Among the individual components operating in the background of most artificial intelligence (A.I.) systems is predictive analytics. The process has proven effective in numerous other business processes and is a driving force behind many of A.I.’s capabilities.
Predictive analytics has been used particularly heavily in the credit scoring, underwriting and appraisal processes.
Blend Enterprise Compliance Manager Austin Miller and John Vong, executive chairman and founder of ComplianceEase, explained to Dodd Frank Update how predictive analytics has evolved and its value in the mortgage business. Vong noted that Fannie Mae and Freddie Mac are among the largest institutions currently using the technology, particularly when handling large refinancing volume.
Credit scoring, appraisal applications
Predictive analytics are used to provide an objective outlook to enable A.I. systems to make reliable decisions about available data. It also can be used to expose faults with current methods of evaluating credit applicants.
Miller pointed to a study by FinRegLab.com that explored issues that exist in common credit scoring models.
A common practice in determining an applicant’s creditworthiness that has proven problematic in some cases relates to the use of consumer credit reports, Miller explained. Relying too heavily on those reports has been found to create impediments to credit access disproportionately affecting certain groups of borrowers. The solution to that issue may lie in using other data to determine a borrower’s ability to repay a loan.
The FinRegLab.com research showed that “cash-flow” data was a more effective predictor of a borrower’s creditworthiness than their credit score.
“The study found that taking in bank account data and using that to determine whether a person has enough cash-flow to support monthly payments actually has profoundly predictive powers in determining a customer’s creditworthiness,” Miller said. “It’s similar to the predictiveness of credit reports but hasn’t demonstrated the same level of biases that have traditionally shut out other groups from credit products that traditional credit reports have proven to have done.”
The study found that the cash-flow data can be at least as predictive of as traditional credit scores. Pertaining to borrowers who have limited or non-existent credit histories, cash-flow data can provide meaningful predictive power that, by extension, can help borrowers with limited access to credit. The study further explained that cash-flow scores and attributes seemingly improved lenders’ ability to predict credit risk among borrowers that are scored by traditional systems as presenting similar risk of default.
Cash-flow information can be used effectively either in combination with, or in place of, credit reports in a company’s predictive analytics model to make such determinations to better serve community credit needs, Miller suggested.
Miller noted that, according to the study, when divided into subgroups based on likely race, ethnicity and gender, the degree to which cash-flow data predicted credit risk was relatively consistent.
“Moreover, when compared to traditional credit scores, the cash-flow based metrics appeared to predict creditworthiness within the subpopulations at least as well as the traditional scores, and better in selected cases,” the study found. “Overall, the cash-flow data appeared to provide independent predictive value across all groups, rather than acting as proxies for a demographic group.”
Miller detailed steps Blend has taken to mitigate fair lending risk and to facilitate a more efficient origination and underwriting process for the benefit of all parties involved.
“We actually analyze the types of document requests that are made of borrowers,” Miller said. “We use that analysis to determine borrowers who have similar characteristics so we can surface document requests to that borrower to ensure they are consistently applied, earlier on in the process. This enables them to sort of speed through the underwriting process.”
By recognizing the many similarities that occur in the compliance process and delegating those to an A.I. system, rather than having a human perform a manual “stare and compare” on pages and pages of documents, the process can be made much more efficient and accurate.
“There are so many different types of paperwork with similar data points that need to be compared to one another and assessed. I think that the combination of OCR technology with A.I. that actually analyzes those datasets is going to be really important to focus on paperwork reduction and improving efficiencies in the industry,” Miller said.
Valuation applications
Similar to how predictive analytics is used to evaluate historical data to make determinations about a borrower’s creditworthiness, Vong noted it also can be used in appraisal underwriting.
“With credit scores, they use predictive analytics to look at a consumer’s credit history and attributes that predict how they will behave in the future,” Vong said. “For instance, an AVM (automated valuation model) looks at historical and current data in an area and predicts the value of a property. But in compliance, it’s pretty dry. It’s black and white. So we don’t use predictive analytics for that.”
By examining data gathered from various relevant sources, predictive analytics allows an A.I. system to make the types of determinations appraisers currently make, but in a fraction of the time.
“AVMs look at similar comparable properties with similar square-footage figures, location, and similar prices and try to predict their value,” Vong said. “When a homeowner goes to a real estate search engine to find out the price of a house, search engines will use comparable data and several models to predict the value of that home.”
Because such technology makes determinations about property values absent an actual visit to the property, many appraisers are quick to note that there are certain anomalies that such technology cannot account for that can have a major impact on the value of a particular property.
Marketing applications
Vong noted that lenders often use predictive analytics in marketing, particularly with regard to refinancing and debt consolidation. He explained that by looking at a borrower’s credit history, a lender can consider a person’s credit rating, their personal debt and other factors and determine whether they are in a suitable position for a refi.
AVMs also come in handy in determining whether consumers have sufficient equity in their homes and may be a candidate for specific refi products.