PointPredictive tool uses AI to spot income discrepancies in loan apps

Register now

Even as mortgage loan application defects continue to decline, data inaccuracies related to income and employment are making up a greater share of discrepancies found in loan files.

Income and employment made up 28% of the post-closing quality control critical defect findings in the first quarter, up from 20% in the fourth quarter and 25% in the first quarter of 2018, according to Aces Risk Management.

Aged, missing or unsatisfactory income documents produced nearly two-thirds of those first-quarter defects, while another 27% resulted from a mismatch in the income-to-employment calculation. While not all income or employment information defects in a loan file are because of fraud, they are a red flag that something is amiss.

A growing number of technology vendors are looking to artificial intelligence, machine learning and other advanced technologies to help lenders with the underwriting process, including detecting potential problems with application data. For example, Freddie Mac began working with ZestFinance, a developer of AI-driven consumer-risk modeling software, which could expand access to credit.

At September's National Mortgage News Digital Mortgage Conference, 46% of attendees said AI would have its greatest impact on changing the underwriting process.

Most recently, PointPredictive has rolled out IncomePASS, which uses machine learning to look at the applicant's employer, occupation, job title, residence and estimated years of experience to determine his or her likely income. That can then be matched with the submitted data. The company's testing gives IncomePASS a 90% to 97% accuracy rate. The lender is able to use the findings to do further data checks as needed. Most applications can be sent through a streamlined process.

"This is not a voice or manual verification of employment or income," said PointPredictive CEO Tim Grace. "It uses sophisticated machine learning algorithms that are highly accurate in order to validate that the income is stated is correct."

Lenders need to understand where their income risk is and where it isn't, so they can route the high risk stuff to higher level underwriters and the low risk stuff can be streamed right through the process, added Frank McKenna, PointPredictive's chief strategist.

It is available for mortgage, automotive and small personal loans. However, the product is not designed to replace the verification of income statement or other compliance checks.

"What we are enabling lenders to do is actually streamline their underwriting process, so we are able to let them know when there are pieces of information that are OK so that they do not have to do a forensic due diligence on 30% to 50% of their loans for income," Grace said. Those files can be sent to a junior underwriting team; other files with suspected income misrepresentation can be routed to a more sophisticated forensic underwriting team.

IncomePASS analyzes stated income information from the mortgage loan application. It doesn't study any bank account documents or pay stubs.

PointPredictive did a cross-industry study where it looked at validated incomes versus what the borrower stated and for "from 15% to 30% of the applications have falsified income," McKenna said, adding that for the remainder "the income was completely reasonable and correct."

IncomePASS is a separately available feature of PointPredictive's MortgagePASS product. MortgagePASS uses machine learning to streamline mortgage application workflow and have less friction in the lending process.

Nearly half of the mortgage applications have extremely low risk. "So all loans should not be treated equally in the underwriting process. What we help lenders determine is what that 50% of loans are, what are the features that make those loans not risky," he continued.

For reprint and licensing requests for this article, click here.
Digital mortgages Mortgage fraud Compliance Underwriting Originations Machine learning Artificial intelligence