As the amount of information held at mortgage companies continues to accumulate, many within home finance have found a business case for bringing more data science into the picture.
By applying advanced analysis techniques and models, data scientists today are “proactively helping you find those needles in the haystack” while providing appropriate context, according to Jerry McCoy, executive vice president, performance management, of subservicer LoanCare.
“We’ve run the models. We have this nugget of insight,” said McCoy, who previously held leadership roles within technology and servicing companies, such as IBM, Nationstar, Fannie Mae and JPMorgan Chase.
“What we also do as part of data science is telling the story with it.”
LEARN ABOUT MAKING DATA ACTIONABLE AT DIGMO 2023
Some companies are already on the bandwagon. In 2023 research conducted by Arizent, three out of five mortgage industry professionals said they had tried using data-science tools or applications of some form in the course of their work. For customer acquisition purposes, 36% indicated they were leveraging it aggressively, with 85% of the subset expressing satisfaction with outcomes. But there’s a clear divide between depository institutions and nonbanks in their relationship to data science, with the former three times more likely to be utilizing data science than their smaller and less capitalized IMB counterparts.
Still, as mortgage companies build up digital capabilities, many view data science professionals as welcome additions to their strategies, relying on their expertise to achieve greater operational efficiencies while moving them closer to a streamlined lending experience.
“The possibilities are really phenomenal, both from an operational standpoint, to new solutions that can be offered to the market filling a number of the clunky gaps that exist in the mortgage lifecycle today,” said Julian Grey, executive vice president of data and analytics at Black Knight.
Currently, professionals using data science and modeling can be found working across mortgage-related segments, including sales, underwriting, servicing and secondary markets. The term itself covers a wide spectrum of potential uses, ranging from simple graphs to generative, large language models applying artificial intelligence, Grey said. Its potential, though, is increasing with the surge of interest in AI.
“We can start allowing our clients and users to use tools like natural language processing, to use tools, like really working with interfaces that are intuitive about their needs,” she said. “And we can also start doing things with complex datasets like photographs and documents.”
While many of the methods used have been available to some degree for decades, two landmark 21st-century events helped underscore the necessity within the home finance industry for quick processing and modeling of datasets on a wider scale: the Great Financial Crisis and the COVID-19 pandemic.
The housing crisis illustrated the need for precise information, such as price indexes narrowed to specific ZIP codes or prepayment default models on the loan level. “The need for that detailed, granular information really started jettisoning the whole movement of saying ‘Let’s be smarter about our information,'” Grey said.
The onset of COVID-19 presented a new challenge. “All the sudden there was a need in the marketplace to understand both forbearance and prepayments at a faster rate than we had done before,” she added, noting the significance of Black Knight being able to produce daily updates during the height of the pandemic, which previously came out monthly.
That power to quickly ingest and analyze large loads of data has become more critical to many business clients, who require quick turnarounds in their decision making, such as insurers, said Ari Gross, CEO and co-founder of True, whose platform uses artificial intelligence in processes involving data extraction and document recognition to serve lending and tax industries.
“If you can’t figure out certain things in minutes, you’ll lose that opportunity, particularly in the mortgage insurance space,” Gross said.
The capability of tools now used by companies like True, who may need to examine up to thousands of pages in a loan document, has come a long way in a few years and illustrate the role science and technology play to improve processes, particularly when reading and verifying data against other sources.
Where Gross once worked with companies using machines with single-core processors 20 years ago, “now we can put algorithms on those machines and run to 300 core,” he said.
“I’ve got 300 processors trying to understand an 1,800-page load. You can start doing stuff you could never do before,” he added. The tools are constantly learning as well, based on both employees’ input and the corrections being made to their findings, leading to improved accuracy.
“For example, a year ago, you could easily come up with many, many documents where there were whole sections of it we couldn’t do data extraction on.” Flaws ranged from excessive data “noise” to instances where a client’s handwriting inadvertently made information unrecognizable to the machine.
“Today, we can probably get into almost all of those, and that’s in the last year,” Gross said.
Data science is also paying dividends for the servicing industry, allowing companies like LoanCare to study behavior based on queries coming from up to 1,000 borrower phone calls a month
“We then start to understand interactions,” McCoy said. “These may not seem like they’re big data science aspects, but it means a great deal because you can then start to understand are there mechanisms that we could do.” The analysis can lead to business practices that better serve clients and borrowers after scientists “connect the dots.”
“We turn that into business context. How do we then turn around and provide more information, maybe on our website?” McCoy said.
“Fundamental to data science is understanding probabilities of outcome. And how do those probabilities of outcomes change as things adjust?” he explained.
For a company like business-purpose lender Kiavi, which works with investors that may purchase multiple properties each year, data can similarly help them identify risk in much the same manner, but down to the individual borrower level. As a business with many repeat clients, Kiavi can tap into the wealth of data on them based on past transactions.
“Since we’re servicing the loan with information across multiple dimensions, we can identify certain patterns in the data that can indicate a potential for higher propensity to default,” said Kiavi CEO Arvind Mohan, who used his training as a computer engineer in leadership roles within banking and fintech worlds, including at Barclay’s.
“If we identify the variables and create the right factors that indicate a higher probability of risk, then you can actually design your default and servicing pipelines in a more efficient manner to address those cases first,” he said.
Data science also provides greater nuanced market information for Kiavi, whose business Mohan described as “fragmented and hyperlocal in nature.” But it is the type of information that might benefit residential lenders and real estate agents as they look for customers.
“In some sense, we’re using data science like a local lender that lives with that market,” Mohan said. “What we want to do is bring a much more scalable experience, but ensure that we can be as quick as a local market lender.”
But large amounts of information means little without quality interpretation behind it, which is where the skills of data scientists come into play the most.
“There’s art in here, and that’s the real part that most folks kind of miss a little bit,” according to McCoy.
“If the business doesn’t know how to operationalize it and can’t get their head around what it means, then it’s not going to do anybody any good,” he said.
