Blogging is a bit like exercise. Do it regularly, at a regular time and a regular way, and it doesn’t feel all that hard. But get out of the pattern, miss a bit of time through travel or sickness, and it feels like bloody hell when you get back to it. So a missing week (combination of business travel and family Christmas-stuff) has me feeling a bit like I’m running up-hill.
It’s lucky that at least I have a topic ready-to-hand – my promised exploration of some tactics in visitor segmentation. As I’ve written often enough, there is no single capability in doing real web analytics that is more important than segmentation.
Visitor Segmentation in web analytics is used for two main purposes. The first is to setup regular reports of key populations. If your site has distinct types of users (like guests, subscribers and premium subscribers), then reporting that describes the numbers and behavior of each type is essential. This sort of segmentation doesn’t have to be very dynamic. It can be done at the tagging level or in software and the segment definitions are usually obvious and fairly simple. I’m not going to spend time on this type of segmentation – it is supported by almost every kind of tool (including GA now) and its uses are well-understood.
The second type of segmentation is the dynamic creation of filters to try and isolate a population for analytic purposes. This type of segmentation is almost always ad hoc, requires significant tool capabilities, and can be quite a bit more difficult. But it’s a capability at the heart of real analysis in our field.
What I thought I’d do for these blogs is walk through some recent segmentation exercises (a little bit disguised but completely real) we’ve done for analysis purposes and show how we used segmentation tools to find an answer. I’m mostly going to draw examples from Omniture’s Data Warehouse and Discover but I may throw in a few NetInsight examples as well. We’re seeing Unica get some definite traction in the market and NetInsight is a very nice tool for this type of segmentation analysis.
Example: Catalog Searchers vs. Site Searchers
Use Case: The client distributes many printed catalogs and, like nearly all catalogers, allows visitors to enter a catalog id in the search field. Visitors are then taken directly to a product detail page.
Client Question: The client wanted to know how many visitors fit this use can if the process was working well. Key questions included:
- Were catalog visitors successful?
- What else did they do on the site?
- Do catalog visitors come back and if so do they come back and do catalog searches?
In addition, there were two types of catalog identifiers and the client wanted to know if there were differences between visitors from these two groups.
Measurement Issues: There were no distinct pages within the process. Catalog visitors used the home page, search and product detail pages. Fortunately, the client was capturing the search term in a variable and the catalog identifiers were distinctly formatted.
Tool: We used Omniture’s Data Warehouse tool for the analysis.
Methodology: Here’s how we built the analysis. We started with a simple visit-based segment that isolated the visits where a catalog search was made:
This segmentation allowed us to report on all the aspects of visits with a catalog search.
A simple segmentation like this let us answer the basic questions about how many visits included a catalog search and how many of these visits resulted in a sale.
One of the things we wanted to understand, however, was repeat visitor behavior for visitors who started out with a catalog search. To find out, we needed a more complex segmentation. One of the really cool features of the Omniture segmentation builder is the ability to define segments at the Visitor, Visit or Page View level. We’d started out with this:
But we added another segment that looked like this:
This filter allowed us to run reports on all the visitor behavior for any visitor that did a catalog search. One of our key questions, though, was the repeat visit behavior of visitors who start with a catalog search. This segment doesn’t answer that question – because it includes visitors who may have been to the site many times before doing a catalog search.
Our next segment combined Visitor and Visit Segmentations to solve this problem:
This segment let us look at all the visitor behavior of anyone who did a catalog search in their first visit to the site.
Incidentally, the segmentation builder also includes an Exclude capability. Excludes are absolutely essential in many filtering contexts – including building comparison populations. Using the exclude feature, I can create groups that DIDN’T have any of these specific behaviors and use them for comparison:
Once we built the segments, we used the Data Warehouse to generate the reports. We could have created ASIs, which give access to all the SiteCatalyst reports, but we were interested in some reports that can’t be accessed from SiteCatalyst. In addition, we weren’t interested in keeping these segments around and generating reports from them – we just wanted our information for the analysis.
Here’s one of the reports we used:
This report broke down the visit number, path length, and exit page for every visitor who started (Visit 1) with a catalog search.
Why this report? Well, I wanted to see how deep visitors who used catalog search went into the site and where they exited from – and I wanted to know if that behavior changed over time.
What I got was a classic pogo-stick pattern (I’ve summarized it here across Visit #):
Note how the Exit Page on even numbered path lengths is usually the Catalog Detail Page and how on odd numbered pages it’s usually the HomePage. It’s also interesting to see that there are quite a few Catalog Detail Page Searches deep into sessions. Visitors were bouncing back and forth from Home Page to Product Detail, doing one or more catalog searches, and then exiting.
The full analysis showed that catalog searchers averaged more than 3 searches per visit. It also showed that they tended to come back to the site and do more catalog searches. Not only that, there pattern of behavior didn’t change significantly in visits 2-x. That meant that even as users were more experienced, their catalog lookup experience wasn’t changing.
Conclusions and Recommendations: There’s no point doing an analysis if you don’t do something about it. In this case, we found that that catalog searches were fairly common and fairly successful. Trivial so far. But we also found that they tended to do multiple catalog searches in a visit and they didn’t tend to look at other products – and that this behavior didn’t change much over time.
When we looked carefully at the product detail page, we found some obvious reasons why this might be true. First, there was no easy way to do another search from a detail page. And you couldn’t go back to a product list to see similar products – because you hadn’t arrived by a list. The designers had assumed that people hitting a product detail page would be arriving by a list and would go back and forth to see multiple products. This worked pretty well for traditional navigators and searchers. But this use case bypassed the process.
To fix the problem, we recommended that this detail page included an easy method for doing additional catalog searches and that it also include a new module to recommend additional products via a product list.
Reflections: This was a significant use-case on the site – and it’s a common one across many multi-channel businesses. But it took some fairly robust segmentation capabilities to isolate the critical behaviors that really showed how these visitors were behaving and helped us figure out how to optimize the experience.
I think this example well illustrates some of the most important aspects of visitor segmentation: it needs to be flexible and software driven – no one is going to create this type of segmentation from a tag; it’s a big advantage to be able to segment on different levels – especially visit and visitor; and segmentation needs to provide access to rich reporting cross-tabulation (such as my visit-number by path-length by exit page report).
In my next posts, I’m going to show more of these examples to illustrate different aspects of web analytics segmentation.