We use cookies and other similar technologies (Cookies) to enhance your experience and to provide you with relevant content and ads. By using our website, you are agreeing to the use of Cookies. You can change your settings at any time. Cookie Policy.


10 Best Practices in Customer Behavior Segmentation

Originally published April 2, 2010

As organizations mature in their customer data collection efforts, they are exploring more ways to monetize and unleash the value locked in terabytes of raw customer transactions stored in data warehouses, data marts and operational data stores. One of the first steps they can undertake is to move beyond reporting and the slicing of customer key performance indicators (KPIs) across various analysis dimensions and into true data mining.
 
Behavioral segmentation is an initial data mining application that can be deployed on a customer data hub. Customer behavior segmentation can reveal insights as to what common customer behaviors are observed in different sectors and what drives these behaviors.  In retail, this could be accomplished by segmenting shoppers by visit and purchase behaviors.  Airline travelers may be segmented by frequent travel behaviors and frequent flyer point redemption patterns. In the banking sector, borrower credit card usage and payment behaviors can be segmented, while call detail records (CDRs) can hone in on usage behavior in the telecom industry. The output of a customer segmentation exercise provides a perspective on various common customer behavioral clusters and the impact of these well-identified discrete sets of behavior on revenue and profit.

However, one needs to be wary of the project land mines along the journey to customer segmentation. In this article, I examine 10 crucial best practices distilled from the author’s experience in customer segmentation engagements in both emerging and advanced markets. (See Figure 1.)

Figure 1: Best Practices for Customer Segmentation

Segmentation Best Practice #1: Using Segmentation to Gain Clarity on Customer Behavior Questions

The practice of segmentation can answer multiple customer behavior-related questions such as: Are we trying to identify the drivers responsible for dropped calls and billing errors for certain customer segments who are switching carriers in the prepaid mobile segment? Are we trying to model segment migrations as a result of a coupon redemption campaign to loyalty cardholders in retail? What are the top 5 behaviors that distinguish one behavioral segment from another? Which variables do not seem to correspond with any segments?

One of the initial tasks of the business analyst in this assignment is to engage the customer in an intensive workshop to identify the various flavors of business questions that can be answered using segmentation. The diagram in Figure 2 is a sample deliverable which illustrates the possible business questions that can be answered using segmentation to understand shoppers’ visit and purchase behaviors using point of sales (POS) and loyalty card data.

Figure 2: Sample Questions that Could Benefit from Segmentation

Segmentation Best Practice #2: Too Many versus Too Few Behavioral Dimensions

It is important to identify the specific customer behaviors that can be measured and fed as an input to the segmentation process. At this stage, a business analyst can map various dimensions of customer behavior that need to be taken into consideration. For example, two important behavioral dimensions to track to understand shoppers’ motivations in the retail sector are visit behavior (identified by time of day and day of week that the visits take place) and shopping behavior (identified by the shopper’s spend dispersion across categories purchased during the trip).

Having too few behavioral dimensions makes it difficult to holistically understand the customer, while too many behavioral dimensions tend to cloud the analyst’s mind. There must be a reasonable balance between the 2 extremes. Here is where the analyst’s domain experience and deep knowledge of the industry and market helps to include the right behaviors to track and to exclude the ones not having an impact on the business questions being modeled. Figure 3 illustrates some behavioral dimensions considered in the credit card industry

Figure 3: Credt Card Industry Behavioral Dimensions

Take advantage of the resources that lie in front of you. Front line business folks can be great resources in defining the most important customer behavior. Since they interact with the customer on a daily basis, they may be able to bring to light behavioral nuances important to customer segmentation. These insights can reveal a lot about customers’ intentions, which may not be apparent to those who are removed from the front lines.

Segmentation Best Practice #3: 5-Step Variable Selection Process

Once you have narrowed down the behavioral dimensions of interest, you need to drill down and identify specific customer attributes that would measure the behavior of interest. These behaviors could be:

  • Customers having more outbound calls than inbound calls
  • Responders to a campaignurchase of a product with the lowest price point in the category

Figure 4 is an indicative catalog of customer behavior measures used to understand subscriber behavior in the telecom industry.

