DSNews delivers stories, ideas, links, companies, people, events, and videos impacting the mortgage default servicing industry.
Issue link: http://digital.dsnews.com/i/1517420
MortgagePoint ยป Your Trusted Source for Mortgage Banking and Servicing News 48 March 2024 F E A T U R E S T O R Y is continuously verified and validated from loan application through close and secondary market sale. Both compliance and scalability can be addressed with this approach. Understanding the Source of Defects A s the housing market becomes in- creasingly challenging for borrowers trying to overcome the impact of higher rates and tighter credit, the chances of loan defects multiply. Currently, most loan defects stem from discrepancies in income verification, appraisal and valuation incon- sistencies, and errors in the underwriting process. Inadequate documentation is another significant contributor, often because manual practices lead to human error and increased likelihood that critical details may be overlooked. No matter the origin, however, every defect becomes a ticking time bomb, threatening to explode in the form of a re- purchase demand from the GSEs or other secondary marketing partners. Failing to address these defects in a timely manner, regardless of investor, is not limited to im- mediate financial losses, either. They also create long-term consequences, including reputational damage and increased regula- tory scrutiny, which can collectively derail a lender's future. The question then becomes, how can lenders plug these loopholes effectively and efficiently? This is where artificial intelligence (AI), machine learning, and business outcome automation technol- ogies come into play. In fact, these tools have already become game-changers for lenders that have embraced them. Machine learning algorithms are adapting and learning from millions of structured and unstructured loan doc- uments, becoming vastly more accurate in the classification of documents from which data is extracted. Meanwhile, business outcome automation solutions are increasingly being used to scrutinize vast amounts of loan data and identify anomalies or inconsistencies that typically lead to defects. This verified data can then feed automation for income calcula- tion, loan processing, loan quality, and other origination processes prior to close. Rules can indicate if a test is true or false, compare values, use fuzzy logic to make determinations beyond exact match data, and validate loan file data and documents against guideline requirements. More advanced uses of AI open up a world of even more interactive and self-learning applications. Use of verified data in consumer chat applications can enable the lending process to be more engaging. AI analyses of borrower data can aid underwriters in making better risk decisions. And fraud detection tools can incorporate AI to evaluate mortgage applications for possible illegal activity. All represent greater efficiency, but they must also strike the right balance of transparen- cy and automation that does not introduce bias or unfair lending practices. Aided by industry and technological know-how, these tools are becoming increasingly effective at pinpointing poten- tial areas of risk with every new transac- tion. This high level of precision not only reduces the probability of repurchases, but also significantly expedites loan process- ing by prepping files for underwriting and the quality control (QC) audit process. Compared to human staff having to slog through countless hours of man- ually reviewing loan documents and keying in data from documents, machine learning, and automated tools can "green light" complete loan files while creating shortlists of exceptions within a fraction of the time. This not only speeds up loan processing and pre-funding QC checks, but also frees up human resources for decisioning tasks and minimizes any defects that might surface in post close audit reviews. Machine learning algorithms are adapting and learning from millions of structured and unstructured loan documents, becoming vastly more accurate in the classification of documents from which data is extracted.