Why mortgage lenders should start looking harder at AI, big data

Register now

Mortgage companies have been slow to adapt as secondary-market buyers, vendors and other types of lenders embrace advances in artificial intelligence, big data and methodology used in underwriting.

For the most part, they've gotten away with it. The evolution of the secondary market has been a long, drawn-out process and alternative lenders have had plenty of other categories of credit to chase.

But 2020 could provide a significant reality check for the go-slow movement. Lower-than-expected rates have bolstered activity in the segment and nontraditional lenders have taken notice. In addition, the dominant government-related mortgage investors are starting to modernize their underwriting in ways that will likely put competitive pressure on the entire ecosystem.

As a result, established players across the industry could find themselves at a competitive disadvantage to newcomers when it comes to capitalizing on new business prospects, managing risk and operating efficiently.

Those mortgage lenders seeking to stay apace will need to understand and act on some fundamental forces driving change if they are to preserve their own relevance in the coming years.

Faster decisioning

Operational costs are still relatively high in mortgage lending and consumers increasingly expect a relatively quick decision when they apply for a loan, making quicker turnaround among the most immediate and urgent matters of attention.

Matt Komos, a vice president in TransUnion's research and consulting division, is among those who see consumer expectations as first among the questions lenders need to address.

"This is my perspective, not necessarily the TransUnion stance, but I think continuing to figure out how to answer the consumer’s needs is the answer," he said. "The consumer has gotten used to a world where I can order food on my phone and get it in 20 minutes, I can have a car come to me instead of hailing a taxi, and I can order through Amazon today and have it tomorrow."

Mortgage decisions aren't instantaneous at this point — and perhaps they shouldn't be, given that many borrowers still need some consultation. But processes supporting those decisions are moving faster thanks to new forms of workflow automation and improved electronic access to bank data and other information.

Workflow management tools that streamline processes by collecting consumer-authorized data used in underwriting, and routing of loans in response to it, are the first step to take in upgrading underwriting. It's an area where insurgents may currently have the edge.

"Some of the new fintechs are looking at data aggregation upfront before getting into fulfillment or underwriting. Being able to do that at the front end allows you to streamline operations," said Michael Grad, a senior partner at Stratmor Group. "If you know the complexity of the loan at the start of the process, you can send that loan to the right part of the factory that processes less-complex loans more efficiently."

"The most efficient way we can receive that data and analyze it, that's what we're looking into," said Tom Hutchens, executive vice president of production at lender Angel Oak Mortgage Solutions.

Broader collection and use of data

Another compelling reason to update underwriting technology and data is that it has helped lenders approve a broader range of borrowers.

"I think this is a big topic among lenders of all shapes and sizes right now, and it's really about expanding the credit profile served to capture additional market segments," said Joey McDuffee, vice president at mortgage technology firm Blue Sage.

Existing mortgage underwriting models can consume at most 15-20 variables. The latest AI-driven versions expand that number and in doing so more readily incorporate larger alternative data sets. In some cases, that expands the potential approval pool.

"AI is better math that can be used to find people around the margins that lenders ought to be saying yes to if they took a more comprehensive view," said Kareem Saleh, an executive vice president at fintech Zest.ai. "What we find is that lenders can increase their approval rate by 10% pretty quickly by just parsing the edges more finely."

The idea that more borrowers could be underwritten if broader sets of data could be considered in models isn't new, but it's gaining considerable momentum across consumer finance.

"Fintechs have successfully done it in the unsecured personal loan space and now we're starting to hear about home equity lines of credit," Komos said. "They're also starting to dabble with auto refinance. So I think there's an opportunity for mortgage lenders to learn from what they're doing. It could definitely be a way to bring more consumers into the mix."

To get a sense of how the use of advanced data and technology can make a difference in a lending sector, consider this: Fintechs in the personal loan market are now responsible for 40% of originations today, up considerably from 6% in 2010, according to Komos.

While the mortgage market is far larger, more beholden to government-related entities and more complex than the personal-loan sector, key stakeholders there also are starting to recognize that alternative data is a reality they must contend with.

"As the consumer continues to show an interest in those things, we'll see different asset-classes move toward them to help meet the needs of the consumer," Komos said. "There's a lot of regulation and homes are a big purchase for consumers, so mortgage lenders are going to be a little more hesitant, but they're also beholden to Fannie Mae and Freddie Mac. It could be a matter of getting the GSEs on board."

At this point, both Fannie and Freddie have tested and are weighing the use of more advanced underwriting data and technologies.

In addition, traditional credit reporting and scoring vendors that serve the mortgage industry have been steadily working to incorporate more alternative data and advanced technology into their product lines.

"We've been working for a number of years to figure out how to get a consumer's bill payment history beyond traditional credit relationships with things like utilities your mobile phone and your cable bill. That's the type of data while it's used by a lot of different consumers hasn’t been factored into a credit score before," said Michele Bodda, general manager for Experian Mortgage.

In addition, traditional credit-score provider FICO also is working on adding new sets of data to its risk model.

"We're looking at alternative data or data outside what the three main credit bureaus typically provide to gain additional insight into positive consumer repayment behaviors that exist, but just haven't shown up in the traditional data," said Joanne Gaskin, a vice president at FICO. "We don't really see a problem with that, as long as the decision and the score to be used are based on true risk assessment."

More advanced risk models

As the financial crisis demonstrated, opening up the credit box ultimately doesn’t help lenders if the incremental loan originations wind up defaulting at substantially higher rates.

So it's imperative that new underwriting technology and data are designed to improve risk assessment. Evidence that they do has been promising to date.

"Artificial intelligence certainly has the ability to find additional predictive lift from looking at existing data as well as finding new correlations," said Craig Focardi, a senior analyst at Celent.

The latest artificial intelligence-driven alternative-data models may not have been around long enough to conclusively demonstrate how predictive they are when it comes to long-term performance, but regulators are starting to take them seriously.

"The use of alternative data in a manner consistent with applicable consumer protection laws may improve the speed and accuracy of credit decisions," five federal regulatory agencies said in a joint statement issued in December 2019.

This admittedly cautious regulatory endorsement was nonetheless a noteworthy breakthrough. Concern that newer risk models used for underwriting might raise compliance questions has been a big concern, in part because it was unclear supervisory agencies would be open to the new approach. The interagency statement suggested regulators are getting there. Mortgage lenders need to, too.

Get ready to grow — or miss out

The housing finance industry needs to start laying the groundwork for new AI-driven underwriting models that can digest larger amounts of alternative data because government-related agencies and large investors are starting to embrace these strategies. In addition, emerging homebuyers increasingly demand a quick, digital turnaround on credit decisions and have nontraditional credit and income histories. Meeting this demand has expanded lending in other parts of the consumer finance market and it could help mortgage lenders grow their business volumes, too.

That growth is going to be accessible to those in a position to pursue it, and the amount of business available to those who fall behind will only get smaller as new technologies and access to new types of data advance.

For reprint and licensing requests for this article, click here.
Artificial intelligence Underwriting Big data Data and information management Risk management Fintech Law and regulation GSEs