The era of COVID-19 has created unprecedented short-term disruption in the lives of millions of individuals and families. Record-high unemployment, reduced work hours, and rising costs have affected millions of people. For many households, these events create difficulties in making timely mortgage payments, and for lenders, are generating a stark rise in the number of requests for loss mitigation (e.g., loan modifications, forbearances, etc.).
Understanding how borrowers perceive their experiences with requesting mortgage relief due to the pandemic may help servicers boost trust, improve their reputation, and prevent the loss of business that often results from negative publicity in public and private forums.
To that end, this article explores the borrower experience during the loss mitigation process. First, we focus on why borrower experience matters during this process. Next, we summarize our findings from mining consumer complaint data and what it reveals about the borrower’s perspective of the loss mitigation experience. Finally, we offer ideas on how mortgage servicers can use this data to design better borrower experiences.
Why does borrower experience matter in the loss mitigation process?
The power imbalance inherent in the borrower-servicer relationship can interfere with the ability and desire to deliver a good borrower experience during the loss mitigation process. Borrowers may already have negative feelings about their situation and their limited ability to select a servicer. Moreover, servicers that receive lucrative fees on foreclosure actions may lack incentive to accommodate the borrowers. Such thinking fails to consider the potential reputational costs, compliance scrutiny and possibility for litigation, as well as the potential loss of business associated with poor customer service.
Loss of business
Mortgage servicers are among the least trusted financial service providers, and customers perceive servicers’ focus on cost-cutting and profits at the expense of customer satisfaction, according to J.D. Power. For mortgage originators that retain servicing rights, poor service quality and negative brand perceptions among current and new customers are major drivers of financial product selection decisions. A borrower with an unnecessarily poor service experience will likely avoid coming back to the lender for a future mortgage or doing any kind of business with the company in the future.
Moreover, the borrower can broadcast this experience to her or his personal network and the greater public. Borrowers have multiple platforms on which to voice their frustrations and distrust with mortgage servicers. Social media platforms, regulatory complaint systems, internet search engines and review aggregation websites allow any borrower to generate negative perceptions against a mortgage servicer’s brand. By sharing their negative experiences in person, borrowers also have important influence on friends and family members’ choice of a financial institution.
This visible erosion of trust in a brand can contribute to a decay of customer lifetime value with every borrower who leaves with a poor experience.
For third-party servicers, a poor servicing reputation may become a factor when borrowers decide to refinance their mortgages. When a borrower decides to refinance, why should he or she risk relying on a servicer known for poor management of property tax payments, inexplicable delays in applying payments against their balances, or just being difficult to work with?
Similarly, a poor servicing reputation may carry over to the brand perceptions of the mortgage originators, harming their reputations and revenue potential in mortgage lending and other products. Lenders are then likely to avoid servicers with poor servicing reputations. Consequently, even for servicers with a financial incentive to pursue foreclosures, poor service quality could be detrimental to long-term business relationships.
Poor experiences can also signal costly process inefficiencies. Lengthy calls, excessive calling, case escalations, manual rather than automated procedures, and other aspects of the borrower experience increase the costs of the loss mitigation process.
Ultimately, all mortgage servicers can protect revenues and profits by protecting their brands; and, part of protecting their brands is providing good borrower experiences.
Compliance and litigation costs
Consumer complaints against a mortgage servicer also attract attention from regulators, who use such complaints as a tool to uncover compliance issues (e.g., fair lending, RESPA, etc.). This scrutiny requires additional resources from the servicer to address the identified issues, such as manual reviews of loss mitigation requests. If regulators are not satisfied, the scrutiny may be elevated to an enforcement action and referral to the Department of Justice.
Individual borrower complaints may also lead to private litigation against a servicer. If there are numerous related complaints against the same servicer, even more costly class action lawsuits could ensue. Avoidable litigation consumes human and financial resources, and, like the cost of dealing with compliance investigations, can erode profitability.
What borrowers talk about in complaints against mortgage servicers
To improve the borrower experience, mortgage servicers must understand the borrowers’ complaints. These complaints tell services a great deal about what does not work in their loss mitigation efforts. ADI analyzed the Consumer Complaint database administered by the Consumer Financial Protection Bureau (CFPB). A summary of our methodology is provided at the end of this article. Following are the highlights from applying that methodology to analyze the CFPB loss mitigation complaint data.
