Picture the ideal customer: a frequent shopper, highly engaged and loyal. 

But most business can tell you even the “ideal” customer can be fickle. She'll turn away (or be lured away) in an instant, without any warning.

Retailers and card issuers agree that one of their big customer management issues is attrition. Attrition that occurs in non-contractual relationships is “silent,” because the consumer does not cancel an account. They simply stop using the loyalty or credit card. 

Silent attrition is a costly problem for businesses. Many experts believe it is five to 10 times more expensive to acquire a new customer than to keep a current one.

When silent attrition occurs, companies have a brief window of opportunity to re-engage the customer before the parting becomes permanent. Rapid detection of silent attrition and fast contact through mobile channels provides an opportunity to hold onto valuable customers, if the business is able to create persuasive and meaningful offers.

Capturing the Moment, Before It's Too Late

Creating such offers and programs is easier to do now using big data and predictive analytics, yet the sheer volume of accounts and data has often limited the frequency and granularity of this process. Many companies rely on monthly data aggregations of customer activity that could be weeks out of date before any action is taken. 

And by that time, customers have often moved on to new cards or shopping patterns before companies become aware of the danger.

By updating traditional attrition models daily with granular transaction data, marketers can have full visibility into a customer’s spending pattern. This process is possible by leveraging machine learning to capture complex, predictive patterns involving thousands of features. These types of dynamic attrition models can produce more precise attrition-risk scores that indicate the earliest signs of attrition risk with the lowest latency so marketers can take immediate and relevant action (see Figure 1).

Customer Attrition Scores

Fig. 1: Daily attrition score’s dynamic response to a customer’s intermittent card use. While customer is active, attrition risk vanishes. Risk starts to build up quickly during a multi-day activity lapse, and returns to near zero when customer uses her card for another purchase.

Designed to address silent and sudden attrition for the card and retail industries, these models can complement existing monthly processes or stand on their own. For example, the monthly process can predict substantial future reductions in card balance or customer spending based on observing earlier declining trends that are likely to continue (“balance” or “spend” attrition). Complementing these slow forms of attrition, the dynamic attrition model detects other customers who abruptly change from active to inactive.

Besides predicting customer-level attrition, dynamic attrition models can also be geared to provide detailed category-level analytics, such as when the customer stops using the card for gas, or at a particular merchant. These models predict exactly how this risk increases with each day that passed since the last activity. Armed with these predictions, marketers can move quickly to pursue customers with targeted counter-offers, say, at a lower price point, as soon as the attrition risk exceeds a threshold.

Improve Your Retention Offers

Timely and accurate prediction of attrition risk is only half of the solution. Finding the right offer that will persuade a customer to stay (while being easy on the marketing budget) is the other half of the problem.

Traditional campaigns are often designed using “single shot” decision analytics, whereby retention offers are based on observed historic customer behavior. For example, is the customer a transactor or a revolver? While past behavior can inform campaigns, using it alone can limit the understanding of preferences that may have changed recently.

This is where sequential decision making shines. Sequential decision making is an analytic process that formalizes and solves for how a sequence of interdependent decisions should be taken to optimize long-term objectives, such as customer lifetime value. Here we see attrition risk as a symptom and we seek additional information from the customer to diagnose the problem before making a retention offer. 

To better understand customers’ changing preferences and avoid customer fatigue, businesses should:

  • Leverage investments in mobile channels and apps to engage in customer dialogue in a timely, convenient manner
  • Ask a simple question or two. For example, would the customer like a new card feature, prefer different reward options, prefer a plan with lower fees?
  • Make a more relevant offer to persuade the customer, based on her or his response

On the flip-side of retention, businesses may be tempted to curtail stale customer relationships in cases where expected future customer value would not exceed marketing spend. The question is, how long should a business try to re-engage with customers before cutting costs? And should this decision depend on characteristics of the earlier active relationship with the customer?

Again, sequential decision analytics, with the objective of maximizing customer lifetime value, tackles these questions and provides an effective approach to organizing and analyzing customer behavioral data.

Advanced predictive modeling and sequential decision analysis can help businesses make more informed retention offers and decide when to cut costs from the “lost causes” who are highly unlikely to re-engage.