The Importance of Retail Customer Segmentation
Even in a digital age when consumers can be targeted at an individual level, customer segmentation strategies have weathered the test of time. Need proof? Think millennials. This generation-based segment is just about as modern as it gets, and it’s the most talked about segment in marketing since baby boomers entered the scene. The point is that segmentation is still a valid technique to use for improving marketing ROI.
Of course, there are more sophisticated ways to generate market segments than simply grouping consumers by generation. Need more proof? Consider this excerpt from an article entitled “New Criteria for Market Segmentation” by Daniel Yankelovich that appeared in the Harvard Business Review circa 1964:
“The director of marketing in a large company is confronted by some of the most difficult problems in the history of U.S. industry. To assist him, the information revolution of the past decade puts at his disposal a vast array of techniques, facts, and figures. But without a way to master this information, he can easily be overwhelmed by the reports that flow in to him incessantly from marketing research, economic forecasts, cost analyses, and sales breakdowns. He must have more than mere access to mountains of data. He must himself bring to bear a method of analysis that cuts through the detail to focus sharply on new opportunities.”
Sound familiar even today? That’s because Yankelovich was describing a method that today we call segmentation analysis.
The Segmentation Schema
When discussing advanced customer segmentation techniques, what comes to mind? Certainly there are “mountains of data” to consider and aggregate (we’ll get to that later). But first and foremost, it’s the overall marketing objective that guides the segmentation schema and technique to implement. The word schema is defined as “a structured framework or plan.” Think of the segmentation schema as the approach to be utilized. A business objective could be as follows:
OBJECTIVE : Partition existing customers into finer segments for development of more relevant offers and promotions.
As a side note, it’s equally important to be sure there is buy-in from project stakeholders. The purpose and success factors driving the segmentation schema must be clearly articulated to those within the business who have a vested interest in the project. These stakeholders need to reach agreement regarding the approach and business outcomes for the segments. Incorporating their feedback will be crucial to gathering final buy-in.
Scientific and Human Elements of Retail Customer Segmentation
The technical building blocks that go into the schema represent the scientific element. This is the period when the analyst encounters mounting issues regarding data to access, filters to use (scrubbing and screening the data), measurements to produce and other specifications required for preparing the data for analysis.
The implementation of a final solution and strategy represents the human element. This is where the right brain kicks into action. It’s the period when the marketer engages in ways to humanize the segments from a more creative perspective. This includes asking:
- What do these segments look like?
- How should we communicate with these segments?
- What might make these segments unique from a demographic, behavioral, and intellectual standpoint?
There will also be some crossover between the scientific and human elements that takes place as the schema is fleshed out. This is the period when the analyst and marketer are linked together, each considering the interests of the other as the project progresses. When the left hand (analyst) doesn’t know what the right hand (marketer) is doing, then there is potential for a dysfunctional outcome.
Data for Retail Customer Segmentation
Earlier we referred to the “mountains of data” available for segmentation. In reality, it’s more like an avalanche of data with slabs of characteristics that describe customers/prospects with respect to:
- Who they are (demographics)
- What they do (lifestyle/activities/interests)
- How they act (behavioral)
- What they buy (purchase preferences)
- What they value (attitudinal)
- … and so on
At the onset of a segmentation plan, it’s a good idea to conduct an inventory of available data. Case in point, many years ago I would introduce segmentation concepts with an example designed to classify people into different types of fish. Using a simple questionnaire consisting of five questions with two possible outcomes for each question (for example, are you a day person or a night person?), there were 32 segments that could be created. Compare this to all the data available today and the chaos it can inflict when trying to translate the information into a manageable number of segments.
Advanced Customer Segmentation Solutions
Now, let’s consider some of the more advanced schemas and solutions that retailers might leverage.
Cluster Analysis (K-Means)
The main idea of cluster analysis is to identify groupings of objects (customers) that are similar with respect to a collection of characteristics and are as dissimilar as possible from an adjacent grouping of objects. The number of clusters is dependent upon the type of algorithm used to identify the clusters. The characteristics can be a set of data elements that describe each customer.
Within a cluster analysis, one widely used methodology for datamining is K-Means Clustering, where K represents the number of clusters to be created. Each cluster is centered around a particular point called a centroid. Think of the clusters as planets in multi-dimensional space, with the number of dimensions equivalent to the number of data elements and where each object has a center point. The center point is made up from the average values of the data elements making up the clusters.