Distinct topics that borrowers write about in their complaints about servicer loss mitigation
We identified 10 distinct topics that borrowers write about because of their experiences in servicers’ loss mitigation processes. These topics are described in Table 1. We consider these 10 topics to be a reasonable summary of the diversity of issues that borrowers raise about the loss mitigation process in their complaints. While these issues are likely to be obvious to any mortgage servicer, the ability to classify each complaint based on its mix of topics provides structure to the text data that supports some useful comparative analyses.
There are potential performance gaps in the borrower experience between servicers
We compared the mix of complaint topics for the ten mortgage servicers included in our analysis. Figure 1 shows four of the servicers with the most pronounced deviations in topic mix from the average complaint included in our analysis.
As illustrated by Figure 1, borrowers for Servicer A were more likely to write about program eligibility and their homes than other servicers, while Servicer B tended to receive complaints about legal matters. Servicer C’s borrowers tended to write about the disposition of their homes and documentation issues, whereas Servicer D’s complaints were more likely to concern communicating with the servicer as well as process and payment issues.
While these differences are driven by various factors – such as variances in the composition of products in the companies’ servicing portfolios – the figure highlights specific weaknesses in the borrower experiences each servicer delivers. For instance, Servicer A may be more likely to deny loss mitigation requests than other servicers, while borrower experiences with Servicer B tend to inspire appeals about legal remedies.
The complaints from Servicer C’s borrowers suggest difficulties during the short selling and foreclosure process and concerns over missing or alleged fraudulent documentation. Servicer D’s complaints suggest potential issues in how the servicer communicates with borrowers and handles their account payments and requests for loss mitigation.
The complaints described above indicate potential inefficiencies in the loss mitigation process and heightened compliance and litigation risk. Such shortcomings increase operating costs and legal expenses above the typical mortgage servicer.
Borrowers with different characteristics often focused on specific complaint topics
The data about the borrowers in the CFPB complaint database offered limited insight about the concerns of some groups. Upon submitting their complaints to the CFPB, borrowers can self-report as older Americans (i.e., 62 years and older) and former or active U.S. service members. In addition, the published data includes the borrower’s three-digit zip code (ZIP-3), thus providing some geographic perspective.
A review of the distribution of the most common complaint topics among different groups of borrowers highlighted some differences in their experiences. Compared to borrowers from majority-White-non-Hispanic ZIP-3 geographies who did not self-report as an older American or service member:
- Borrowers from majority-minority ZIP-3 geographies were significantly more likely (at a 95 percent confidence level) to focus on legal issues (25 percent more likely) and program eligibility (20 percent);
- Older Americans were significantly more likely to focus on their homes (84 percent) and documentation (31 percent); and
- Service members were significantly more likely to focus on their homes (73 percent), legal issues (43 percent) and account payments (19 percent).
Understanding such differences in the experiences of borrowers based on their characteristics can provide important insights to mortgage servicers. The skew toward the topic of the home among older Americans and service members potentially signals the unique importance and emotional attachment these borrowers have with their homes. Special training among mortgage servicer representatives to proactively recognize and address these unique concerns from older borrowers may be necessary.
Separately, the focus on legal issues (majority-minority geographies and service members), program eligibility (majority-minority geographies), documentation (older Americans) and payments (service members) may indicate differences in how protected classes of borrowers are treated versus their comparators. Such differences in treatment can lead to Fair Lending risk if these experiences correspond with differences in adverse outcomes from the loss mitigation process.
Profiling complaints to understand and enhance borrower experiences
So far, we have explored differences in how borrowers write about their experiences based on differences in the mix of topics. Comparing the mix of complaint topics by mortgage servicer and the borrower characteristics illustrated potential performance gaps, elevated compliance risk and training opportunities. However, it can be a daunting task to formulate a meaningful abstraction of thousands of complaints in a manner that offers actionable insights for improving borrower experiences during the loss mitigation process.
To help make the raw complaint data more actionable, we distilled the thousands of complaints into a small subset of complaint profiles, each of which contains a distinct mix of the 10 complaint topics but focused a few key topics that distinguish that profile. Each profile represents a specific complaint scenario that highlights potential pain points and needs of borrowers in that subset.
