March 2009

Data mining for customer loyalty
Creating meaningful loyalty measures
is no easy feat but it is achievable. By Richard Boire

The discussion of what constitutes true customer loyalty has been an ongoing debate within the marketing community. The debate is not so much about the semantic definition of customer loyalty. After all, most marketers would agree that customer loyalty is behaviour that is primarily driven by a strong affinity or attachment to a given company’s products or services. Where differences arise is in the methodology marketers use to measure or evaluate customer loyalty. Marketers need to define those metrics and measurements first and then determine whether the data is readily available from the current information environment to create such measurements.

Is RFM a good measure of loyalty?
The most common way to quantify customer loyalty is to examine prior purchase behaviour. One very common, quick and pragmatic way is by calculating RFM scores, whereby “R” represents the recency of last purchase, “F” stands for frequency of purchase and “M” is the monetary value of a purchase. From each of these metrics or measurements, RFM indexes or scores can be developed. Customers who have higher scores or indexes are viewed as “more loyal” than their counterparts. Many loyalty experts and pundits view this process as overly simplistic and ineffective at capturing the true essence of customer loyalty. The argument is that there are groups of customers with high RFM scores who are not really loyal and will whimsically switch companies or brands at the drop of a better offer. These customers have high purchase behaviour not out of any real affinity to the organization or brand but because they need the company’s services and products. Their so-called loyalty often stems from the fact that switching to another organization for the same products and services involves too much effort and risk. This situation can change on a dime if such a customer is presented with a more compelling option. But should we really care that these high RFM customers are not loyal? Does loyalty really matter if they are our most profitable customers? In many cases, emphasis on the value that these customers bring to the organization will result in marketing to this group regardless of the absence of loyalty. Programs could easily be created around the concept of “Best Customers”.

Marketing activities as a short-term stimulus for loyalty
But can we obtain even some basic learning around loyalty within this high RFM or Best Customer group? The answer is yes, if we view marketing activities to these customers as a means of stimulating customer attachment and affinity to the company’s product/service. The next question is how to measure or evaluate this stimulation. One way is to look at customer behaviour that is truly incremental as a result of the marketing activity. Overall, customer behaviour can be broken down into three key behaviours:

• lift
• shift
• retention

Lift behaviour can be described as increased usage of a product or service. Shift behaviour represents the acquisition of new customers to a product or service. Retention behaviour deals with maintaining customers’ activity relative to a given product or service. In all these cases, we want to identify the behaviour that is truly incremental as a result of the marketing activity. The way to identify these behaviours (lift, shift, and retention) as incremental is by using test cells and control cells. (Test cells represent the group receiving marketing activities. Control cells represent the group receiving no marketing activities.) Comparing the activity between these groups allows us to identify the difference and ultimately, the incremental behaviour of both groups.

The use of RFM allows us to identify the most profitable groups, thereby providing us with better ROI potential. The use of specific marketing activities allows us to be proactive in stimulating certain customer groups. Arguably, it is these “stimulated” groups that are the more loyal customers, at least within the specific marketing campaign.

Defining customer loyalty over the long term
But how about if we want to identify loyal customers from a much longer-term perspective? We know that previous purchase measures are simply not adequate in attempting to capture the real essence of customer loyalty. If we think about customer loyalty, then what are some of the other non purchase behaviour traits that define it? Let’s look at examples of three traits which might be considered as keys to identifying longer-term loyalty:

• overall marketing response
• customer inquiries
• customer complaints

Overall marketing response represents the customer’s total number of responses to all the marketing campaigns from a given company. A customer’s increasing willingness to respond to campaigns over time could certainly be construed as becoming more engaged with the company. These responses could be related to both purchase and non purchase activities.

Customer inquiries represent positive behaviour, since the customer is requesting more information from the company. With this activity, there is an implied sense of trust between the customer and company. This trust factor can only grow as each of these inquiries is handled in a positive manner. Conversely, customer complaints represent a negative activity between the customer and company. With each complaint, erosion starts to occur within the level of trust.

As database marketers, we want to identify these traits or characteristics as information that is recorded on a database. Marketing response can be captured if there is some kind of campaign management system within the organization. In this type of system, all campaigns and responses to those campaigns are captured at an individual customer level. This allows us to identify the overall marketing response history for each customer. Meanwhile, the Internet represents a channel for capturing customer inquiries. By examining companies’ Web logs, we can identify customers’ page views. In most cases, these are basic inquiries for information. If customers are exhibiting more page views within a given company’s Web site, they are generating more inquiries about the products, services, and other company information. Arguably these customers have a heightened interest in the company. In the telemarketing world, both outbound and inbound calls deal with customers who often inquire about other services and products as well as expressing complaints.

Creating meaningful loyalty measurement
So how do we properly organize this information into meaningful loyalty measurement? As previously mentioned, information on the Internet (page views) and marketing response history (taken from the campaign management system) are more easily organized since the data is located within structured data fields. The analyst can simply go to the appropriate field, whether it is a page request field on the Web log file, or the marketing response flag related to some specific campaign. In both cases, this information can be summarized from each of these fields to obtain specific views of customer activity or engagement. However, in the case of a telephone conversation, the information of interest arises from the dialogue between the sales or customer service rep and the customer and it is essentially contained in an unstructured format. Unstructured data means that we cannot point to one specific field but instead must search through the entire text of the dialogue to extract meaningful information. Extracting meaningful data from unstructured data has become a primary focus of today’s data mining research and [extracting meaningful data] is commonly referred to as text mining. Through this discipline, conversations, comments and anything that is of a textual nature can now be mined in terms of extracting meaningful information. In the case of customer loyalty, text mining can be used to identify customer complaints or customer interest levels based on the conversations or comments between a customer and sales or service representative.

The use of campaign management systems, the Internet and text mining provide new ways of looking at customer loyalty beyond just prior purchase behaviour. Instead of creating programs based on just prior purchase behavior, which can be profitable in the short-term, the longer-term perspective of defining customer loyalty allows the development of marketing programs designed around a definition which reinforces this notion of customer engagement and attachment. By stressing initiatives that encompass this philosophy of customer engagement, marketers are, in fact, creating long-term customer profitability, which is the overriding objective of any marketer.

Richard Boire heads up the Boire Filler Group, a Pickering Ontario based firm specializing in customer database analytics and predictive modeling. He can be reached at 905-837-0005.