Using Data to Identify Members Ready for a Mortgage Refinance

As states reopen after the COVID-19 restrictions, mortgage refinance originations are expected to grow at a rate that exceeds twice the Mortgage Bankers Association’s initial Weekly Applications Survey projections – jumping 36.7% to approximately $1.23 trillion. This is great news for the mortgage refinance market, but how can credit unions take advantage of this turnaround? By leveraging data.

A credit union’s data, augmented with external information, can identify members who may be considering a mortgage refinance. Automatically and accurately identifying a subset of the total population will allow for efficient and targeted messaging to potential borrowers. Leveraging data to target the right audience not only reduces the manual effort required to find these members, it also allows for control over the advertisement noise members receive. Noise reduction increases the effectiveness of marketing, and enables a more precise, personalized outreach approach that members have come to expect from their credit unions.

If data is the solution, these are the next logical questions to ask, followed by answers that provide some general approaches to how your credit union could respond.

What data should I use and where can I find it?

To identify members most likely to be considering, or would be most helped by a refinance, a credit union should use two distinct data source families: Internal and external. Internal data illustrates how, when and why a member interacts with the credit union and defines who the member is. External data sheds light on a member’s tendencies and preferences when they are not interacting with the credit union, and who else holds a portion of their wallet share.

Internal data sets that have useful information can be found in the core, lending and mortgage origination systems; online banking platforms and transaction streams. Your internal data can even be found in the unstructured data lakes filled with text and speech data from NPS surveys, social media, call center recordings and comments. External data that should be used to enrich and provide context to a member’s action can readily be sourced from census data, credit reports, firmographics providers and global mortgage trend analysis such as the MBA’s Weekly Applications Survey.

How do I do it?

Once all the data is sourced, the next step is to draft a systematic methodology for using that data. Drafting this methodology is dependent on clearly defining the problem to solve (for example, “identify the members most likely to need a mortgage refinance”), and incorporating business resources and your people to develop the data definitions. Once the data is collected, defined and documented, the next step is to prepare the data for analysis by your internal teams. For some credit unions, the creation of machine learning training sets may also be considered.

The most important piece of data preparation is setting the boundaries of expected values for consequential fields, and then transforming the incoming data to match and solving for outliers. This process is called normalizing. Once the data is prepared, continual evaluation and optimization of the models and analytical data sets is crucial.

This is all in service to developing models to determine which members are likely to take advantage of a mortgage refinance, and a credit union leader can accomplish this in a variety of ways. Using the old stand-by spreadsheet or more advanced techniques like predictive analytics and machine learning are all valid. It is the outcome that matters – a list of members that meet that criteria, which is defensible and grounded in data.

Like the data ingestion processes, this process should be iterative so it returns results that meet expectations for accuracy, precision and depth. The output for these models should include a probability score for acceptance of a refinance by an individual. These scores are important as a credit union matures in its data journey, as model scores will become an objective metric for assessing how close the credit union was in its prediction. A high level of member engagement means it was a good approach and a low level of member engagement means the approach needs to be refactored. For credit unions focusing on personalization and creating offers on an individual basis, extra data preparation may be required to consolidate individuals with multiple accounts.

How do I act on it?

The end goal for machine learning models and predictive analysis should be easy-to-leverage datasets and a strong indication for action. These datasets then can be imported into a marketing campaign manager and explored using a data visualization platform. Each member should be pre-qualified for an offer and presented with an avenue for easy acceptance of the pre-qualified offer. Once the data is ready to be acted upon, a credit union should deliver messaging to the member, through their preferred channel, about the offers they qualify for (and given an easy way to accept them).

The key to a high level of success is a low amount of friction – making everything as easy for the member as possible, which is a secret Rocket Mortgage knows well. While people will always be a differentiator for credit unions, automation and digital transformation are also key enablers. They allow for an increase in internal operations efficiency and scale, and near instantaneous fulfillment, which members now expect.

How do I know if it is working?

Creating a data-informed list of individuals who can be closely monitored over time is crucial. To know if the analysis is working, simply identify if a member engaged with the communication and accepted an offer in accordance with the probability score. More members, with higher scores, should accept more offers. Gathering external data and noting any behavioral, transaction or balance changes to the member’s account internally can also help speak to the accuracy of the model.

To improve the performance of the model, the end-users should ask the following questions: Did we expect the member to refinance? Did the member accept a refinance offer through the credit union? Did the member accept a refinance offer from another financial institution? Were there any behavioral or situational changes that occurred that influenced the member’s decision between now and the time the model was run? With these questions answered, the targeted population can be compared to the credit union member population as a whole to determine the efficacy of the targeted refinance outreach program.

It should always be expected that early versions of analysis, predictive analytics or machine learning models can be improved over time. However, with enough good data, even early versions can prove to be scalable and a far more potent way of reaching members with attractive offers than any ad hoc approach. Using algorithms, a credit union can intelligently engage with those members most likely to be interested in the credit union’s products and services, while reducing costly misdirected efforts.

The data that can be used to determine who will accept a refinance offer – or perhaps more importantly, who would benefit from a mortgage refinance the most – is readily accessible to credit unions today. Using data and a few data science techniques, a credit union can better understand their members’ needs and create offers that will improve the financial well-being of both the credit union and the community they serve.

Ray K. Ragan, PMP is the co-founder of Clear Core, a data cleaning and transformation provider focusing on increasing the value and accessibility of data for financial institutions, in Tucson, Ariz.

Timothy “Buck” Strasser is the founder of Clear Core in Tucson, Ariz.