ADI conducted an exploratory analysis to evaluate the value of profiling borrower complaints. Using the segmentation analytic methods described at the end of this article, we identified eight distinct complaint profiles based on their unique mix of topics.
Figure 2 compares the mix of complaint topics of four of the 28 complaint profiles with the average complaint and illustrates how similar complaints tended to focus on specific topics relative to other complaints.
As an example of how these complaint profiles can be used, we can see in Figure 2 that Profile A complaints tended to discuss issues related to payments, notifications, and legal matters. These overall statistics and a review of a sample of the written narratives indicated that Profile A complaints typically alleged the servicer failed to apply payments or provide notifications in accordance with contractual or regulatory requirements. In many cases, this type of complaint leads to threats of legal action against the servicer.
Being able to identify this type of complaint and all of the touchpoints that lead up to it will allow the mortgage servicer to recognize where the borrower experience broke down and how to fix the underlying issue (i.e., payment and notification processing).
Profile B complaints tended to focus on eligibility for servicers’ loss mitigation programs and the borrower’s home. A review of narratives for this complaint indicated these complaints generally were about the loss of the home through foreclosure after being denied loss mitigation. Determining the touchpoints that lead up to this type of complaint may provide training opportunities for representatives to help borrowers through the painful process of losing their homes before a complaint is submitted to regulators.
Reviews of Profile C complaint narratives indicate these relate to the subset of servicers’ solutions, such as processing short sell requests and foreclosures, while Profile D complaints tended to focus on being approved for loss mitigation requests and receiving the appropriate notifications of those decisions. In each case, being able to identify and study these scenarios helps servicers uncover specific problems with their processes.
How mortgage servicers can enhance the borrower experience
Our analysis of complaint data in the loss mitigation process was a means to explore these borrower experiences, and uncovered potential performance gaps, elevated compliance risk and distinctive experiences. Analyzing CFPB complaints, however, is just one of many actions that mortgage servicers can take to truly understand and improve borrower experiences for their own servicing portfolios.
Gather data about borrower experiences from multiple sources
Surveying borrowers in the portfolio can provide crucial information about the borrower experience. A well-designed survey can provide structure to the collected data by homing in on specific touchpoints and other elements in the mortgage servicing process.
Mortgage servicers can also analyze existing internal data to understand how representatives interact with borrowers. This data includes communications logged in a customer relationship management (CRM) system, emails, virtual chats, and transcribed phone conversations.
Gathering data outside of the organization also contributes to a more complete picture of the borrower experience. The CFPB complaint data analyzed for this article is one source. Other sources that can provide data with both positive and negative sentiment include reviews on third-party sites, social media posts and competitor surveys. These external sources provide context by highlighting the gaps in service quality between competitors.
Analyze the data to understand and enhance borrower experiences
With the internal and external data collected, various analytic techniques can be applied to create useful insights that can lead to improving borrower experiences.
Different borrower experiences – such as the monthly payment and loss mitigation processes – can be clearly defined by developing customer journey maps that identify the phases, touchpoints, and desired outcomes of each process. Figure 3 provides an example of such a customer journey map of the loss mitigation experience.
Profiling borrower survey responses will uncover how different types of borrowers feel about their interactions via the various touchpoints in the experience. This knowledge allows the mortgage servicer to identify the weaknesses in the overall experience, sparking ideas from employees throughout the organization about how to achieve the desired outcome less painfully.
For example, a possible solution to test regarding the distinctive topics raised by older Americans and service members as identified in our analysis would be to screen for these customers and escalate them to responders that are specially trained to manage the typical problems and concerns that these borrowers will likely raise.
The collected data may also be modeled to identify which experiences present some risk to the mortgage servicer. For example, a predictive model may be developed to flag interactions that have a high degree of likelihood of resulting in a complaint to the CFPB or another regulator or other governmental body focused on consumer protection. Such knowledge could help the servicer address the issues before regulatory scrutiny is raised.
Similarly, borrower histories and interactions may be modeled to identify cases with an elevated chance of entering loss mitigation. Identifying those cases proactively may allow the servicer to provide a more profitable resolution than if the case enters the normal loss mitigation workout process.
