Note: This post has moved from Leapthree.com to Ayima as part of the 2018 acquisition.
The first two parts of this series covered two key methods in analysing web analytics data, comparing performance against a set of reference numbers and trending data over time to expose patterns. This third and final part covers segmentation, what I would commonly refer to as drilling into the data. Writing on this does sadly date me to being nearly five years behind Avinash with my thinking. As he has written extensively about segmentation in numerous blog posts and books such as Web Analytics 2.0, this post will be more of an overview.
Segmentation is an integral part of analysing web analytics data. Web Analytics reporting contains aggregate numbers which don’t provide a true picture of the range of visitors who access a website. There is no such thing as the average visitor as reported in your monthly dashboard, accessing your website 2.4 times a month and viewing 6.2 pages each time taking them 4 min and 39 seconds.
Segmentation at its essence is very simple, you just look at a subset of data in an attempt to gain more insight. This is similar to someone going to the doctor, saying my head hurts and having the doctor ask “what part of your head hurts?” and “how bad is the pain?”. The doctor applies the segments of “part of head” and “intensity of pain” to aid in their diagnosis of the problem.
The number of ways that you can segment web analytics data is near infinite and only limited by your creativity. Some examples of segments are:
- Traffic Source – probably the most common segmentation to be applied, splitting traffic by the method used to access the website e.g. paid search, direct, email.
- Site Section – which content is most popular?
- Browser – knowing what browsers your visitors use is highly important, learning that your website has a significantly lower conversion rate in one browser can be vital.
- Country – you can only sell products to those countries you can ship to, does the data suggest you should be increasing this list?
- Search to Book days – how does the number of days between the date a visitor is searching for a hotel room and the date they require the room influence their likelihood to make the booking? And how does this impact your business strategies?
As with everything in web analytics, you can usually get an answer out of the web analytics tool three or four different ways. There are however two basic approaches to using segmentation to analyse web analytics data. The approach you use may depend on what is available within the web analytics tool you are using or on your personal approach to data analysis.
View multiple subsets at one time
The first approach involves splitting a set of data into multiple subsets, reviewing a limited set of metrics for all subsets in one go. This actually describes nearly every web analytics report. Your basic Traffic Source & Page Content reports are the set of all data segmented by traffic source and page respectively. In most tools, clicking through on one value to a new report is the same as applying a segment of that initial variable to the new report.
It is relatively straight forward to interpret segmented data when the dimension is something like traffic source or day of week as you only have a limited number of segment values. It becomes much more difficult when the number of potential segment values is very high. This can be the case for dimensions like Country where there are 100+ options or for numerical fields like the number of pages viewed or time on site where it is theoretically unlimited.
The solution is to group similar values. I discovered the scientific explanation for this from a recent Google Conversion post on using Information Architecture to improve conversion rates. It references Cognitive Load theory which explains how people can only retain 7 “chunks” of information in their memory at any one time. By grouping like values, data derived through segmenting on this dimension becomes understandable to users.
There are three ways to do this:
- Group similar values into categories
- Create chunks of values using some logical system
- Group the long tail in an “other” category
The first option relies on natural categories being available - an example for this was my recent post on Search Term Categorisation. If this is not the case, the next two options come into play with the decision depending on the distribution of values within that dimension. If only a few are heavily represented with a long tail, then represent those key values individually and group everything else under “other”. Otherwise create artificial categories through chunking values similar to what was described in this post for enhancing the reporting of numerical dimensions.
Isolate a particular subset of data
The second approach is to define and isolate a particular subset of data, applying that segment of data to multiple reports in order to understand performance. Most web analytics tools now offer a feature where you can define your own segments e.g. visits that spent greater than 4 min on the website AND viewed a product page AND didn't use internal search.
An alternative is to use in-line filters. These work through only displaying values within a report that match the defined conditions of the filter. They are a great way of quickly applying a segment like Brand to your Search Terms report or Blog Posts to your Landing Pages report.
This post has covered the third key technique in analysing web analytics data, segmentation. This can be performed by refining the set of data being reviewed or by isolating a subset of data, whichever is appropriate to answer the business question under investigation. Without going into detail, I mentioned the potential need to group variables referring to other blog posts that go into more detail on this topic. As mentioned, Avinash's books and blog are a great source for learning about segmentation while if you use Google Analytics, Justin Cutroni has written about the six ways you can do segmentation in GA.
The three techniques described in this series of blog posts (which was not meant to cover a 3 month period) are the basic tools for any web analyst. There is a lot more to it of course, this only covers clickstream data. Optimising the performance of an online business will require familiarity with areas such as usability studies, VoC or A/B testing. But these are some of the building blocks for any web analyst new to the game and are vital if in answering that common question - how was the business performance last week?
Links to other posts in the series