How to Improve Customer Retention with Predictive Modeling

How to Improve Customer Retention with Predictive Modeling

Part 3 in a 3-part series on customer retention.

(Updated April 25, 2019)

As you know, retaining your customers is critical to your success. There are a number of ways to improve retention, including several ways to fall short. However, there is a proven, effective, data-driven way to improve customer retention.

Predictive modeling uses data and statistical algorithms to predict certain outcomes. Historical data, such as direct mail responses, spend, or donation amounts, are incorporated into the building of predictive models to predict the future.

This is beyond descriptive statistics or analytics such as response rate for direct marketing or Click-Through-Rate (CTR) for digital ads.

Predictive modeling aims to identify patterns and paint a picture from which a decision maker can draw actionable insight.

Simply collecting large sums of data is not necessarily going to help much on its own. Raw data is simply the starting point. Your data needs help telling its story. 

“The goal is to turn data into information, and information into insight.”
— Carly Fiorina, former CEO and Chair of Hewlett-Packard

Prudent decision makers in various industries use predictive modeling and analytics.

  • In manufacturing, Reliability Engineers combine it with predictive maintenance practices to develop maintenance strategies that prevent unplanned equipment downtime.

  • In financial services, lenders use it to predict the credit risk of consumers.

  • In marketing, data-driven marketers use it to predict consumer behavior and target promotions. They leverage predictive modeling to attract, retain, and grow their customers.

Predictive Churn Modeling focuses specifically on reducing customer attrition. (Sometimes this is referred to as Attrition Modeling.) Modelers can improve your customer retention through this specific analytics application.

This well-established approach works in large part because we are creatures of habit, which allows practitioners of data science the ability to predict behavior.

In fact, approximately 45% of what we do every day is habitual. [1]

Think about it.

Nearly half of what you and your customers do each day is out of habit.

The challenge lies in predicting consumer behavior accurately and consistently. Though this is no easy feat, the underlying premise is simple and intuitive—our behaviors are consistently associated with other aspects of our lives.

Those aspects are used as variables by modelers who statistically correlate them to the behavior they want to predict, such as defection. The variables that the modeler eventually chooses are known as the “predictive indicators.”

...we are creatures of habit, allowing practitioners of data science the ability to predict behavior.

In order to maximize the number of variables available for modeling, combining enterprise data with a robust third party database is essential.

The enterprise data should first go through a data hygiene process to optimize the append process. Then hundreds of variables should be appended to the enterprise data for proper analysis to begin.

Data variables can include demographic, financial, Zip+4, credit, lifestyle, hobbies, etc.

When identifying customers who are likely to defect, a modeler uses this data and predictive modeling techniques to build a model and measure the level of “disloyalty” in each customer.

The modeler analyzes the customers who have left and compares them to the rest of the customer base, identifying the predictive indicators that make those who fled the company unique from the rest.

...approximately 45% of what we do every day is habitual.

The predictors reveal a pattern that can be applied to any consumer record so that the modeler can contrast any given consumer with the others in terms of disloyalty, or attrition likelihood. 

The modeler then scores the database, ranks the customers, and identifies those most likely to leave.

The customers with the highest levels of disloyalty are the at-risk customers.

One of the major advantages of Predictive Churn Modeling is the ability to target these at-risk customers with specifically tailored messages and programs.

In collaboration with you, a modeler can also identify those at-risk customers who should not be saved because they are not profitable.

This predictive modeling approach allows you to target the profitable, at-risk customers and improve the ROI of your retention program.

Focus on the most profitable customers who are likely to leave.

Focus on the most profitable customers who are likely to leave.

If desired, the modeler can also identify the best timing for offering the right promotion. This option would require a designed experiment to obtain the necessary data.

This type of custom predictive model is crafted specifically for a particular company and their unique customer base. It is not a generic one-size-fits-all tool or software. Rather, it is “hand-crafted” by a modeler experienced in data, statistics, and marketing with a broad business acumen.

This approach helps you eliminate error in your marketing, focusing your dollars on the best customers who you want to keep.

Once a Predictive Churn Model is developed, you can tie it into the company’s acquisition program. Any direct mail acquisition list can be suppressed for potentially disloyal customers.

Predictive Churn Modeling also enables actions that feed into a larger strategy.

Predictive Churn Modeling can lift you out of a deluge of data and help you achieve your retention goals.

A modeler can collaborate with an company to build a long-term solution by gathering and evaluating the links between attrition and company action.

This approach evolves the initial model into something more. It becomes a conscious structuring of the actions taken by a company and the careful monitoring of the effect on consumers.

In this case, a modeler aims to reveal the cause-and-effect relationships between corporate actions (e.g. communications) and consumers’ behaviors and perceptions.

The modeler seeks insight into the interaction between the company and its customers.

In other words, a skilled practitioner can statistically identify company actions that trigger customer attrition among certain sensitive parts of the customer base. Price changes, for example, may affect lower income households more, and a change of layout or font may irritate older segments of the base.

This goes beyond looking at individual customers and looks at the relationship between a brand and its customers.

This brings about superior customer insight.

Once clearly understood, measurement of attrition is more thorough, and the business can proceed with better information about the effects of their own communications, pricing, layout, etc.

This approach addresses the problem in a more complete way than a single, one-time model with a consumer-only approach. It involves ongoing modeling and the evaluation of recommended retention campaigns. This refined insight can integrate nicely with acquisition, cross-selling, and upselling initiatives to improve their effectiveness.

Predictive Churn Modeling can lift you out of a deluge of data and help you achieve your retention goals.

If you haven't already, read the first and second part of this 3-part series on customer retention. 

If you're interested in learning more, download our special report by clicking the link below.


[1] Wood, W., Quinn, J.M., & Kashy, D. (2002). Habits in everyday life: Thought, emotion, and action. Journal of Personality and Social Psychology, 83, 1281–1297.