Delivering good loss mitigation experiences is good for business
As discussed earlier in this article, delivering good borrower experiences can minimize costs, reduce exposure to compliance risk and protect mortgage servicers’ brands for sustained, maximum profitability. Accomplishing these goals requires having a thorough understanding of borrower experiences.
In this article, we have shown how analyzing unstructured data, such as the CFPB complaint data, can offer useful information. Complaints, by definition, and the loss mitigation process are guaranteed to generate negative sentiment from borrowers, yet the insights hidden in these data can be leveraged to understand breakdowns in the borrower experience during the loss mitigation process. In many cases there will be no resolution, but other issues, such as misapplied payments, inaccurate documentation, excessive collection calls and poor customer service, can be resolved.
Based on our experience assisting clients across several industries, mortgage servicers can expect a wealth of new insights from data gathered via borrower surveys, competitor surveys, internal communications, and other sources. The knowledge gained will be invaluable to optimizing the borrower experience and, in turn, protecting the bottom line.
We would be pleased to discuss how Person to Person QualityTM could assist your organization in understanding and enhancing the experience of your customers.
Addendum: Text Mining Methodology
Following is a brief overview of the methodology used to analyze the complaint data referenced throughout this article.
The source of our data was the Consumer Complaint database administered by the CFPB as mandated under the Dodd–Frank Wall Street Reform and Consumer Protection Act. These data consist of complaints submitted by customers of financial institutions related to various products and types of issues.
For this article we focused on issues related to borrower requests for assistance with difficulties making payments on their mortgages. In the database, these complaints are classified as issue types “Loan modification, collection, foreclosure” and “Struggling to pay mortgage,” the latter replacing the former in 2017. In total, we extracted 11,766 complaints with narratives that were received by the CFPB between March 2015 and April 2020.
To make some of the data processing tasks more manageable, we filtered for the top 10 servicers in the dataset by number of complaints. The resulting dataset consisted of 6,992, or 59 percent of all complaints with narratives extracted from the CFPB website.
Processing the data for text mining
To analyze the dataset, we applied common text mining techniques using the R statistical language. Following is a general outline of the steps taken to prepare the dataset for analysis.
- First, the narrative data were converted into a table where each row represented a word or number from the corresponding narrative.
- Next, we eliminated common words in the English language that generally do not provide value when analyzing text data, such as articles and prepositions. These are known as stop words in text mining. Examples include “and”, “the”, “a”, “is”, “at” and “on.”
- Then, references to numbers and coded redactions (i.e., “X” in the narrative data) were removed.
- Next, we excluded words unique to the subject of the narrative data (i.e., complaint data regarding mortgage modification and loss mitigation requests) that we concluded did not provide meaningful information. This included the words “modification”, “mitigation”, “mortgage” and “loan.”
- Then, we applied an algorithm to reduce words to their base word stem. For example, words like “require”, “required”, “requires” and “requirement” are related to the same underlying word and concept that can be reduced to the common word stem “requir.” This was done to reduce the redundancy of different variations of the same root word in the dataset.
- Finally, we excluded words referencing the 10 servicing companies included in the analysis. This involved excluding any word that was a part of the servicing company name for the corresponding narrative as well as acronyms and misspellings that we identified.
The result of these processing steps was a final analytic dataset consisting of 704,889 distinct words and corresponding attributes about the complaint, such as the name of the servicing company, the type of product and borrower characteristics.
We processed the final dataset using a modeling technique known as Latent Dirichlet Allocation (LDA) to identify abstract topics from the words. We calculated several models testing different numbers of topics and removing sparse terms until we identified a model with topics that offered a reasonable interpretation. The final model that was selected represented 10 topics as discussed earlier in this article.
The topic modeling generated estimates of the probability that each complaint could be classified to each topic. Put simply, for each complaint, a probability estimate was calculated for each of the 10 topics included in the model. We interpreted this as reflecting the fact that a complaint typically contained several topics, not just one.
With the interpretation that each complaint represented some mix of the 10 topics identified in our analysis, we applied a model-based clustering algorithm to classify each complaint to a small subset of complaint typologies. Applying this algorithm on the topic probability mix automatically identified 28 distinct types of complaints. Analyzing the topic mix and a sample of complaints in each of these subsets would allow for the development of complaint profiles that could be used to monitor and train for specific complaint scenarios.