In the end, the K-Means solution might yield five manageable segments. A customer is assigned to a segment by calculating a distance measurement to each of the centroids. The score is commonly calculated from the Euclidean distance. The centroid with the closest distance becomes the customer’s home segment.
K-Means Clustering Analysis Diagram
CHAID (chi-square automatic interaction detection) or Tree Analysis starts with partitioning each object (customer) into one of two outcomes (this is the dependent variable). Examples include response to a marketing promotion (0=No, 1=Yes), purchasing a specific product or making a repeat purchase. This is the starting point for the tree.
Other characteristics (the independent variables) are used to further partition customers. Each partition creates a new branch of the tree. For example, customers who responded to a marketing promotion might be significantly more likely to fall within a specific age range (say, age 24 to 30) compared to customers who did not respond.
Further analysis might show that customers age 24 to 30, living in urban markets, who do not have any children living at home, are even more likely to respond. This creates a new branch of the tree (perhaps called “Childless Metro Millennials”). The branching process continues until there are no more statistically significant partitions that can be made. Each ending branch (called a node) represents a customer segment. The number of final nodes depends on how well the independent variables are able to differentiate the behavior under study.
As an example, below is a tree showing survival of passengers on the Titanic (“sibsp” is the number of spouses or siblings aboard). The figures under the nodes show the probability of survival and the percentage of observations in the branch. The tree analysis shows that your chances of survival were good if you were part of segment (1) a female or segment (2) a male older than 9.5 years with three or more siblings.
Tree Analysis: Titanic Survivors Example
The concept of neural networks for segmentation is derived from an imitation of neural networks in the human brain. In short, it’s a sophisticated version of connect-the-dots, where the algorithm deduces certain outcomes based on multiple layers of input data.
Neural Network Example
A neural network must be trained by being presented with numerous examples of the behavior to be understood. Much like the human brain considers a multitude of inputs to come to a conclusion or make a decision, the neural network algorithm connects all the available data in order to recognize patterns in the information that can be most associated with an expected outcome.
Some applications of neural networks include:
- Creating buyer segments from the total population of consumers
- Developing customized content for specific segments
- Making predictions regarding certain customer preferences/behaviors
No matter what schema is used, the resulting segments should be validated before being implemented into any type of marketing campaign. Validation is a reality check to make sure the segments hold up in a real-world setting. Some schemas allow for the use of a “training” dataset, used for developing the segmentation model, and a “validation” or “test” dataset, used for testing the overall accuracy of the solution. The training and validation datasets are mirror images of each other with respect to the input data and the output behaviors. This ensures the statistical relevancy of the end solution.
Implementing Retail Customer Segmentation
Now that the solution has been validated and is ready for action, it’s time to set parameters for implementing the segments into your marketing strategy. Let’s go back to the sample objective we stated earlier: to partition existing customers into finer segments for development of more relevant offers and promotions.
Specific promotional offers can now be targeted based on the composition of each segment. For example, offers developed for “Childless Metro Millennials” should emphasize particular product features that can be differentiated compared to offers for other segments. The impact of a marketing plan that incorporates the segments should be tested against a more generic strategy. This allows the ROI to be quantified against the cost of building the segments.
Away We Go
No segmentation solution is perfect and, of course, there is a limit to the amount of resources, time and money to be allocated to a segmentation strategy. But hidden inside and beneath our customer data is the probable answer to some of the most complex problems that a marketer needs to solve. We have at our disposal many alternatives to “master the information” and move in a direction that “cuts through the detail,” as Daniel Yankelovich first suggested. Segmentation analysis, in its many forms, is a track to continually keep in mind for more focused and relevant target marketing.
As Principal Analyst for IntelliStats Analytics Solutions, Bill Schneider leads internal teams and works with external business partners, including CCG, in the development of advanced analytical solutions that utilize transaction level data analysis, predictive modeling, consumer segmentation and other emerging marketing science techniques. He has worked with clients across a wide spectrum of industries including retail, financial services, sports and recreation, telecommunications, hospitality and non-profit organizations.
Let our experienced retail marketing group help you leverage your data to provide quantifiable, actionable results. Our data and analytics services include customer segmentation and cluster analysis, qualitative and quantitative research, modeling and marketing performance analysis. Click below to schedule a free consultation with one of our retail marketing experts or call 800.525.0313 today.