Figure 4: Telecom Behavior Dimensions

Once these variables are identified they are fed through a 5-step funneling and variable filtration process, which ensures that the right variables end up in their corresponding segmentation pool as shown in Figure 5.

 

Figure 5: Final  List of Variables

Filter 1 – Availability. Are there transactions available to derive this behavior? If you want to tag a shopper as a morning shopper or late evening shopper, you need the timestamp of the bill from the POS transactions. If it is not available, then this variable cannot be added into the pool of variables for segmenting shopper behavior.

Filter 2 – Data Hygiene. If you need to add the age of the shopper and if the date of birth field in the loyalty card details is missing, then you will have to eliminate the shopper’s age as a variable from the segmentation scheme.

Filter 3 – Actionability Test. Essentially, you must question whether variances in the customer behavior result in differentiated outbound action. Variances in spend dispersion across categories can influence targeting of customers for a coupon campaign through which experimentation of categories not shopped before can be stimulated by coupon redemption.

Filter 4 – Variable Dispersion Test. Does the customer behavior variable have a huge spread or scatter factor to be used for segmentation? If most shoppers seem to have a shopping frequency of two visits per month, it does not make sense to include it in the segmentation pool, as this variable is not going to disperse the segment for differential action.

Filter 5 – Correlation Test. Identify whether there is any correlation between variables in the variable pool. In one retail segmentation exercise, it was found that the average purchase value per visit was strongly correlated to the number of visits per month. In this case, the business analyst chose to remove one of the variables after consulting with the customer team.

Consider a variable for the segmentation process once it is determined that the variable is available, believable, actionable, dispersible and not correlated.

Segmentation Best Practice #4: Choosing the Right Clustering Technique

Once the variables have been identified that are ripe for the segmentation process, it’s time to identify the appropriate clustering technique to disperse and group customers in the “N” dimensional behavioral space. There are multiple algorithms for segmenting customers like k means, orthogonal clustering and hierarchical clustering. Depending upon the platform, all or some of those clustering techniques may be supported. For example, Oracle Data Miner supports orthogonal clustering and k-means clustering. SAS Enterprise Miner supports hierarchical clustering and k-means algorithms. A framework to identify the appropriate clustering technique needs to be evolved taking into consideration the business objectives of the segmentation exercise and the reality of data constraints (categorical versus numeric data, missing values, etc.). Figure 6 is an indicative segmentation technique selection template to help narrow down the right clustering technique.

Figure 6: Determine the Appropriate Clustering Technique

Segmentation Best Practice #5: Iterate, Iterate, Iterate

Segmentation is an iterative process in which evolved segment formations need to be observed and recalibrated. It is akin to tuning an analog radio where you bring the needle to a zone where a signal is heard and slowly fine-tune it to maximize the signal and reduce noise. Similarly, after observing the initial clusters of customer behavioral patterns in each segment, you need to assess the segmentation parameters and decide how to recalibrate the segmentation process (see Figure 7). In the screenshot in Figure 8, you see the output of clustering stores using Oracle Data Miner to assess model effectiveness. It shows confidence, which is essentially a measure of the probability that members in the segment adhere to the business rule expression. For example, Cluster 3 has the maximum number of store members adhering to the business rule. Support indicates the number of members adhering to that cluster.


Figure 7: Iterative Segmentation Process


Figure 8: Output of Clustering Stores

Once the statistical process for clustering customers is run, it creates clusters that are statistically significant. But are they significant from a business perspective? In order to find out if a segment is significant from a business perspective, it pays to ask the following questions:

  • Do you have enough customers in a segment to warrant a marketing intervention?
  • Is the intrasegment variance low (measure of segment homogeneity)?
  • Is the intersegment variance high (do they vary in the spend dispersion across categories purchased, in the visit frequency or channel used to purchase, etc.)?

