For many successful lenders, part of their “secret sauce” depends on the partners they decide to work with during the loan origination process. Choosing the right set of data providers can be a rigorous endeavor that requires months to complete. Even after lenders choose the right data providers, it can take several months to get the new data provider integrated into the loan origination process. A key decision that lenders need to make is how to gain access to data sources. Centralized hubs provide the flexibility to connect to multiple data sources easily. Industry leaders tap multiple external data sources to improve the predictive power of their credit scores and to explore other sources of alternative data. Therefore, to code and deploy new models quickly, the connectivity solution selected should enable fast, low-cost, seamless integration of new data providers. BEST PRACTICE 2 Use tri-bureau attributes in your credit scoring models for maximum flexibility Credit scoring models that are used in production at the time of application can reside in various locations depending on the digital ecosystem of the lender’s application and decisioning platform. For example, a lender might choose to have one or more credit bureaus code and install the model at the bureaus. Alternatively, a lender might decide to house it inside the LOS. In certain cases, the model may reside inside a centralized data connectivity hub where it can incorporate data from multiple bureaus and other data providers. Wherever the credit scoring model is deployed, the raw credit bureau data (inquiries, tradelines, public records, etc.) must be converted into a set of predictive attributes that can be used to calculate a score. Several options exist in the marketplace that can compress raw bureau data into a set of predictive attributes that model developers and data scientists require to build their credit scoring algorithms. Some options are more limited than others. Each has unique advantages and disad- vantages depending on the needs and resources of the lender. Certain predictive attribute sets are designed to be calculated on only one bureau, or one bureau format, whereas others can work on multiple bureaus or multiple bureau versions. Some sets can merge credit data files across multiple
applicants, or multiple bureaus, removing duplicate tradelines, inquiries and public records to create a consolidated view of the applicant. Many industry leaders choose to use tri-bureau attributes to gain maximum flexibility in the data sources that can be used as inputs to the model. Using tri-bureau attributes often results in signifi - cant efficiency gains for the lenders that choose to do so. These attribute sets synthesize credit information for easier analysis, transforming redun- dant data into valuable information. Credit-based decisions can be made more consistently with attributes that are uniformly leveled across each of the major credit bureaus. BEST PRACTICE 3 Store both the raw data and predictive attributes used to build your credit scoring model for improved efficiency Industry leaders are especially adept at mining their internal performance data, as well as combining it with external data sources to identify highly predictive credit signals. This requires advanced analysis of existing sources of data. One example is mining trended data that reveals spending behavior that might involve using attributes that are more complex than the number of delinquent tradelines or recent inquiries. Keeping the raw credit bureau data allows lenders to explore innovative ideas for future attributes that might be useful in the next model build. During the Covid-19 pandemic, lend- ers that were able to take a deeper look into raw tradeline data were able to deploy new attributes that focused on identifying hard-to-find loan accommodation situations. An analogous situation exists with lenders that want to identify and track the usage of buy now pay later, credit builder or medical tradelines. Storing the complete set of predictive attributes, not just the ones used in the current production models, empowers data scientists to efficiently prepare model-ready datasets with updated per- formance information so they can quickly build and validate challenger models. A lender can use machine learning (ML) or artificial intelligence (AI) techniques to develop challenger models in parallel with the champion credit scoring model to discover where other credit signals might provide potential lift. Savvy data scientists can also use ML models to understand specific segments where they might
As lenders continue to enhance their application and decisioning platforms, they need more sophisticated and automated credit scoring models that can incorporate varied data from inside and outside the organization.
improve a traditional regression-based model. They can carve out these segments and build dedicated models to increase the overall predictive power of
than just a basic service relationship. Partners willing to do the “something extra” are the ones lenders can trust to deliver results. Good partners will have extensive industry experience, be able to function as a central connectivity point to data providers, will be capable of supporting legacy partners, and offer a mature API infrastructure. The four best practices discussed in this article can help any lender elevate its credit model to manage changing economic conditions, and do so more quickly. ▪ About Digital Matrix Systems (DMS) Digital Matrix Systems (DMS) works with banks, credit unions, auto finance lenders and insurance companies to extract value from data and make informed decisions for their business. We provide centralized connectivity to credit bureaus and alternative data providers. In addition, we help power the analytics and processes used for marketing, loan origination, dispute resolution, collections, fraud mitigation, risk management, and other critical business activities. Our tri-bureau DMS Summary Attributes ® are standardized, reduce redundant data, and can be combined with internal performance or archived data for predictive modeling. About the author Michael Sogomonian is the Senior Analytics Consultant at Digital Matrix Systems (DMS) . In his role, he helps clients gain optimal value from their data through the DMS suite of data access, storage and analytics solutions. Michael joined DMS in 2016, bringing over 20 years of prior risk management experience in both the banking and insurance industries. He specializes in building and integrating credit scoring solutions into legacy and emerging decision engines with minimal disruption.
the scoring system. BEST PRACTICE 4
Leverage in-house business expertise for new insights
AI and ML are remarkable tools, but credit scoring models should not be based solely on traditional statistical nor newer “black box” methods. For a truly robust and high-quality model, lenders need to leverage their internal business expertise through- out all phases of the model development process, from data acquisition to model deployment. This will help them understand what credit signals are missing and then recognize and validate new ones. For example, data scientists and model develop- ers should interview risk managers and underwrit- ers on credit issues and collaborate with them on how to convert those insights into qualitative questions for improved credit signaling or novel attribute design. Business experts can also assist in validating the usage of certain attributes based on their own real-life interactions with customers, knowledge of underwriting and approval processes and understanding of compliance. Final Thoughts As lenders continue to enhance their application and decisioning platforms, they need more sophisticated and automated credit scoring models that can incorporate a wide variety of traditional and nontraditional data from inside and outside the organization. In the face of intense pressure from both fintech and challenger banks, this will make them more competitive and resilient in challenging economic times. To develop the infrastructure to rapidly develop and deploy new credit scoring models, lenders should find the right technology partners that are willing to form a strong relationship. It goes deeper
34
35
THE FINANCIAL BRAND INSIGHTS SPRING 2023
THE FINANCIAL BRAND INSIGHTS SPRING 2023
Powered by FlippingBook