[This post is the second post of the Web analytics in practice series - practical posts on various topics based on my own daily experience – as a practitioner. It aims at providing tips, advices and examples that – I hope – may inspire and help you – whether you are a beginner or more experienced Web analyst]
If you really want to do true analytics then segmentation is essential. I like to think that if you are just looking at aggregated data, you are only doing reporting. If you want to do analysis, segmentation is the way to do as it leads to valuable insights that, in turn, will drive business actions.
While there are several “common” ways in segmenting online data – true segmentation requires putting in the effort to have a good understanding of your business (what does matter, key goals…) and to find your own meaningful segments. Such exercise will help you sharpen your business expertise – always a good thing.
In this post, I propose a step-by-step simple example (based on my own experience) to illustrate how to apply “standard” and context-related segments, to drive insights and the resulting actions.
The case: analysing the performance of a key landing page
(Disclaimer: for confidentiality purpose, actual figures and results have been modified but I have kept the general order of magnitude and the resulting learnings are true ones)
The mission is to analyse the performances of a landing page that is the entry point for new customer acquisition. The page depicts the product offering and key benefits. The main goal (or outcome if you prefer) of the page is to drive traffic to the acquisition process. (Note: for this illustrative example and to keep it simple, I will limit the scope to this simple “micro” goal).
The metrics used: traffic volume to the page, number of conversions (outcomes) and of course the conversion rate.
Start: no segmentation – aggregated level
A quick query in the Web analytics tools returns raw numbers that look like this:
These are just…er, numbers. What do they tell? Nada, nothing, niets, rien! The number may be nice for some or ridiculous for others. Who knows? (Even me, I couldn’t tell if these were good or bad as it was the first time I got my hand on this area).
1st level segmentation – by traffic source type
If you are a regular reader of Avinash Kaushik’s blog (or you have read his book – “Web Analytics 2.0“), you may have heard about what he calls the “best Web analytics report” (page 85 in the book ). A good way to start is by segmenting by traffic source. It is quite an “universal” way to segment.
I start first with key traffic source types: “direct” (in this case coming from site), search (all) and paid channels (banners, affiliates, emailing but not SEA). This already gives a different picture:
At this stage, there should be some alarms ringing in your head. Look at the paid channels! It performs 8 times less than the other ones while it brought almost 90% of the traffic but account for less than 50% of conversions. Ok, we know that typically such channels so not perform so well but still! Like the detective we are, we got a first clue so let’s further investigate!
2nd level segmentation – by paid channel
Focusing on the paid channels, we can segment it one level deeper – by paid channel and I trend the result over time to add more context to the analysis. (Again, in order to keep the example simple, I have limited it to 3 paid channels). Now we are seeing something very interesting:
Affiliates traffic suddenly went through the roof – causing at the same time a massive drop of the conversion rate. Almost time to talk to the digital campaign manager but before let’s gather some more insights and let’s segment a level deeper: by actual sources.
3rd level segmentation – by referring source
I look specifically at the affiliate channel and split it by referring site to end with a report very similar to the Avinash’s “best report in the world”:
The results clearly highlight big discrepancies between referring sites – those related to saving & financial topics are performing rather well while the rest is doing very poorly in spite of bring huge amount of traffic (typical of affiliates).
Now I have enough and I can go to talk with the digital campaign manager (and not just sending an email with figures) in order to share the insights. If cost is related to traffic then urgent action is needed (= wasted money). If not related to volume but per conversion then the question is: is it really useful? What is the lead quality? Is it positive for the brand (for visibility and awareness)?
At this point it is up to the manager to decide. My role is not to tell her how she should do her job – I would not dare to – but to provide her with insights so she can take data-informed decisions.
Using contextual segment – customers vs. Prospect
Similar segmentation can be applied to search channel – organic vs. paid then at keyword level but I guess you get the picture. Now let’s apply a different segmentation – what I call contextual segmentation i.e. that relates to your own business context.
Working in the bank industry, there are two main types of visitors to the website: customers and prospects. I can easily identify the former group using a visitor-based segment: visitors who have logged in the online banking area in current or former visits.
Why such segment? Because there is no way customers can be “converted” so by aggregating these visitors in my results, I don’t get a true representation of the actual performance of the content and process for the intended audience i.e. prospects.
The analysis shows that customers represent a significant share of
visitors coming from search – these land on the page and they use the
site navigation to log in or to browse the content. Now calculating the
conversion rate for the prospects gives a much different results as
See the difference between the “average” performance and this segment? Amazing, isn’t it? The insight here is that prospects coming via search are highly qualified. What can I do? Get more of these segment to the site. How? By reviewing SEO aspects of the page (meta tags, copytext…) and optimize it (and there was work to be done in that area ) and considering putting some more money on SEA.
The analysis also demonstrates that the page is maybe not doing so bad after all. Just looking at aggregates could have lead to a typical (bad) reaction: “let’s redesign the page – it sucks!”.
Segmentation - a powerful analysis tool
Segmentation can be applied on many criteria: sources, behaviour and others. Be inventive! Segmentation is a way to translate a business question into an analysis that will help bring the answer. Typically, segmentation will lead to new questions that in turn, will be translated in more detailed segmentation.
I hope that I have illustrated how segmentation moves you from reporting to analysis. I must confess that I fell in love of segmentation. Once you start, you can’t stop. This stresses the importance of having tools that give you such ability, tools that make it possible to do in a flexible and dynamic way. Segmentation on the fly is not an option. It’s a must!
Taking segmentation to the next level
This post was a just an illustrative example and there is much more behind segmentation. If you have the possibility of linking your online data with CRM or customer data then you can do even more advanced segmentation: by customer segment, by product, by demographic, by persona… It will pave the way to more valuable insights and customer intelligence. Segmentation climax will not be far away! More on that in a future post…
So, was the example a good illustration of segmentation? What are your favourite segmentation criteria? What “own” segments are you using? Do not hesitate to share your feedback and own experience on that topic. I am very curious to know.
Related resources & posts:
- “Web Analytics in practice (#1): Campaign tracking & offline advertising” (Sept 2011)
- “Web Analytics Segmentation: Do Or Die, There Is No Try!” by Avinash Kaushik (May 2010)