Once the above questions have been sufficiently answered, a determination can be made on whether or not there are customer behaviors that are both statistically significant and important enough from a business perspective to warrant an intervention.

Segmentation Best Practice #6: Make It Real. Evolve Segment Personas to Evoke a Mental Picture of the Segment

Once the segmentation exercise is complete and there are significant clusters and customer memberships in each cluster, it is time to make the segments “real” to the business users so that they are able to get a mental picture. List a couple of customer names and their behavioral profiles along with the statistical variables so that they can get a mental picture of each segment, as illustrated below.

Segmentation Best Practice #7: Overlay Customer Behavior and Segments Geospatially

Enhanced visualization can greatly benefit business users in the detection of affinities between customer behavior and the geographical location where they interact with the business. This can be done simply by geotagging each customer. Create a simple map of the customer, the segment he or she belongs to and the customer’s ZIP code. Once this is created, this information can be overlaid on Google maps so that you can visually identify any concentrations of certain behaviors in certain regions and ZIP codes as shown in Figure 9.

Figure 9: Geotag Customer Behavior

Segmentation Best Practice #8: Overlay the Results of Text Mining from Sentiment Analysis on Customer Behavioral Data

While customer segmentation works on structured behavioral data to clue organizations into customer behavior, unstructured text mining can provide a clue as to what the customer actually feels about his or her experience. In a recent exercise, I segmented customer behavior for a large U.S.-based card company and overlaid the key themes emerging from sentiment analysis through the text mining of inbound customer calls to 1-800 numbers on top of it. This allowed the business to correlate customer behavior with the underlying themes in conversations. The top 5 keywords used frequently by each behavioral segment can reveal a lot about the service levels and product coverage that influence the behavior reflected in the segment classifications.

Segmentation Best Practice #9: Actionability

For each segment, determine the important actions to ensure customers move toward the desired behavior. Figure 12 illustrates a segment treatment strategy that was evolved to target post-paid mobile subscribers whose data usage was greater than voice usage (“Gizmo geeks”).

Figure 10: Segment Treatment Strategy

Segmentation Best Practice #10: Track and Quantify the ROI of Segment Migrations

The segmentation exercise does not end with the unearthing of buried customer behavior segments. Once the segments are harvested, it is time to track segment migrations and the effect these behavioral migrations have on revenue and the company’s bottom line. The stimuli which cause segment migrations could come in the form of an online campaign or targeted relationship calls to cross-sell services that are most likely to be consumed. The following template provides a framework to capture the essential elements of an ROI framework which has segment population targeted, desired response from each segment, and the cost/revenue recognized from segment interventions.

Figure 11: ROI Framework

These 10 best practices, if followed rigorously, have the potential to reduce the risks inherent in a customer segmentation project by more than 80%. Creating behavioral maps using segmentation to understand the customer’s portrait is an important first step in an organization’s journey to derive value from its customer repositories. The 10 key segmentation best practices detailed here can serve as the guiding light to navigate a first-time customer segmentation initiative through the perilous land mines that are inherent in a customer segmentation project.

  • Derick JoseDerick Jose

    Derick Jose is the vice president of Advanced Analytics/Research within MindTree's Data & Analytic Solutions (DAS) Group, one of the world’s largest information management practices, which offers customers a one-stop-shop to capture, analyze, enhance, and view their business information. The DAS practice combines MindTree’s proven analytics, business intelligence, information management and research services for customers in the consumer packaged goods (CPG), retail, financial services, insurance, travel and media markets. Derick has 20 years of experience spanning consulting, advanced analytics and business intelligence solutions. He has worked extensively in the CPG, banking, telecom and retail industries. Derick can be contacted at Derick_Jose@mindtree.com.

    Editor's Note: More articles and resources are available in Derick's BeyeNETWORK Expert Channel. Be sure to visit today!

Recent articles by Derick Jose

 

Comments

Want to post a comment? Login or become a member today!

Be the first to comment!