This piece originally appeared in the March 2024 edition of MortgagePoint magazine, online now.
Taking advantage of machine learning and other emerging technologies can streamline loan production, loan acquisition, and QC processes
The old saying, “when it rains, it pours” perfectly captures the current mood among mortgage originators. Not only has loan volume fallen significantly over the past couple of years, but mortgage lenders are also dealing with a storm of new quality control requirements—including new repurchase demands from Fannie Mae and Freddie Mac.
Unfortunately, tighter guidelines for pre-funding audits and faster post-closing reviews from one of the GSEs come at a moment when lenders are already struggling with thinner profit margins. In fact, according to the latest data from the Mortgage Bankers Association (MBA), independent mortgage bankers lost an average of $1,015 per loan during the third quarter of 2023. This situation has created a precarious balancing act that has lenders reassessing their business operational strategies and looking for anything to tip the profitability scale back in their favor.
Yet, necessity often proves to be the mother of invention. For lenders willing to adapt, there are ways to navigate today’s turbulent market safely and comply with GSE and investor requirements. And few tools are proving more effective than AI-powered business outcome automation.
How We Got Here
Since the 2008 crisis, the mortgage industry has undergone a significant transformation to meet ever-changing regulations and expectations involving loan quality. Most recently, it’s been Fannie Mae and Freddie Mac that have tightened their quality control (QC) requirements, pushing for more stringent audits and for lenders to evaluate loan quality sooner rather than later.
However, while these new stipulations have created a hill to climb, they also represent an opportunity for sellers/servicers to leverage automation to verify and validate loan quality from loan application through close and secondary market sale.
For example, Fannie Mae now requires pre-funding QC audits on at least 10% of loans or up to 750 loans per month sold to the GSE. Lenders must also complete post-closing reviews in just 90 days, compared to the previous 120-day window.
At a time when many lenders are already wrestling with rising loan defects and heightened fraud risk, these new requirements are timely but may also be straining already limited operational capacities.
The GSE’s decision seems to stem from concern over the quality of the enormous amount of loans originated during the height of the pandemic, when lenders were dealing with massive capacity issues.
Indeed, the agencies’ noticeable uptick in repurchase requests reflect that, but may also be an indication that both are concerned about counterparty risk and the financial stability of some institutions in the current market. Going forward, seller/servicers need to uncover quality issues sooner and pay meticulous attention to a host of variables in their loan production processes that could potentially trigger a defect or repurchase demand.
As expected, many lenders have had to cut back on staffing amid lower origination volumes. In fact, according to the MBA, the average number of production employees for independent bankers fell from 362 to 331 between the first and second quarter of 2023. While a necessary expense measure now, cutting staff and then rehiring when market volumes return should be a way of the past.
Instead, there is an opportunity for sellers/servicers to leverage automation much more pervasively across their production processes. This way loan quality 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
As the housing market becomes increasingly 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 inconsistencies, 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 repurchase 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 immediate financial losses, either. They also create long-term consequences, including reputational damage and increased regulatory 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 technologies 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 documents, 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 calculation, 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 transparency 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 potential areas of risk with every new transaction. This high level of precision not only reduces the probability of repurchases, but also significantly expedites loan processing by prepping files for underwriting and the quality control (QC) audit process.
Compared to human staff having to slog through countless hours of manually 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.
The benefits of these technologies are not theoretical. In addition to reducing the number of repurchase requests from the GSEs and other investors, lenders that have adopted them have achieved significant cost savings and substantial ROI.
In fact, our own clients have been able to reduce some income calculation costs by $400 per file, scale operations with automated classification of 689 document types and extraction of more than 10,000 data fields, leverage 20 component reviews to meet new prefunding requirements more cost effectively, and complete post close reviews within 90 days.
Of course, weeding out loan defects and meeting investor requirements more cost-effectively is not just about dollars saved and efficiencies gained. It is also about elevating a lender’s profile in the secondary market at a time when demand for higher-quality mortgages is at a premium. But this begs the question, what is the best way to take advantage of machine learning and other emerging technologies to streamline loan production, loan acquisition and QC processes?
Getting From Point A to Point B
For most lenders, the first step in process improvement and risk mitigation is shifting one’s focus from reactive to proactive. In practical terms, this means performing an honest and thorough self-assessment of the organization’s existing production and QC processes, hunting down vulnerabilities, and determining what works and what does not. If a lender has not audited its processes recently, now is the time. This audit should be as inclusive as possible, ranging from processing to underwriting to documentation to QC reviews.
It is equally vital for mortgage lenders to scrutinize the capabilities of their existing vendors and evaluate more automated tools to improve their own process efficiency.
For lenders, automation exists to accurately calculate borrower qualifying income, pre-screen loans for high-risk factors, and identify missing documentation or inconsistent data before they reach the closing stage. Post-close, technology can help to validate data integrity from documents and systems, leverage that data to power audit automation, and complete post close loan file reviews more quickly to align with a shorter 90-day review window.
For correspondent and servicing operations, where time is money, ensuring that loan acquisition and loan on-boarding is done efficiently and in a highly transparent way is critical. Missing documents and data issues can result in costly delays, holdbacks and tied up warehouse lines.
Finally, for any well-equipped operation, reporting and action plan tracking can uncover the root cause of defects and incorporate what has been learned for continuous process improvement. A lender’s future workflows, a correspondent’s guidance to sellers, and a servicer’s onboarding experience for borrowers can all be informed through reporting and trend analysis that detects troublesome patterns early for proactive action.
Taking the Next Step
In the challenging market we find ourselves in, adaptability to originate different types of loan products with quality while satisfying the timelines and requirements of any secondary market loan buyer represents an opportunity for transformative growth and heightened operational efficiency.
Lenders should act now on implementing automation to balance staff resources more effectively during market upturns and downturns, leveraging loan quality insights to understand defect trends related to shifting market dynamics, and validating loan file data to avoid investor repurchase risks.
For most organizations, it is more than just avoiding losses. It is about creating a more resilient, compliant, and profitable mortgage business for the long haul. In this sense, the sooner lenders conduct a deep dive into their processes and embrace new technologies, the faster they can protect their businesses and start turning the odds in their favor for increased profitability and growth.
