April 20, 2011 ↘︎

Moving beyond business-based segmentation

Loading the Elevenlabs Text to Speech AudioNative Player...

FOYL_DashboardOne of the most powerful ways to enable an audience connection is through behavioural segmentation.

Many companies today segment from a business standpoint.  Don’t get me wrong, this is a good strategy and aligns your measurement and optimisation strategy with your business segmentation model.

Customer / non-customer segments.  Product A owners / product B owners.  Mosaic-based segments.  Geographic segments.  Lead / Non-lead segments.  These are all typically business-based segments, and you should definitely be segmenting using this methodology if your overall business does.

But I think there’s a higher level of segmentation – behavioural segmentation.

Behavioural segmentation in action

Every time someone visits your site they’re telling you what they’re interested in.  These behavioural segments are measurable and usable.  And moving from business-based segmentation to behavioural segmentation truly allows you to relevantly engage with your customers based on what they’ve told you about themselves, through either their activity on your site, or through forms they’ve completed.

At Murdoch, we recently re-developed our primary lead generation tool, called “Figure Out Your Life”.

FOYL, as it was affectionately known to us, was the University’s primary lead gen tool – it essentially asks you a number of questions and then suggests courses or careers that may be of interest to you.

Overall, the redesign was based on user feedback from focus groups but we took the opportunity to introduce behavioural segmentation into it, to not only measure those segments over time, using a combination of SiteCatalyst and SAINT classifications, but to enable more refined and relevant content targeting to each segment for optimisation and engagement purposes.

The new FOYL, now called Figure Out Your Course (or FOYC) asks 6 personality-type questions, and allows the user to rate their answer to each question, on a sliding scale, from 1 to 10.

The combination of their answers to the 6 questions results in 3 recommendations for courses they might consider.

I’m not going to reveal the methodology behind tool – that’s confidential IP and very complex, but I will talk about the segmentation opportunities it allows.

How we achieved behavioural segmentation

From a segmentation standpoint, we decided that we wanted to firstly measure the segments to see what “types” of users are completing the tool, and secondly, through integration with Test & Target, we can then customise content to each user, based on how they answered each question.

To do so, we needed to capture each of their answers in an eVar (conversion variable), which must be remembered over multiple visits.

To enable more realistic clustering of segments, we decided that while the tool works on a rating between 1 and 10, we’d actually work our segments on a rating of 1 to 5.  So we “bucketed” each answer accordingly:

Moving the slider to position 1 or 2 makes them segment 1,
3 or 4, segment 2
5 or 6, segment 3
7 or 8, segment 4
9 or 10, segment 5

In doing so, each user now scores (from a segment standpoint) between 111111 and 555555, which makes a total of 15,625 combinations (111111, 111112, 111113, etc through to 555555), or 5^6.

By combining their scores for individual answers into a single variable, we end up with a per-person score, where each digit in the variable represents their chosen response to each question.

So for example, if a person answers question 1 with a rating of 2, question 2 a rating of 10, question 3 a rating of 7, question 4 a rating of 10, question 5 a rating of 6 and question 6 a rating of 1, their overall segment score would be 154531.

We store this in an eVar and a persistent s.prop, and pass them to SiteCatalyst.

Results

What we end up with is a list of scores in a SiteCatalyst report as follows:

FOYL_segments_raw

Not a lot of use, except that we can see the raw scores and the number of instances of each score.  You can see as well that they’ve started to cluster.  50 people have been through the tool and selected 141515 as their answers to each question.

Using SAINT classifications, we then created every possible combination and classified every result.  Despite it being 15,625 combinations, this only took a few minutes – gotta love Excel for this.

We’ve classified each digit location as the question, and each digit value as the score from 1 to 5.

For example, the classification for digit location 1 is “Indoors (1) or Outdoors (5)”.  The values range from 1 to 5.

Our classification file looks like:

FOYL_Classification_File

We set “No FOYL” as the default for people who haven’t been through the tool yet – which also allows us to use T&T to engage them into the tool.

Using the classifications, we can now see a report that shows all responses to digit location 1…i.e. how did people answer that question:

FOYL_Question_1_Responses

So we can see that most people are answering question 1 with a segment rating of 3 (mid way).  Unfortunately, we can’t order the values numerically, they are ordered highest to lowest on their instances.

If we chart that as a pie chart, we see:

FOYL_Question_1_Pie

Now we repeat that for every question asked and pop it into a dashboard for an overview of results, and we can see:

FOYL_Dashboard

We can also now sub-relate those individual segments with the courses they’ve viewed:

FOYL_Question_1_Courses

Our true measure by segment is lead generation, so we just add in those conversion metrics too, Lead Start and Lead Complete and we can see conversion by segment.

Discover

Using Discover, we can create those segments individually and compare them.  Here I’ve compared search terms against anyone that answered the Behind Scenes/Lead a team or Indoors/Outdoors with a 5:

FOYL_Discover

The same segment can be used in SiteCatalyst 15 to segment your reports.

We also pass these segments into our email marketing platform so we can segment our ongoing communications further, generating a more personalised and relevant experience for the user.  We can now customise parts of the message dynamically, based on the segment (or segments) that they told us they’re in.

Test & Target

Now the really useful stuff.

Because Test & Target can also see those segments and values, we can target content to each segment as they re-engage across our content.  For example, we can vary messaging to people who prefer to work outdoors instead of indoors.

We can target course content to them based on their selections.

We can build multiple targets and combination targets to get the content to them.  If they haven’t been through the tool, we’d also know that too, and we can target messaging to try to engage them that way.

In summary

Every company has the opportunity to segment. And that’s how you move beyond reporting to insight.  Understanding how your segments interact with your content or how they convert differently allows you to further optimise their experience.

There’s plenty of opportunity to segment for measurement purposes, but when you combine Test & Target with those segments, and start to target information based on those segments, it’s really powerful.

When you move away from just reporting, to segmentation, and beyond to segmentation based on user behaviour, you can really start to generate user engagement and relevance of content.

You just need to open your mind to the segmentation possibilities and find out what works for you.

Let me know how you’re segmenting your site visitors and what you’re doing around optimising their user journey based on their segments.

DB logo
DB logo
DB logo