Data-Driven Marketing Analysis: Profiling, Performance-based Segmentation and Modeling
Data-driven marketing analysis can be bewildering; the multiple tools available for collecting and interpreting data and putting them to use to shape new campaign strategies at lower cost and increased revenue leaves many marketers confused as to which approach is right for their organizations. Marketing professionals want tools and techniques that will help them make better decisions about what prospects or customers to contact, how to contact them, and what offers will yield the highest response and revenue.
The discussion below highlights four available analytical tools or techniques of increasing complexity and sophistication for interpreting data in an organization’s customer base, and for identifying its best customers (Profiling, Performance Segmentation, Descriptive Modeling, and Predictive Modeling). Professional database and modeling experts can help marketers understand how these tools fit together and assist in the choice of the most appropriate approach for their organizations’ specific situation.
Profiling
Most companies know the overall average demographics, lifestyles, and psychographics of the customers that comprise their database. But these averages usually mask important differences among customers. For example, the average customer may earn $50K/year, but there may be a group of customers who earn more than $90K/year, and another group of customers who earn less than $25K. These two groups are quite different from those with the $50K average income.
Profiling describes such differences with "cross tabulations." With this tool, customers can be profiled along many dimensions, including age, income, and number of purchases. Sometimes a simple comparison to outside data, such as a national average, is included. The results can be presented as a table, but are often shown as a histogram, like this one:
This kind of data-driven descriptive analysis is easy to conduct and is a good and helpful starting point. However, profiling is limited to examining just one or two customer attributes at a time. When grouping customers by three or more attributes at once, the tables and graphs quickly become complicated and confusing.
Performance-based Segmentation
Performance-based segmentation takes profiling to the next level of complexity. Among its tools are three 3-dimensional techniques, based on Recency of purchase, Frequency of purchase, and Monetary value of the purchase (represented by the initials RFM).
RFM - RFM organizes customers into groups according to their Recency, Frequency, and Monetary commitment. These variables are important because they predict future behavior:
- Customers who bought recently are more likely to buy again.
- Customers who bought more often are more likely to buy again.
- Customers who spent more are more likely to buy again.
RFM-A incorporates the average purchase amount instead of total purchases, making it possible to identify individual purchase behavior versus overall spending.
RFM-P monitors specific products or profit instead of aggregated total purchases. This approach enhances the basic RFM analysis by recognizing that not all products sold by a company are equally profitable.
- Visual RFM - Target Resource Group has developed a unique approach to displaying the results of the RFM technique. It is called Visual RFM™ because it provides “at a glance” the more complex relationships revealed by the basic analytical tools.
- Advocate, Buyer and Tryer Analysis - The most sophisticated performance-based segmentation technique is Advocate, Buyer, and Tryer (ABT) Analysis™. Also developed by Target Resource Group, ABT Analysis provides a multi-dimensional tool to identify and describe customer segments. This data-intensive approach segregates customers into groups of Advocates, Buyers, and Tryers and identifies ways to increase profits from existing customers, improve customer retention, and direct sales efforts towards the most promising prospects. (For more detailed information on ABT, see TRG White Paper: “A Deeper Look at Database Segmentation”)
Descriptive Modeling
Descriptive modeling is an effective tool for revealing how to reach customers and what to offer them. The two approaches to modeling are both based on a descriptive statistical technique called “Cluster Analysis.”
Static Clusters - A static cluster is a collection of many characteristics that are observed to fit within one pre-defined group within a physical neighborhood. One popular list service (Prizm™) has defined 62 static clusters based on attributes including average income, average age, and average number of children. Prizm assigns every neighborhood in the United States to one of these 62 predefined clusters on the basis of zip code.
Static clusters have a list of attributes grouped together under names such as “Money & Brains” or “Upward Bound.” The underlying assumption is that if a customer lives in the zip code, he or she is like everyone else in that cluster. Sometimes this is true; more often, it is not.
Dynamic Clusters - A dynamic cluster is defined only after analyzing actual customers. The process assembles customers into clusters, based on their specific demographic, lifestyle, and psychographic characteristics at a household, not zip code, level. Dynamic clusters are defined by specific customers, thus creating more accurate, descriptive, and precise clusters.
Dynamic cluster analysis typically reveals that a company’s customers can be very accurately grouped into far fewer clusters—five or six, instead of dozens, as is the case with Prizm. This simplifies the job of selecting the appropriate medium and crafting a message best suited for the specific need.
Predictive Modeling
Predictive modeling is an extension of the descriptive modeling process, and is used to determine how likely customers and prospects are to respond (response rate) or the degree to which they will likely respond (how much they will spend). This knowledge allows marketers to focus their efforts on their most likely prospects and prospects most likely to purchase higher-priced product, and brings those efforts closer to a high return on the marketing investment.
Predictive Modeling is the preferred method for an organization to determine:
- Who are the best prospects for its promotions
- How many prospects exist that “look like” its best customers
- Which of its existing “smaller” buyers are most likely to become “bigger” buyers
- Which prospects are most likely to buy?
Predictive modeling uses the demographic and behavioral characteristics of a company’s best customers as a “model” for identifying qualified prospects from purchased, rented, or traded lists. The results of this process are based upon the organization’s data, which are, therefore, unbiased opinions that can be safely used to predict future prospect behavior.
Predictive modeling can use a variety of statistical techniques, including regression analysis, CHAID (Chi-Square Automatic Interaction Detector) and C&RT (Classification & Regression Tree), among others.
But above all, good data (i.e., accurate response, sales, and profit margin data) and a thorough understanding of customer behavior are essential for the success of a predictive modeling analysis.
Conclusion
Each of the analysis techniques outlined can play a part in a company’s marketing program providing useful knowledge about its customers and their behavior. The complexity of these analytical tools ranges from simple (Profiling) to complex (Predictive Modeling). And the cost of performing each type of analysis varies, in turn, with the complexity and amount of outside data needed.
It is usually best to start with a simple “cross-tab” profile analysis. Depending on the results of that analysis, more sophisticated analyses can be conducted, based on an organization’s budget and extent of marketing efforts. It should be remembered, however, that marketing analysis is both a science and an art, best accomplished with the assistance of experienced database and modeling experts.
To learn more about database analysis and modeling and how they can benefit your organization, please contact Will Lester, Vice President, Target Resource Group at 719.314.5835 or email wlester@trgarts.com.
