I never quite finished my series on building profiles of visitor segments. Where I left off, I’d put together a pretty nice little profile of the sourcing for a single visitor segment that I was examining (from a Financial Services site) using Unica’s NetInsight and Excel:
In building this profile, the key concepts were the creation of a control group and the use of indexicals comparing the target segment to the contro to represent where the target segment was distinct.
In this post, I’ll use some of the same concepts while I explore the more complex issue of profiling content interests.
For visitor segments, content interests are the single most important defining characteristic. In this case, my target segment was actually defined as those who showed interest in a particular product. The fact that the segment includes a content selection complicates the task of building a profile since we have to make sure the differences we see between our target and control group aren’t simply the result of viewing our targeted area (and the navigation necessary to get to it).
Another complicating factor is the way NetInsight works for segmentation. If I simply build a segment of visits where my target pages were viewed, the only page views contained in the segment will be my target pages. That isn’t what I want at all! So for this analysis, I had to build a Visitor Profile in NetInsight and then apply the Profile. Profiles allow you to select all of the Visitor and Visit behavior that applies to some filter condition – in this case viewing my target pages.
For our target population, a large percentage of our traffic came directly to the product area and then left. That won’t make for a very interesting content profile. So I started by selecting only Home Page Entries for this group and then evaluating the content viewed (minus the target pages) compared to the content viewed by the control group when entering on the Home Page.
Here’s what I got, shown in a simple columnar chart:
While I see some fairly interesting differences here, I don’t think this the best view I can get. The problem is that while I used the Home Page, many of my target group visitors viewed the Home Page and went directly to my Target Content via the Target Category page – viewing nothing else along the way. That distorts the comparison between my target and control groups.
So I further refined my segmentation by picking only visits that included the home page plus some other type of page. Here’s how the bar chart looks after this step:
The impact to this wasn’t as profound as I expected but it did bring most of the numbers into slightly closer alignment.
I think this is probably a pretty adequate basis for comparison, so now I’m going to tune the presentation a bit and incorporate this content interest into my profile. Here's a first cut:
This is very consistent with the way I chose to display sourcing. But I want to explore a few additional options that are often particularly useful for profiling content & tool usage by segment.
One method that really highlights differences vs. the control group is to use the control group as the base and show the target group’s behavior as a plus/minus % difference compared to the expected control value:
To my mind, this makes the difference to control very easy to understand and gives a more impactful view of how significant the differences are.
Here’s another alternative I borrowed from an EU-based web analytics package (Nedstat) that makes very nice use of Radar-charts to show content interest:
These 3 views are all derived from the same basic data. The first shows both target and control with unadjusted data. The second shows only the target values. The third is the Radar counterpart to my plus-minus bar above and shows the target values vs. control.
I’ve always thought Radar charts looked cool but were a little hard for people to grasp. And I wouldn’t favor the third alternative here because it doesn’t give a good sense of the extent of the negative values – the places where our target users do less than expected. The second chart would probably be better if I made it indexical instead of unadjusted. The first chart is my favorite of these. Would I use this? Probably not unless I was showing a series of segments and wanted an easily comparable visual benchmark. For a single profile, I’d probably stick with the plus-minus bar view.
Here’s a combined view:
This profile still lacks a sense of how large this group is and how successful the visitors in this group are. I’ll fill in both next time around!
In building this profile, the key concepts were the creation of a control group and the use of indexicals comparing the target segment to the contro to represent where the target segment was distinct.
In this post, I’ll use some of the same concepts while I explore the more complex issue of profiling content interests.
For visitor segments, content interests are the single most important defining characteristic. In this case, my target segment was actually defined as those who showed interest in a particular product. The fact that the segment includes a content selection complicates the task of building a profile since we have to make sure the differences we see between our target and control group aren’t simply the result of viewing our targeted area (and the navigation necessary to get to it).
Another complicating factor is the way NetInsight works for segmentation. If I simply build a segment of visits where my target pages were viewed, the only page views contained in the segment will be my target pages. That isn’t what I want at all! So for this analysis, I had to build a Visitor Profile in NetInsight and then apply the Profile. Profiles allow you to select all of the Visitor and Visit behavior that applies to some filter condition – in this case viewing my target pages.
For our target population, a large percentage of our traffic came directly to the product area and then left. That won’t make for a very interesting content profile. So I started by selecting only Home Page Entries for this group and then evaluating the content viewed (minus the target pages) compared to the content viewed by the control group when entering on the Home Page.
Here’s what I got, shown in a simple columnar chart:
While I see some fairly interesting differences here, I don’t think this the best view I can get. The problem is that while I used the Home Page, many of my target group visitors viewed the Home Page and went directly to my Target Content via the Target Category page – viewing nothing else along the way. That distorts the comparison between my target and control groups.
So I further refined my segmentation by picking only visits that included the home page plus some other type of page. Here’s how the bar chart looks after this step:
I suspect this view represents a closer approximation but now I feel like I may have gone too far. My definition for this group within the Target population automatically forces every visitor to have looked at 2 different things (my target group plus something else). I don’t have a corresponding condition on the Control Group – so it stands to reason that my Target group may be a little bit biased. Remember, the single most important task when building a good visitor profile is choosing the best possible control group to highlight what's truly interesting and distinct about the target - not what's different because of the exigencies of site navigation.
So in my next cut, I controlled for that by segmenting within the Control Group and selecting only visitors with 2+ areas. Here’s the new comparison after this segmentation:The impact to this wasn’t as profound as I expected but it did bring most of the numbers into slightly closer alignment.
I think this is probably a pretty adequate basis for comparison, so now I’m going to tune the presentation a bit and incorporate this content interest into my profile. Here's a first cut:
This is very consistent with the way I chose to display sourcing. But I want to explore a few additional options that are often particularly useful for profiling content & tool usage by segment.
One method that really highlights differences vs. the control group is to use the control group as the base and show the target group’s behavior as a plus/minus % difference compared to the expected control value:
To my mind, this makes the difference to control very easy to understand and gives a more impactful view of how significant the differences are.
Here’s another alternative I borrowed from an EU-based web analytics package (Nedstat) that makes very nice use of Radar-charts to show content interest:
These 3 views are all derived from the same basic data. The first shows both target and control with unadjusted data. The second shows only the target values. The third is the Radar counterpart to my plus-minus bar above and shows the target values vs. control.
I’ve always thought Radar charts looked cool but were a little hard for people to grasp. And I wouldn’t favor the third alternative here because it doesn’t give a good sense of the extent of the negative values – the places where our target users do less than expected. The second chart would probably be better if I made it indexical instead of unadjusted. The first chart is my favorite of these. Would I use this? Probably not unless I was showing a series of segments and wanted an easily comparable visual benchmark. For a single profile, I’d probably stick with the plus-minus bar view.
Here’s a combined view:
This profile still lacks a sense of how large this group is and how successful the visitors in this group are. I’ll fill in both next time around!
Comments