Mortgage lending is a data-intensive business. That’s obvious to anyone who has applied for a home loan. Among the documents you need to hand over are two years’ worth of tax returns, W-2s, bank statements, copies of identification plus any name changes, proof of other income, and so on. As laborious as it is for borrowers, all that paperwork requires much time and effort on the other side of the lending process.
Mortgage lenders must assimilate, sort, evaluate, and weigh all that information to ensure the loan will be repaid. There’s a lot at stake at an average mortgage of $260,386.
That’s why I was intrigued by a recent survey of mortgage lenders and their use of artificial intelligence (AI) and machine learning (ML) in lending. The survey found that the use of these technologies in the mortgage industry will boom over the next two years. And that will be a good thing for borrowers and the market.
Fannie Mae, a federally sponsored agency that backs mortgages, surveyed senior mortgage executives at 184 lending firms on their interest in AI/ML last year. The survey found that while nearly two-thirds of lenders are familiar with AI, only about one-quarter (27%) are using it in their businesses now — and only half of that group are currently using it with customers (the rest are doing trials).
Yet when asked to look ahead two years, 58% of lenders expect to use AI/ML in their mortgage business. Of the rest, 22% predict they’ll investigate AI, and 19% foresee being in a “wait and see” mode. Only 2% of lenders stated outright they wouldn’t be using AI. The divide between the low number of banks using AI/ML right now and the number who expect to deploy it in two years shouldn’t be a surprise.
AI is just starting to make inroads. While AI voice assistants such as Alexa and Siri may be part of our everyday world, more sophisticated AI is only beginning to make its way into industries such as health and finance, where there’s more at stake than your Alexa adding toothpaste to your shopping list. With financial information and loans, lenders and regulators rightly have an explicitly regulated duty to ensure these new technologies are used appropriately. AI and ML systems need to be effective.
They must also be transparent about deciding why applicants are denied or get different mortgage rates than others. But AI and ML still have enormous potential for the mortgage space. For one, much effort goes into organizing and reading all the paperwork. Analyzing those documents can be managed far more efficiently with computers, freeing lending professionals to use their experience and judgment on individual applications—a % of lenders, 42%.
I told Fannie Mae their primary objective with AI/ML is to improve operational efficiency. Nearly as many, 41%, expect these technologies to enhance the borrower experience. For instance, while you could expect to wait about three weeks for a mortgage application approval, AI will probably reduce that to one day within a few years.
Bankers also envision smart online applications where interview questions are responsive to your specific answers. These methods shorten the basics of the process to focus more on your qualifications and needs. The bigger picture is that AI/ML can better evaluate the ability of people to repay their mortgages. This should open up loans to a wider variety of people who are good bets to repay a mortgage but, for whatever reason – typically lack credit history – would be rejected today. ML can take thousands of data variables.
Through its advanced mathematics, it finds hidden correlations that better indicate whether someone will repay their loan. All that insight allows lenders to view borrowers as more than the handful of numbers in traditional lending models. AI/ML gives them a fuller picture of applicants as people. And history suggests lenders need better tools. It’s worth remembering that at the height of last decade’s mortgage boom, several lenders with a high volume of loan requests streamlined their process down to using just one variable – three-digit credit scores – to decide on applications.
While the financial crisis that followed had many complex causes beyond that fact, relying on one useful variable as the sole determination of risk was a mistake. Experience with ML shows that people with the same credit score can be, in fact, less (or more) risky than each other when additional pieces of information are used in the lending process.