Get in touch with us. We'd love to hear from you.
Utilizing Segmentation in Digital Analytics
This blog provides a brief overview of segmentation, discusses some best practices and provides tangible examples. We’ve kept things tool agnostic so we can focus on the fundamentals.
What is segmentation?
Segmentation is a systematic process of creating meaningful subsets of the total population based on predefined Factors or Attributes. Segmentation allows marketers or the leaders in the decision chain to parse the data on a more granular level. Doing so provides a much better insight on how each of these segments is interacting with the digital asset, be it a website or an application. For the ease of reference we will only discuss segmentation in context of website performance but the underlying fundamentals are the same.
The best way of answering this question is to provide a simple example. Let’s say you are in charge of an e-commerce site that sells over 150 products. It goes without saying that not all site visitors are created equal. Some visitors convert (in our example, purchase) upon the first visit to the site, some visit the site multiple times before making a purchase, and some visitors only purchase when provided a coupon or promotional content. As the person in charge you need to understand how these groups differ – in relation to the site goal(s). Figure A-1 depicts this hypothetical (and simplified) example.
The next step in the segment analysis would be understanding how these segments are different. The ultimate goal is to move the visitors from C to B and finally to A. To do so you need to understand the unique attributes that define each of the segments.
The diagram in figure A-2 depicts each of the 3 groups mentioned above. However at this level we start associating these groups with different attributes. Please note that attributes are not mutually exclusive to each group. For example, ‘Group B’ has elements that are influenced by ‘attribute 2’, which is also an influence factor for Group A. The reason is that in real life examples there is always some overlapping amongst groups (note that we are not using the term ‘segment’ yet). To overcome the overlapping issue some mathematical procedures could be performed to assign weights on the ‘Attribute’ level to normalize the data. One of the most common practices is to create a standardized weight system by developing multiple regression models that can explain the in-group variations by a linear or non-linear relationships between attributes. We will explore the applications of different forms of Regression Analysis in developing predictive models in a separate blog. Now, back to our hypothetical example; we are on our way to create our segments. After assigning appropriate weights to normalize the effect of attributes we end up with a model like Figure A-3.
Now that we have a better understanding of the basic principles of segmentation let’s take a closer look at ‘Attributes’. There are some general rules for studying attributes. By definition, all attributes should be mutually exclusive. If you find an attribute that doesn’t satisfy that requirement it is advised to combine the overlapping attributes and create a dummy attribute that contains the overlapping attributes. In these cases special attention needs to be given to ensure the weight system for the entire attributes pool is adjusted or you will run the risk of multicollinearity. In this context, multicollinearity is referred to as a scenario where there is a linear or near linear correlation with standardized attributes and if the weights are not adjusted properly (in case of combining attributes) you run the risk of miscategorizing visitors into the wrong segment.
Attribute selection is the most important part of the segmentation process. For our hypothetical example, the marketing manager needs to understand how visitors from different sources (think multi-faceted campaign programs) are interacting with the site and converting site goals. In this example the mode of arrival to the site could be an attribute. Then the marketer can segment the traffic based on how the visitors got to the site. On a higher level he or she could compare the effectiveness of channels by each segment. For example the marketer could find that visitors that come from banner ad campaigns have the highest conversion rate. Then further analysis could be performed on that one segment alone. The marketer could evaluate the performance of each creative or banner type. Another attribute could be the ‘Device’ used to access the site (Mobile V.S. PC). Some other attributes like Geography also could be used if appropriate for the business. For example if the business is conducting hyper geo targeting they might want to know how each location is performing.
Segmentation Scorecard Development
Once the business has identified some segments of interest it is advised to track these segments over time. The best way of doing this is by using scorecards. As the nature of each business is different so are the forms of the scorecards needed to track segments. Figure A-4 is an example of a Segment Scorecard:
The scorecard example measures the following elements over time and presents the data in a coherent and easy to understand manner. It is recommended that these scorecards get updated at least once a month and have quarterly reviews. The data elements presented here are as follows (and feel free to change the report elements based on your business’s needs):
- Rank: The rank of the segment for the reported time period (based on monetary evaluation scale or other methods)
- Proportion: this should be represented as a ratio of
- Weight: is an optional metric but in cases where there are a lot of disparate segments one needs to understand the relevance of the segment to the entire traffic.
- Site Metrics: These are the overall clickstream data for the segment. These metrics could either be presented alone or with the % change from the previous time frame (i.e. last week, last month, last quarter ….)
- On the right pane; each conversion for the previous time period is listed and in the yellow box all the current conversions with % change from last period.
This format provides a mechanism to track and monitor the segments of interest and make timely adjustments if needed to ensure the growth of the segment and by extension the conversions on the site.
To summarize, we looked at some fundamentals of segmentation and some high level explanations and a few practical examples. The goal was to showcase the power of segmentation as a tool available for the decision makers to modify / fine tune the strategy based on the interaction of the segments of interest with the site.