Mac), the largest guarantors in the market, have already taken the leap toward automated underwriting.

In November 2020, Freddie Mac announced it had partnered with software maker Zest AI to try out its platform for credit underwriting, with the hope of expanding access to credit without substantially increasing risk. Founded in 2009, Zest AI has raised $232 million in funding from investors like Insight Partners and Chinese tech giant Baidu, according to Crunchbase. The firm said that $150 million was returned “unused,” therefore Zest AI has effectively raised $80 million as of March 31.

Zest AI uses data that “customers already have on hand, [but] just because of machine learning, we’re able to deploy better math and be able to consume more data,” Mike De Vere, the company’s CEO, told BAN. Instead of looking at indicators like debt-to-income ratio at a given point of time, Zest’s model looks at the ratio through a machine learning model across 12 to 18 months, which provides a more comprehensive outlook of the borrower.

This model also continuously monitors the data for changes to accommodate shifts in the macroeconomic environment, like the government sending stimulus checks to its citizens, for example. “All of our customers are asking not only for monitoring the model in production, but also understand the economic impact in production,” De Vere said.

Wells Fargo, on the other hand, has taken a more measured approach toward mortgage automation. “We are just in the front door of leveraging data to drive process,” said Longfellow. He added that the bank has been experimenting with automation by trying out proofs of concept, but the implementation is still in its infancy. “I’d say [in about] about six weeks, we’ll have our kind of refined roadmap and target state,” he said.

Longfellow called the decision for Fannie and Freddie to use automated underwriting tools “great advances.” because the more such secondary buyers of mortgages use the technology, the more they “advocate to enable the leveraging of data versus, friction documents or other types of manual interpretations.”

Fannie Mae and Freddie Mac declined to comment on their automation for this story.

The hybrid model

Automation techniques currently can be applied to tasks that follow rules, but not necessarily a lot of context. So, while tasks like inputting data, classifying it, and moving it from one system to another can be performed by bots, others — like underwriting and appraisal — tend to more tricky.

“This industry … relies a lot on intellectual experiences and knowledge to process and underwrite a loan,” Longfellow said. “As an example, I don’t see the underwriter going away.”

While market liquidity providers like Fannie Mae and Freddie Mac can rely on automated underwriting because they don’t directly originate loans, lenders like banks and credit unions are responsible for vetting the information’s veracity, and hence rely more on manual processes that so far cannot be automated.

“How do we make sure what’s in that data file that goes to the [government sponsored enterprise] is accurate and true data that’s substantiated by the right documentation?” said Longfellow. “That’s kind of been the pursuit that we’re all on.”

Many banks — and some credit unions — have recently partnered with mortgage automation fintechs to automate lending procedures, largely those that involve processing documents. U.S. Bank, Wells Fargo, BMO Harris Bank and Navy Federal Credit Union have all announced partnerships with Blend, a fintech platform that helps lenders digitize pieces of the mortgage process, allowing borrowers to complete them online.

While the current spike in demand has increased demand for bots to extract and process information, eventually the costs start to tick up. “A lot of the vendors have been pushed to provide more of a consumption-based pricing that was based on utilization,” Le Clair said, adding that vendors are told by customers that they “only want to pay when the bot is actually doing something.’”

On average it takes about 45 to 60 days to close a mortgage. In August 2020, it took about 51 days and that increased to 58 days in January 2021, according to data from Ellie Mae. In many ways, this timing is indicative of the multitude of moving parts involved. . Automation is entering the sector, but its progress has been deliberate.

“How do we make decisions and automate actions in a consistent, repeatable way,” Longfellow said. “If they truly were doing that, the mortgage wouldn’t cost $9,500 for us to originate — and that’s an industry number.”

While the cost of originating a mortgage — and the time it takes — may provide ample opportunity to automate processes, some current applications have also been related to cutting costs at a time of economic uncertainty spurred by the pandemic. But while this may have triggered the push to automate, analysts say the trend is here to stay.

“It’s easier to buy RPA, as compared to updating your core systems,” Nicole Sturgill, an analyst at research firm Gartner, told BAN. Efficiency is the main proposition for companies looking at these techniques and, after implementing it in a few processes, “now they’re scaling it through the organization,” she added.

The mortgage automation process is likely to be one of “progressive replacement,” at the outset, instead of a new end-to-end design, Sturgill said. But speeding up tasks at scale isn’t completely free from risk, she cautioned. Identifying exactly which tasks to automate, and how, will be imperative to prevent bots from changing a “slow, bad process to a fast, bad process,” she said.

The story was updated on April 1, 2021 to reflect the change in funding raised by Zest AI.

Bank Automation Ignite, on April 13-14, is the event for inspiring automation initiatives and investment in financial services. At the virtual event, financial services professionals can discover new use cases and technologies that are accelerating automation in banking. Learn more and register at www.BankAutomationIgnite.com.