Halt Customer Defection
Keep Your Customers Using Analytic Models
Imagine being able to predict which of your customers will stay your customers. To achieve this heightened understanding of customer behavior, many marketers are turning to modeling. When performed properly, modeling can help organizations predict which customers will remain and which ones will defect—or churn. An effective modeling approach takes into consideration a wealth of information from a variety of sources, and can help marketers determine the best allocation of their marketing dollars to improve ROI.
Defection erodes your customer base and occurs with the cancellation of an existing contract, non-renewal at the termination date of a contract, or cessation of purchase behavior over a period of time. Models can help you define what defection really means within your organization. Take a look at your historical data to determine which customers did—and did not—renew contracts or continue purchasing from the previous year or two.
Modeling also enables you to create a profile, or fingerprint, of the two groups, which will allow you to see how they differ. This fingerprint also will help you predict which customers are likely to defect by including differentiating characteristics around purchasing behavior and lifestyle demographics. When added to a model, transactional data provides an analytical tool that can give you insight applicable to your entire customer base.
Internal and External Data
Several key types of internal and external data are needed to build successful models. Here’s a look at key data to consider, and why:
• Customer interaction transaction data lends insight into what a customer is thinking by examining his or her past interaction behavior. This data typically includes promotion history, and sales and service interactions. Transactional data can come from multiple sources and should include data from all relevant customer touch points, including direct mail, outbound telemarketing, e-mail and all other inbound responses.
• Purchase behavior demonstrates not only who purchased from you, but also what they bought and how much they spent. It is important to note when the first contact took place and when the first purchase took place, as well as the frequency and content of subsequent purchases. Whenever possible, it also is beneficial to examine your customer’s purchase behavior over a number of years. With all other things held constant, consumers who purchase across multiple product categories are likely to have lower attrition rates than single product purchasers.
• Customer interaction frequency provides a look at a customer’s interactions with your company or organization that could add to the “stickiness,” or staying power, of the relationship. Customers who buy frequently, but don’t interact and ignore promotional activity tend to have a high attrition rate.
• Demographics and lifestyle information provide important customer background such as marital status, age and income. This information can help you tailor campaigns to the groups that are most important to your organization.
• Data segmentation incorporates existing customer management segmentation. For example, some organizations divide customers into segments based on purchase incidence. Other segments could include new to file/new customer or multiple purchase, which then can be broken down into medium and heavy purchasers. Typically, these three segments behave differently. Transactional history won’t exist for new customers, so their risk profiles are largely dependent on non-transactional data such as demographics.
• Metrics should include the dollar value of purchases to determine the current value of each customer to the organization. In many cases, when using retention modeling, you will want to take a different marketing approach to risky customers who are of high value vs. risky customers who are of low value to your organization. A risk-value matrix provides a sound framework for managing the financial value of risk of attrition.
Depending on your industry, you may have different types of customer and company relationships that affect the definition and development of data models.
• Service providers. If your company provides a service as part of a fairly predictable, written contract—such as insurance companies and wireless service providers do—you need to take into account contract dates. While a customer can drop the service at any time, the greatest risk for customer attrition occurs when a contract is up for renewal.
• Contract dates define a critical marketing window for conducting retention marketing. It is important to evaluate customer value prior to contract anniversary date so the marketing department can develop a set of offers or a renewal strategy to communicate with these customers. Micro-marketing, or database marketing, which is the concept of using all available data on a prospect to maximize the relationship with that prospect, then becomes beneficial by combining all factors that are needed in creating and implementing effective marketing programs.
• Catalogs. Catalogers often hold “looser” relationships with their customers, and retaining them requires a different type of marketing, because there is not necessarily going to be an event or contract that drives retention or attrition.
Rather, the “contract” that exists with the customer more closely resembles a non-binding brand-loyalty relationship. There is an implicit agreement that the cataloger will provide an on-going suite of products that the customer agrees to purchase. This loyalty is defined in a steady and consistent pattern of purchase behavior over time. Any distinct disruption or absence of purchase behavior over an unusual period of time can be interpreted as a defection or attrition. For example, if a customer has purchased at least one product a year from a catalog for the past 10 years, the absence of a purchase in the eleventh year could be defined as a defection. Other factors, such as rewards plans and bundling a variety of product discounts, can be more useful in strengthening the relationship and are important in retaining customers.
• Nonprofits. As with catalogs, there commonly is a looser relationship between nonprofit organizations and donors. Again, any donor who has a history of donating a minimum of $25 to a nonprofit institution for the past five years, but does not donate in the sixth year, can be defined as a “defection.” While many organizations focus on creative and premiums to attract and keep donors, modeling will help determine which donors are likely to remain with the organizations and will allow marketers to focus more closely on the donors that are more likely to give.
• Local and long distance providers and ISPs. These companies have a “contract” with the customer that is defined as a signed agreement to pay for service one month at a time, with generally little or no penalty for switching providers. The competitive forces in telecommunications create a sense of urgency to proactively identify at-risk, valuable customers and provide appropriately customized offers and services to them.
Once you’ve applied the appropriate data and determined your differentiating factors, you then can rank your customers into 10 groups of equal sizes. The first group is made up of customers most likely to defect within the next six to 12 months, while the tenth group will be your most loyal customers. Within each group, rank the value of every customer as low, medium or high.
So, now you have a matrix that has 30 different groups that you can combine or rearrange to design cost-effective marketing programs that maximize profitability.
Model Life Expectancy
While the shelf life of your models can vary, it is best to do a model evaluation every six months. If the models still are robust and still are predictive—that is, if there has been very little slippage in effectiveness—leave the model alone, and re-evaluate in six months. A model is still robust if it’s accurately detecting defection rates consistent with the original results.
You can determine the accuracy of your models by conducting in-market testing of all risk groups, with special focus on marketing to a statistically representative group of consumers in the higher deciles—those who are least risky. Normally, these are consumers you would not need to market to if you are using attrition models, but it’s necessary to market to some to calculate save rates.
In the lower deciles, develop smaller test cells to measure both defection and save rates. From these test results, you can determine if the model still is valid by measuring defection and save rates to ensure they are rank-ordering the way the model was designed. If the results stop behaving as expected in terms of rank-ordering and size of deciles, it is time to re-evaluate. The ultimate measure of model validity is that the cost savings far out-strip the lost revenue due to more restrictive customer promotional campaigns and programs.
Use Models to Target Investment
A key part of the evaluation process is looking at the ROI taking place across the model deciles. If you use the same marketing approach and select the same spending in each retention decile, the ROI in the higher deciles will be much greater than the ROI in the lower deciles. If you currently are spending money to market to everyone in your customer base, you likely are wasting money by targeting the lower-performing deciles. You will be better off using the money to reinvest in top-quality marketing materials or loyalty programs aimed at your upper-decile customers.
You can use the information gleaned from your models to determine the best use of your marketing dollars and to proactively focus your efforts on consistently active and loyal customers. Modeling also can give you better insight into developing messages, creative, new offers or prices to more effectively engage these key deciles.
It is amazing how many companies continue to market to all active customers regardless of their contribution to the bottom line. Others rely extensively on the recency, frequency and monetary approach, which is not the most accurate way to do retention marketing.
A well-defined modeling strategy can fine-tune marketing efforts to reduce marketing expenses and maximize response rates.
Steve Briley is vice president of analytical services at Merkle Direct Marketing. He can be reached at SBriley@merklenet.com. Scott Cone serves as vice president of Merkle’s Database Group, and can be reached at Scone@merklenet.com.