In my most recent post, I addressed the strengths and weaknesses of marketing personas and analytic segmentation. To continue that conversation, I would like to step back and address a more common database marketing tool: clustering. Keep reading for a brief overview of what cluster analysis is, its strengths and weaknesses, and how it compares to profiles and models.
What is Cluster Analysis?
Clustering is the process of taking a database (usually of prospects) and dividing them into groups according to common characteristics. Cluster analysis can provide insights into how a particular population may be similar to each other. Many of the largest and most well-known compilers offer clustering systems, such as PRIZM, Equifax, KBM, and others.
Cluster analysis is typically done using a block group approach (ZIP+4). The fundamental assumption for many clustering systems is that “birds of a feather flock together,” meaning that there are similar demographic attributes of people in similar geographic areas. Not all clustering systems are geographically constrained, although most clustering systems are still based on U.S. Census data.
The primary method of creating a clustering system is to append data. Lifestyle data, demographic data, and financial data are most often used to create clusters. This data is then analyzed and descriptions are created. Clusters are indexed to allow comparison of one cluster to another.
Historical Use of Clusters
Traditionally, the primary method of utilizing cluster analysis was to code a mail file and then use the cluster as a post analytics tool to determine whether certain clusters were more responsive than others. This typically resulted in sending more mail to responsive clusters and less mail to non-responsive clusters. This is similar to how a model is used today, but clustering is drastically different from true models in both performance and complexity.
Models are developed on record level data and use multiple factors to define and score prospects or customers and include extensive amounts of statistical analysis and probabilistic science. As such, a model can help you not only understand who responds or buys, but potentially predict response rates and much more. A clustering system can be one variable input into a targeting model.
How Models and Clusters Can Cooperate
Today we find that rather than using a cluster analysis approach as a targeting methodology to define responsive and non-responsive groups, we tend to use it as a proxy for demographic or audience insight that is not necessarily predictive. We prefer to let models speak to predictable results and clusters speak to creative definition when analytic demographic segmentation is lacking.
This method allows us to create and score a model that will define a target audience and cross-tab that audience with additional descriptions of that group. In short, a clustering system becomes a proxy for an analytic segmentation system applied to a modeled target audience.
Three Ways to Make Cluster Analysis Work for Your Direct Marketing
Cluster analysis is just one of several methods marketers have of adjusting the scope of their campaigns to more effectively reach the audience they want. Here are the three most effective ways we have found to use cluster analysis:
- The historical method, as an early state substitute for a model (when no other analytics are available);
- As a method to define messaging to separate groups, when record level or analytic segmentation is not available;
- As a cross-tab against an already modeled and segmented universe.
Models are the most appropriate way to target prospects and to rank order them for efficiency. Clusters can be a viable way to approach an audience, but pale in comparison to models and can be best combined with models to help creative messaging.
In my next post, I will spend some time taking a deeper look at personas and models. Until then, if you want to improve the performance of your direct marketing, IWCO Direct can help. Get in touch with me to learn more about how we integrate cluster analysis, modeling, personas, and more to boost response rates and ROMI.
Subscribe via email to our Stevie® and Feedspot award-winning blog and get a fresh post delivered weekly to your inbox. We promise to keep it interesting, but you can easily unsubscribe if we don’t.