The concept of segmentation, is of course, well-known to marketers but it is a term increasingly applied within web analytics. While analytics specialists may be up-to-speed on online segmentation options, most marketers won't have time to get-to-grips with the options.
So, I'm pleased to be present this interview with analytics specialist Hugh Gage to enlighten us about the online segmentation options.
In this interview, I talk to web analytics consultant Hugh Gage of Engage Digital who was previously a senior analyst at Logan Tod & Co and before that Head of Online Planning and Buying at Manning Gottlieb OMD. Hugh also contributes to the Web Pro Analytics page in .NET magazine.
What is segmentation through web analytics?
Q1. Please explain what segmentation means when using web analytics systems.
Segmentation in data-centric web analytics is essentially the capability to analyse a site"€™s performance against a given set of criteria which can be isolated using filters available in your web analytics tool.
For example: an ecommerce site might have an average conversion rate of 1.5%, but this tells us very little about where the strengths and weaknesses lie.
By using the data segmentation features in your web analytics tool it is possible to partition off specific groups of visitors and compare relative performance against the site average using the same key performance indicators.
All the leading web analytics tools on the market place today (HBX, Visual Science, WebTrends, ClickTracks, Google Anlaytics, Gatineau, Omniture, Coremetrics) provide some level of data segmentation within their product line-up, however the level of sophistication and flexibility varies significantly from product to product.
How can site visitors be segmented?
Q2. What are some of the most common attributes of visitors to segment by? Could you give some examples of the benefits of these?
What you want to segment by and what you can segment by depends on the analytics tool you have. That said, typically one of the most common attributes used in creating a segment will be referring source of the visitor.
Effectively a web site can be seen as one entity and the visitors that come to it can be seen as another entity.
In order for performance to be good the web site, must among other things, meet visitor expectations on arrival. But visitor expectations will vary depending on the referring source.
A very simple example would be comparing pay per click search with banner advertising. Visitors from pay per click search advertising have already expressed a level of interest by virtue of typing a relevant search term into a search engine and therefore the level of expectation is likely to be more refined than a visit coming from a banner ad which is practically a cold call.
In the context of an e-commerce site, separating out these two referring sources of traffic using segmentation will reveal both the comparative performance based on overall conversion rates and additionally it will help identify other technical and behavioural characteristics - such as landing page bounce rate* - which may present some actionable insight.
Taken to the next level, the same could be done for groups of referring key words and phrases. Visitors entering the site from a brand search term will have different characteristics to those entering from product specific search terms.
Segmentation attributes don"€™t have to be limited to referring source. Visit length, number of pages viewed per visit, entered on a certain page, saw a certain page (e.g. internal search results page), internet connection speed "€“ these are all potential attributes against which a segment may be created depending on the questions being asked by the analyst and the capability of the tool.
- Bounce rate refers to the percentage of visits that see only the page entered on before leaving again (or the percentage of visits lasting no longer than 10-15 seconds) "€“ it is effectively a retention measure that reflects quality of referring traffic and quality of the landing page. Comparing Bounce rates within different segments can be particularly revealing.
How can you combine different segmentations?
Q3. How readily can you combine different segmentations?
This is an excellent question. Again, the answer must start with "€œit depends on the analytics tool"€. Naturally, the ability to cross reference segments and filtering attributes lends greater weight to any analysis and provides a deeper level of actionable insight.
Google Analytics makes it harder to combine segments since the analyst must chose from the list of pre-defined segment options and then navigate through the features menu in order to get the desired combination. ClickTracks uses a labelling system to identify segments, the advantage here being that any two labels can be cross referenced to create a new label that combines the attributes of the originating two.
This process can go on until a "€œsuper segment"€ has been created with multiple attributes effectively allowing the analyst to create a persona. Visual Science operates in a similar but more sophisticated way.
Other analytics tools allow the analyst to assign multiple attributes to each segment in the initial set up stage.
It"€™s also worth noting that segmentation in web analytics is a very powerful feature which requires quite a bit of computing power. Customers using a tag based version of an analytics tool where data is hosted by the vendor, may find there is a limit to the number of available segments in the initial customer package.
Consequently the ability to delete and re-create is also restricted in some instances. This means that in any given period it may not possible to continue to create multiple segments as answers give rise to new questions unless more segments are acquired "€“ usually at a cost. This again varies from vendor to vendor.
Q4. Are there some segments that are typically more powerful or easier to create / apply than others? Perhaps those that can be used to identify quick wins?
The ability to produce a quick win depends on the ability to take action and implement change.
When it comes to segmenting data in web analytics, quite often marketers will focus their attention on paid for advertising because this will help them get a better ROI and also because it can be easier to change the marketing mix than to change the site itself; this is especially likely to be the case in larger organisations where site changes are sometimes limited to scheduled update periods. In setting up these segments it is also necessary to ensure that campaigns are being tracked by the analytics tool, this usually throws up its own set of questions but there is not time to go into those here.
When looking at segments I think it is also important to consider weighting. This is helpful in creating context and can prevent budget being needlessly misappropriated. For example: visits from an email list may convert at 4% whilst visits from pay per click advertising may convert at 3%. Over a 1 month period that email may account for only 3% of total site traffic whilst the pay per click activity may account for 15% of total site traffic.
In this instance even though the email is performing better it may make more sense to concentrate efforts on refining and improving the conversion rate of the pay per click activity on the grounds that there is a greater volume of traffic and even a small increment in conversion may yield better overall sales volume than the email.
This is a highly simplistic example of course and there are other factors that may come into play but the idea of weighting segments taken in context with the conversion rate and the ease and cost involved in implementing changes should make for an easier decision regarding which channel to pursue.
The future of online segmentation
- Q5. How do you see the applications of web analytics segmentations being extended in the future?*
I think there are two key areas segmentation will extend into, and in fact are already doing so; those of sequential segmentation and segmentation based on external sources of data.
By sequential segmentation I mean the ability to filter based on multiple attributes whilst simultaneously being able to stipulate which event occurred first or last.
Say a campaign consists of email, PPC and banner advertising. A customer may finally purchase after seeing all 3 elements, 1st the email, 2nd a banner and 3rd a PPC search ad. They also visit the site 3 times, once from each communication channel however, they may only buy after visiting the site from the PPC ad.
Traditionally the sale is awarded to the PPC ad, however this is unfair since it may be that the heightened awareness generated by the other two campaign elements which were seen first played a significant part in the persuasion process. Being able to play with segments that include various combinations of the 3 campaign channels whilst also being able to stipulate that only visits from the 3 campaign channels that entered the site prior to a purchase should be included in the segment will help identify the most efficient combination and therefore help improve ROI on media spend.
The second element of segmentation based on external sources of data refers to the growing ability of today"€™s web analytics tools to be hooked up with other client-side databases to provide additional levels of granularity. For example, on a site where registration is required it may be possible to link the analytics tool to the registered users database and utilise information from there to segment by attributes such as age, gender, actual location (not just the location a visitor enters the internet through their ISP), purchasing habits etc.
Thanks Hugh - some great insights there!
You can read more from Hugh in this blog posting on making more of segmentation