Predicting Customer Churn

In today's fast-paced, visceral competitive environments, holding onto existing customers can be hailed as the ultimate business achievement. As a result, customer-centered data play a pivotal role in modeling approaches aimed at enhancing and optimizing service offerings across various industries, a more profitable alternative to previous retention programs that emphasized offer-centric strategies over personalized marketing models.

A key element in the search for customer retention, for example, is to understand churn probabilities. Customer churn, otherwise known as customer attrition, is a measure of the propensity of existing customers to switch service providers based on many factors ranging from interpersonal persuasion to overall dissatisfaction. Churning behavior is a classification challenge, so making intelligent choices about which data represent the most typical indicators is essential to identify high-propensity populations.

This article exposes the value of Gigasheet, an online CSV Viewer, in examining the credit card industry through its ongoing struggle with churning behavior; this isn't the prosaic schtick or practice of indiscriminately opening and closing credit cards to earn quick bonuses and similar rewards, but the benchmarking of relevant customer behavior (without the need to employ complex algorithms) to predict customer turnover. Let's take a look.

First steps to Predicting Customer Churn

Loosely described, our dataset consists of over 10,000 credit card customers divided into multiple categories such as marital status, age, and salary.

A bit of exploratory data analysis can help us coalesce these features into a cohesive picture—the idea here is always to reduce the dimensionality of data without sacrificing helpful information. With Gigasheet, grouping and graphing by specific variables (e.g., educational level) is as simple as using the Group feature, like so:

For any targeted retention program to work as intended, demographics such as these can help any marketing efforts steer clear of unnecessary tactics. If we know, for instance, that graduates comprise our most significant customer segment, these can now be prioritized for retaining purposes and similar claims.

Reviewing more data

As hinted, customer churn is a complex topic in that it implies the role of subtle customer-dependent facets, roughly categorized into voluntary vs. involuntary churn. The latter amounts to some form of canceled services (or subscriptions) upon unsuccessful payment, while voluntary churn entails a conscious decision based on unconventional yet more complex, measurable opportunities. Aspects such as better services or quality play a pivotal role in voluntary churn, whereas involuntary churn is typically more cut and dry.

Given the existing Attrition_Flag attribute, let's see if we can arrive at any additional conclusions or insights. We'll begin by using the Group feature once more to separate our data between "existing" vs. "attrited" (churned) customers:

Using the Attrition Rate formula for this period (1 year), we know that the churning rate amounts to roughly 16.07% (1627/10127 x 100). If we can’t tie this figure to any other transactional or behavioral factors, building any prediction will be difficult, if not entirely impossible. However, American credit card companies usually deal with churn rates anywhere from 20% to 25%, so a defection rate of 16% falls well below the average—already a valuable finding.

Let’s move on by associating average utilization ratios with customer status. Could there be a link between (frequent) credit card usage and customer churning? A simple data aggregation of the Avg_Utilization_Ratio column and a potential indicator emerges: Good-standing customers are almost twice as likely to engage in regular transactions than their attrited counterparts.

Equally foreboding descriptors point to married and single populations having the highest attrition rates, surpassing divorced and “Unknown” marital status customers 5 ½ times over. Using a pivot table, we can quickly identify these “risky” customers, giving departments like marketing ample room for decision-making:

Finally, we can reuse the Pivot Table mechanism to attempt to predict which credit card category (Blue, Gold, Platinum, or Silver) is ranked highest in the number of churned customers. This prediction based on credit card type isn’t novel, as most credit card companies are known to increase personal benefits to attract new customers or keep existing ones.

As the data show, Blue customers are at a significantly high risk of attrition. Adding to the evident problem of customer loss, this may indicate a severe flaw in how management handles this specific area or department.

Summing it up

Predicting attrition rates in the credit card industry is both a matter of profitability and survival. The pace at which many of these companies fight the innovation battle leaves little room for error and ambiguity, with customers having an ultimate say over much of the outcome.

To understand why we at Gigasheet have a different solution to the problem of quickly and effectively working with large datasets is to understand the modern business idea of entering predictive analysis without relying on highly complex applications or techniques. Do we sense a bit of skepticism?

Give Gigasheet a try today and see the power of this incredibly simple yet powerful tool as you make the most of your data in a handful of clicks.