A Sample Two-Tiered Segmentation for Media Properties

In the last couple posts in this series on the “Convergence of Digital Analytics and Database Marketing” I’ve been describing the details of Semphonic’s Two-Tiered Segmentation scheme and giving examples; first from Financial Services and then from Travel & Hospitality.

A two-tiered segmentation starts with a traditional audience segmentation (Visitor Type) and then sub-divides it with a second layer based on visit intent (Visit Type). In those posts, I showed several sample Two-Tiered Segmentations and explained how Two-Tiered Segmentations should be at the heart of almost every Web analytics activity.

For some industries, the creation of the first tier (Visitor Type) is conceptually straightforward. In verticals like Financial Services and Travel, customer segmentation is deeply ingrained in the business model and marketing genes. Because these industries often have logged in sites, they have a ready-made method of mapping these Visitor Types to visitors on the Web. Of course, the second tier (Visit Type) still needs to be created in these verticals, but this, too, is often easily mapped to basic site function.

I’m not saying that creating a Two-Tiered Segmentation in these verticals is easy, but it does flow naturally out of the traditional business and Website design.

That isn’t true for every vertical. Media properties, in particular, can struggle to create a Two-Tiered Segmentation. Media properties don’t often benefit from enforced registration and they don’t have a long-time culture of customer data warehousing and customer segmentation. So how practical is a Two-Tiered Segmentation in Media?

It is possible; and the effort to create a two-tiered segmentation may be particularly worthwhile for Media properties. It can be a foundational part of shifting the digital media culture away from a focus on inventory to a focus on audience.

In Financial Services, I showed a Visitor-Type segmentation based on the role/relationship to the business. The segmentation broke out Advisors, Financial Planners, and Investors; it then subdivided each of these by value of relationship. In Hospitality, I started with the basic relationship (Customer / Prospect) and then sub-divided this by booking channel and finally by Club Membership and relationship value.

Does Media have an equivalent type of segmentation based on role/relationship? To some extent, it does. Most Media sites could start with something like this:

 
Increasingly, this type of Segmentation might be supplemented by your relationship via delivery channel:

 

This channel-based relationship view becomes particularly interesting when the creation of channel relationships becomes a core digital goal.

Tne drawback to all of these sub-divisions, however, is that they are focused on the Registered population. In some cases, that’s a tiny percentage of traffic and traffic value. If you’re only segmenting 2-3% of your population, you haven’t really created a Visitor Segmentation.

For some Media Properties, there’s a fundamental Visitor Segmentation based on Geography. Any property with a split between a Local, National, and International audience should surely capture that in their segmentation.

 
A more universal strategy for segmentation of anonymous visitors is to focus on their behavioral site relationship:

Here, I chose Primary Finding Method as a primary segmentation level. This isn’t my favorite type of Visitor Segmentation since it overlaps heavily with Visit Type. But it does capture real elements of a Visitor Type segmentation. For Media Properties, the “Homepage” setter has a different relationship with the site than the regular (daily) entrant from a bookmark and they are both very different from the Occasional SEO Visitor. Obviously, for my third tier of Visitor-Type Segmentation I used a categorization based on Frequency and Recency. Combining Frequency and Recency provides a nice behavioral Segmentation not too different than the Relationship Value I used as a sub-segmentation for both Financial Services and Hospitality.

Another approach (and one I generally favor though it is more complicated) is to develop your Visitor Types using a Behavioral Segmentation. This is a full-on cluster analysis that combines all sorts of variables to create distinct site personas that then become the basis for the Visitor Type Segmentation tier. The types of variables used with this technique can encompass everything from usage, to topicality, to content-type preference (all the things I talked about in my post on meta-data and implementations):

• Recency and Frequency of Visits
• Time and Day of Week of Usage
• Type of Content Consumed by Mindshare (Text/Slide/Video/Audio)
• Rich Meta-Data about Content Consumed (Topic/Tone/Source/Recency/Navigation Path)
• Arrival Methods
• Duration of Relationship
• Velocity of Relationship
• Consistency of Visit Type
• Community / Social Propensity

A behavioral segmentation can live under (or encompass) more rigid segmentations like “Local/National” or “Registered/Anonymous” that I’ve already described. Combining these elements might produce a 1st Tier Visitor Segmentation like this:
Behavioral Cluster Segmentation will inevitably blend elements of Visit types with Visitor Types. The "Daily Horoscope Checker" is a Visitor whose primary Visit Type is a quick "Horoscope check". Given that we want to create a matrixed segmentation, this will create a certain amount of co-linearity (internal dependence) between the Visitor and Visit Type segmentations. That’s not all bad, but it’s not all good either.

