Residential Magazine

Predictive Analytics in Underwriting

Art and science must remain in balance when using technology

By Brian Kucab

Underwriting has traditionally necessitated a fine balance between “art” and “science,” requiring insights from historical trends as well as assessments of individual circumstances. As underwriting departments and companies seek further efficiency and process enhancements, some fear this balance could be threatened by the increased reliance on technology in the underwriting process.

One change causing this fear is the increased confidence in predictive analytics, which leverages the use of big data to make smarter and quicker decisions. These decisions are offering borrowers, originators and underwriters process efficiencies, decreased risk and better customer experience.

Without a doubt, however, predictive analytics are here to stay. Research performed by LIMRA, the Life Insurance and Market Research Association, states that “nearly nine in 10 financial services companies have or are exploring the use of big data analytics to streamline automated underwriting processes.” With this in mind, the question becomes whether or not predictive analytics is truly a threat to the underwriting balance.

Cognitive insight

The use of predictive analytics does not reduce the qualification elements reviewed or eliminate components of the underwriting process, but instead increases the efficiency of how those elements are reviewed. Rather than simply automating the review of submitted qualification documents, predictive analytics streamlines the process to maximize efficiency and accuracy.

This is a critical distinction. The adoption of predictive analytics is not intended to cut corners in the evaluation process. Instead its aim is to drive efficiencies and allow companies to be smarter about risk.

There are two benefits of predictive analytics: process automation and cognitive insights. Process automation streamlines underwriting by allowing the system to handle pieces of the process where humans are not needed. Ingesting information such as borrower reports or asset statements from loan files, for example, can turn this info into data and run it against a set of rules designed to test acceptability. The end user is then presented only with a decision as well as anomalies or discrepancies that need to be addressed.

Cognitive insight drives efficiency and accuracy by offering information on more complex portions of the process to help in the decisionmaking process, but ultimately leaves those decisions to the human users. Models that run during the origination process and provide granular visibility into specific credit or manufacturing risks for given loans, for example, can inform under- writers about items needed for their evaluation. Further, cognitive insight equips and teaches human users, making users faster and smarter.

The addition of predictive analytics to the underwriting process is not transformational, but incremental. It allows underwriters to focus on the subjective items that require human judgment and intuition, while allowing the system to handle administrative items that would otherwise reduce human efficiency.

Predictive analytics offers multiple ways to improve underwriting, including through loan-classification modeling, which provides valuable borrower insights; volume forecasting models, which estimate underwriting capacity; and rules-based modeling, which creates waterfalls in workflows. Let’s look at each of these in turn.

Loan-classification modeling

Loan-classification modeling uses historical data, loan characteristics and market variables to classify loans based on specified characteristics, such as risk. This classification gives underwriters insights on how deeply they want to review a loan.

This model can be used to drive efficiencies and improve the customer experience. These benefits can be further increased by introducing process automation through cross identification of documents and data within automated classification and extraction software. This model can help accurately predict the probability of outcomes, increase efficiency by directing underwriter focus and identify indications of misrepresentation.

Accurately projecting the probability of expected outcomes early in the process can help underwriters set more accurate expectations for timelines within the process. The origination process can vary significantly, depending on the form of both data and documents — which are often known early in the process. Using these data points to estimate likely outcomes — and updating outcomes as information changes — can help reduce unnecessary wait times.

Even without fully delegating the decisionmaking to a predictive model, predictive analytics can still introduce efficiency into the process by driving the underwriter’s areas of focus. If the likelihood of an outcome is within an organization’s risk parameters, for example, the model could be leveraged to point the underwriter toward only those portions requiring analysis.

Similarly, specific loan characteristics or documentation patterns could select loans for a heightened, but still targeted, review at the beginning of the process, rather than the end. Targeted, concurrent review processes can shorten the overall processing time while reducing the risk of a manufacturing defect.

Similar modeling techniques also can be used to identify highly predictive indicators of misrepresentations. By analyzing large amounts of data and document characteristics, businesses can track patterns between loan parties over both time and geography to more effectively target forensic reviews. As red flags and items for concern are either confirmed or cleared, the models are refined and accuracy increases.

Volume forecasting

Volume forecasting models use historical data to predict underwriting volume in the short-term (daily and weekly) and long-term (monthly and even quarterly). When built on experiential data and predictive macroeconomic variables, these models can drive multiple approaches to maximizing staffing, which is critical for ensuring fast turn times and processing rates.

These approaches can have daily, weekly and monthly impacts and opportunities:

  • Daily. Volume forecasting allows managers to anticipate periods of peak volume, allowing cycle times that customers desire to drive staffing, rather than having staffing drive the turn times that customers experience. If managers can anticipate a volume surge on Wednesday and a drop on Friday, schedules can be aligned accordingly.
  • Weekly. Managers can communicate volume and staffing needs with their teams weeks in advance to ensure adequate coverage and maintain a quality customer experience. By anticipating surges far in advance, managers can secure overtime commitments before personal schedules are set and schedule meetings, training, etc., so as to not conflict with peak volume periods.
  • Monthly. By anticipating volume trends before they occur, managers can address staffing gaps ahead of time. Waiting until the volume is realized can leave managers competing for available resources, thus increasing costs. Being ahead of the market creates a more ideal hiring environment.

Volume forecasting allows managers to be proactive and forward-thinking when managing staffing. Not only does this minimize stress for both managers and employees, it also helps underwriting teams to consistently stick to offered turn times and improve customer satisfaction.

Rules-based modeling

The third model type, rules-based modeling, uses decision trees to create workflow waterfalls. This means the model can be used to route specific types of files — such as files needing a particular type of review — to specific underwriters.

This type of modeling is attractive because of its simplicity. Rules-based modeling does not require a large information technology budget, so it is an ideal tool for smaller companies seeking increased efficiency. Despite their simplicity, rules-based models can introduce significant benefits when combined with automated data collection and workflow.

Specialization at a granular level can help reduce training time and allow for quicker utilization of resources. A prime example of this is the processing experience on a loan from a single salaried borrower compared to a loan with multiple borrowers who are self-employed. The ability to quickly route each of these files to a separate underwriter who has experience with those particular loan parameters improves efficiency all around.

As evaluations of loan data and documents against historical experience and anticipated outcomes becomes more granular and complex, the opportunities for segmentation grow.

• • •

Underwriting technology is critical to delivering a quality product and a quality customer experience in the mortgage industry. And happy customers are more likely to return or give referrals to a mortgage company’s originators in the future.

Ultimately, predictive analytics’ greatest contributions to the underwriting process are increased insight, greater efficiency and improved accuracy. Through various models, predictive analytics offer opportunities to improve the underwriting process by automating the science and allowing underwriters themselves to focus on the art.


  • Brian Kucab

    Brian Kucab is director of underwriting at Genworth Mortgage Insurance. He has held various leadership roles with a focus on risk assessment and mitigation, as well as process enhancement. Kucab began his financial services career as a residential underwriter before joining Genworth in 2007 to focus on contract-underwriting recourse. In 2009, he moved into loss mitigation as director of investigations, with responsibility for delinquent loan reviews and forensic underwriting.

You might also like...