How a Mortgage Lender Used Advanced Analytics to Improve Retention Rates
Overview
From its outset, the company, a top five wholesale mortgage lender, recognized the importance of customer loyalty and retention for long-term growth. Nurturing the customer relationship throughout loan servicing was a bet on increased repeat business and cross-selling.
While customer loyalty is a crucial aspect of any business, it can vary greatly across industries. For example, according to a survey by the consulting firm Accenture, 87% of consumers in the retail industry are loyal to a brand, while, according to JD Powers the average customer retention rate for automotive brands is 52%. Accenture reports further that 47% of customers reporting being loyal to their bank, and 31% of consumers in the telecommunications industry exhibit brand loyalty.
However, customer loyalty in the mortgage industry is notoriously low, with only 15% to 20% of homeowners using the same lender when refinancing or purchasing a house, according to industry-wide statistics. The cause of low customer loyalty largely stems from the fact that the loan origination experience is expensive, complex, time-consuming, and often stressful (“We’ve got to close next week or we’ll lose the house!”). Moreover, homeowners are faced with dozens of lenders to choose from in a fragmented industry that often sells the right to service the customer, thus disconnecting from the customer they hope to retain in the future.
During the refinance boom of 2019 through 2021, the company saw the opportunity to realize the value of its strategic choice. As interest rates fell, customers took advantage of the opportunity to refinance their homes at a lower rate, potentially saving tens of thousands of dollars over the life of their mortgage. The company anticipated that its retention rate of these refinancing customers would be significantly higher than its peers given the resource allocation to its “customer-for-life” (CFL) strategy.
However, after reviewing early data, the company felt that the retention gap between the company and its competitors was not as large as it could be given the investment in CFL. The company’s leadership saw an opportunity to deploy advanced analytics to understand the differences between customers that were retained and those that weren’t. Leveraging data analytics is an essential tool for businesses that want to stay competitive in today's digital landscape. By analyzing customer data, companies can gain valuable insights into their customers' behavior and preferences. Consequently, the company tasked us with providing insights and recommendations for optimizing retention.
Approach
We formed a cross-functional retention team tasked with increasing the retention rate by 25%. The team included representatives from customer service, fulfillment, sales, marketing and, of course, data science and business intelligence.
We began by gathering data. Pulling together customer attributes from across the company’s systems including sales history, product types, financial profile, demographics, call center history and marketing. The goal was to identify meaningful differences between customers who were retained, meaning they returned to the company for their new loan, and those who originated with a different lender. We ran statistical analyses comparing the two groups across 17 loan attributes, including estimated payoff month, original loan amount, balance, FICO score, loan age, and other factors. Surprisingly, we found no statistical difference across these attributes between the two groups.
Next, we turned to customer survey data. We had built a robust “Voice of the Customer” platform that automatically deployed surveys throughout the customer experience in order to monitor the company’s Net Promoter Score (“NPS”). The surveys included both closed-ended questions (e.g. " How likely are you to recommend the company as a lender?") and open-ended questions (e.g. "What other feedback do you have about working with the company?").
When comparing the customer experience of retained and churned customers, we found a significant difference in NPS between the two groups. The NPS remained consistently different over time, even as the NPS changed for both groups. In addition to providing a framework for understanding the difference between retained and churned customers, we validated for the company that NPS served as a measure tied to its economic outcomes. However, none of the additional questions on the customer surveys produced any differences between the groups, and the difference between NPS scores wasn’t actionable—it merely surfaced distinct customer experiences for retained and churned customers.
To gain a better understanding of the customer experience, we analyzed the open-ended feedback from the NPS surveys to identify emerging topics that might account for the difference in NPS. Multiple natural language processing (“NLP”) approaches were used to investigate. Manual coding of a subset of responses was used to train gradient-boosted decision trees. We also used simple bag-of-words frequency counts of single words and word pairs. Finally, we deployed Latent Dirichlet Allocation ("LDA") for unsupervised topic identification.
Results
Our analysis produced several key findings. First, training decision trees using hand-coded survey responses did not yield satisfactory results due to the small sample and document (i.e., survey response) sizes. However, differential frequency counts proved to be surprisingly informative, and LDA provided additional valuable information despite the low explainability of the identified topics (“explainability” in machine learning is the ability to explain to stakeholders the logic driving an output). Second, the convergence of outputs from frequency counts and LDA produced a strong signal identifying a key driver of NPS, and, thus, retention.
By analyzing the words and phrases used in the open-ended feedback, we were able to identify key topics that were associated with high or low NPS scores. For example, we found that retained customers often praised the company's account executives and loan processors, while churned customers frequently complained about the closing process and communication issues. By identifying these topics, the company was able to gain a deeper understanding of the factors driving customer behavior and use that information to improve the customer experience. As a result, the company made operational adjustments based on our findings, and retention increased beyond the 25% improvement goal.
Conclusion
The results of our analyses demonstrate the benefits of using NLP to drive business value. By analyzing survey responses using advanced analytics, the company was able to gain more valuable information than traditional statistical analyses of customer and loan attributes. Moreover, NPS was confirmed as a valuable economic measure in the mortgage market, and leveraging NLP allowed for a more comprehensive understanding of customer needs and preferences.
However, the explainability of the results remains an open problem that businesses must address as they increasingly use NLP to analyze large amounts of data. With our help, the company is now taking on the challenge of automating the customer experience discovery cycle. We’re implementing a fully automated VOC platform including auto-identification of topics embedded in open-ended survey responses. The platform will then use these identified topics to dynamically insert multiple-choice Likert-scaled questions, and the results of these questions will be correlated with the NPS score to identify product or service attributes that are the strongest drivers of customer repurchase.
The company's use of machine learning techniques to analyze natural language data helped the company gain insights into the factors driving customer behavior. By identifying key topics associated with high or low NPS scores, the company was able to gain a deeper understanding of customer needs and preferences and use that information to improve the customer experience. As the financial industry continues to evolve, we can expect to see more companies leveraging the power of machine learning to gain insights from unstructured data and drive business growth.