Including some Visit Type data in the Visitor-Type Segmentation often makes “magic moments” – times when a digital visitor steps out of their daily patterns and does something new and interesting particularly apparent. When a "Daily Horoscope Checker" shows up with a “Top Story” Visit Type, you know something special is up. On the other hand, it can reduce the value of the matrix down to a single dimension. If all you’ve captured in both dimensions is that a Visitor is a "Horoscope Checker," you may have lost something interesting about their sourcing, frequency, or duration of relationship.

As with most segmentation questions, there is no one right answer. A Visitor-Type Segmentation can consciously choose to exclude the data and variables you are going to use to describe Visit Type or it can encompass them. It can even encompass but underweight them. The most appropriate direction is dependent on the site and business focus and, ultimately, the intuition of the analyst.

So what about Visit Types - our 2nd Tier of the Segmentation? The 2nd Tier is not quite so tricky for Media properties. Visit Types will typically track to topic with some layering of daytime parting and source. Topic I assume to be self-explanatory, but why daytime parting and source?

For some media properties, daytime parting provides significant clues into the nature of the visit. Weekend visits may be significantly different than Weekday visits. Early morning visits are often different than lunchtime visits or 10PM visits. Daytime parting, in conjunction with broad topic classifications, can help drive significant visit segmentation. Having said that, I want to emphasize that our experience of daytime segmentation is that it is NOT universal. For a fair number of Media sites, visitor behavior is not as differentiated by daytime parting as might be expected.

Source, on the other hand, is nearly always significant. The most common and significant source differentiation is by SEO vs. Direct but others may also be interesting. SEO visits are typically distinct from Direct Visits even when they consume similar content. Not only is the shape of the visit different, but the success of the visit is measured differently.

I want to highlight this point.

When building a segmentation, it’s important to think about whether there are types of visitors within a matrixed cell (an intersection of Visitor and Visit Type) that you know perform differently. If there are, then you’ll want to create an additional layer of the segmentation to split the cell. For every media site we’ve ever measured, SEO sourced visitors perform at a significantly different level from Direct-sourced visitors. That being the case, at some level, your segmentation scheme should break them apart or else you risk muddying your success numbers.

I’ve said over and over that metrics and KPIs are dependent on and should only be viewed within the context of a Two-Tiered Segmentation. But here’s an interesting example of the converse; sometimes the Segmentation Matrix needs to be dependent on your knowledge of what constitutes success. If you’d be happy with an SEO Visitor to the Local Politics section viewing 1.5 content pages and you expect a Direct Sourced visitor to view 3.5 content pages, then for heaven’s sake you need to make sure they are in a separate Segmentation cell!

Here’s a partial sample of what a Media Visit-Type Segmentation might look like:

Media Segmentations, in particular, tend to be large because the content on the site is often incredibly diverse. This may lead you to create an Uber-Segment approach where topic is extracted from the base-segmentation and then added in as an additional layer:

This approach gives you considerable flexibility in reporting and analysis as you can choose the most appropriate level for any given problem or audience. Because media properties are often organized along topical lines, you might also consider switching the levels, so that the topic level lives at the top and is then sliced by sourcing and generic visit type descriptions (such as Daily Check or Researcher).

Combining this type of Visit Segmentation with your preferred method of Visitor Segmentation and you’ve got a complete Two-Tiered Media Segmentation scheme.

Another long(!) post, but I hope an illuminating one. Media sites have been reluctant users of segmentation and often with good reason. Building a Two-Tiered Segmentation for media is less obvious and more challenging than in many other industries. However, as I said at the beginning, the rewards of creating a full Two-Tiered Segmentation in Media can be considerable.

I often talk about Two-Tiered Segmentation this way. When someone walks into your office and says “Traffic is up by 5%,” your first question should always be “With whom?” and your second question should be “And what are they trying to accomplish?”

There is a world of difference between knowing that your Infrequent SEO-based National Traffic for Major Stories is up versus knowing that your traffic has increased. Media sites, more than any others, should understand that to tell any story well, you must always lead with the “Who” and the “What”. A Two-Tiered Segmentation is the best way to make sure your reporting, your analysis, and your database marketing answer those two questions in every single case.

 

Comments

[Chicoli]

Thank you for this informative post.
Could you tell us what Omniture add-ons you have used to help define recency/frequency segments? We're a little stumped trying to do this with just visit number and days since last visit.

[Gary Angel]

It's true, the default Omniture Plug-ins don't provide true RF variables and it can be nearly impossible to construct them in Segmentation engines.

The best alternative is probably to write your own. I described the basic techniques in a previous post on Paywall Analytics. In that same post, I discuss the use of eVar counters. They won't quite do the same job, but they are much easier and used carefully can create something like good Frequency numbers.

For Recency, I think a custom plug-in is really the only good solution.