|
MARKETING PROFESSIONALS ARE AWARE that customer segmentation (also known as "clustering") is a popular advanced descriptive modeling technique. |
Clustering goes beyond profiling (See Profiling - Seeing Customers in New Ways) and performance segmentation tools like RFM, RFA or RFP analysis (See Segmentation - Finding Your Best Customers). Clustering incorporates many dimensions of customer behavior in the analysis -- not just Recency, Frequency and Monetary, Average Order or Profits.
Because cluster analysis simultaneously evaluates everything you know about your customers, it can reveal useful insights that simpler tools cannot uncover. Clustering can be an effective tool for:
- Moving from a mass marketing or "brand awareness" approach to highly targeted advertising.
- Tailoring product offers to specific customer segments based upon likely need.
- Choosing appropriate media and distribution channels to reach each customer segment.
- Designing the most relevant "look and feel" for creative communications and advertising.
WHAT IS CLUSTERING?
A "cluster" is a descriptive modeling tool. That is, it describes (but does not predict) how a segment of customers behaves as a group -- and how groups of customers are alike and different. Clustering organizes customers who are to some degree alike into segments ("clusters") that that are more or less different from the other customers hwo are grouped into other segment clusters.
There are two very different ways to define a "cluster": statically and dynamically. Both approaches have their uses. Choosing the one appropriate for your specific situation requires some careful thinking.
WHAT ARE STATIC CLUSTERS?
Static clusters assume that people who live close to each other are pretty much alike. Static clusters describe relatively large groups of individuals upon the basis of where they live geographically as defined by zipcodes and/or carrier routes. There is in fact a general tendency for people to choose to live in neighborhoods that are full of people very much like themselves. So, this approach works when neighborhoods are largely homogeneous in terms of income, education, ethnicity and spending habits.
An example of static clusters are those defined by the PrizmT cluster methodology. This clustering technique defines 62 static clusters, each with an intuitively descriptive name like "Upper Crust", "Old Comrades", "College Students" or "Hard Scrabble". Every zip+4 neighbornood is assigned to one of these 64 clusters based upon an analysis of census tract data and other variables. The result is that everyone living in each zip+4 neighborhood is assigned to the cluster having characteristics most like those of their neighborhood.
Historically, this approach has worked fairly well because most neighborhoods were not very diverse. More recently, however, cultural and economic diversity has been increasing across all cities and neighborhoods. Today, many of the people living in a zip+4 neighborhood assigned to the "Country Cousins" cluster may be quite different from each other in terms of attitudes, interests and purchasing preferences.
Further, the description, "Country Cousins," is based upon a generalized overview of the people assigned to that cluster. So, static clusters thus tend to gloss over many important differences among the people assigned to the cluster. As with freezing water and boiling water, referring to the average temperature as "temperate" is misleading.
WHAT ARE DYNAMIC CLUSTERS?
In contrast, dynamic cluster models do NOT pre-define customer segments. Dynamic clustering starts with a list of a company's specific customers. Census data and information purchased from list compilers is then matched to the company's customer list to expands this basic information to include demographic data such as age, income, marital status, etc. Sometimes psychographic data such as hobbies, lifestyle, and preferences for products, services and brands are added to develop a more complete "picture" of the customer. This is called "overlaying."
Finally, a sophisticated statistical technique called "K-means cluster analysis" is used to define which factors best describe each cluster ("segment") of customers. Often, only five or six clusters result - but these customer segments can be very different from one company to another! Dealing with only five or six customer groups also greatly simplifies campaign management, media selection and offer creation. The result is higher campaign response rates and improved marketing ROI.
WHICH KIND OF CLUSTERING IS BETTER?
Dynamic clustering specifically analyzes a company's own customers and it does it at the household level. So, differences between neighbors are fully recognized. For example, young families living near older neighbors or singles living among married couples are assigned to different clusters. As a result, the dynamic clustering approach is far more descriptive of a company's actual customers than "static" approaches that rely only upon neighborhood averages.
Although dynamic clusters often indicate which customers are likely to be "best", they are not especially useful for selecting mailing lists to purchase. To use a dynamically-defined cluster to purchase or rent a mailing list, you must obtain a “master list” which includes demographic data, usually for a specific geographic area. Then the demographic data of the master list names is matched to the names in each dynamic cluster to derive the demograhic profile for the cluster. Frankly, this can be an expensive and complex process.
In contrast, Static cluster models are designed primarily for directly selecting rental names from a vendor's database. So if your objective is primarily to locate a suitable list of names to rent, static clusters offer an advantage.
WHAT'S THE BOTTOM LINE?
Both static clustering and dynamic clustering are useful descriptive modeling tools. Each helps in designing and implementing an effective marketing plan. Static clusting is usually better for selecting mailing lists. For example, static clusters are good for retailers wanting to contact prospects in areas around specific stores. Further, static clustering offers the advantages of relatively low initial cost.
Dynamic clustering is generally superior to static cluster analysis for direct marketers, however. Dynamic clustering provides insights for designing offers and campaigns that are very specific to a company's unique customer segments. Although it costs more, dynamic clustering is more powerful for aggressively controling campaign costs and boosting campaign response rates. These advantages almost always more than offset the additional work needed to perform dynamic clustering.
Remember, though, that all descriptive tools are usually less effective than predictive tools for making targeted marketing decisions. Learn more about predictive tools in Prediction - Finding Gold in the Data.
Questions? Just ask! [Back to How-to Resources] [Top] |