Perhaps because it’s such a common task in web analytics, it doesn’t get much attention. It’s more like delivering a baby than brain surgery. And in our field, everyone wants to be a brain surgeon. But if you don’t deliver babies, there won’t be any brains to work on. And there is more to this seemingly mundane task than meets the eye.
Compared to What?
If the key to a good visitor profile is to show what’s different and interesting about a specific set of visitors, the obvious question is: “Different compared to…?”
The obvious answer – the site average – is not often the right answer.
We all know that averages can hide what’s interesting and site averages are particularly prone to this problem. Many sites exhibit a bipolar distribution around most key behaviors: there is a group of non-regular visitors who skew below all of the site averages and there is a group of highly-engaged visitors who skew far above all of the site averages.
Each of these groups may exhibit something like a normal distribution when considered on their own, but when viewed as a total population they would produce a distribution that looks more like this:
Not every site is like this. If your site is “logged-in”, your visitors are more likely to follow a normal distribution. But this is far from certain – there are as many different visitor distributions as there are sites. You should start with a fairly good idea of the shape of your distribution and the key populations on your site. And you’ll typically compare your target population to one of these key populations.Selecting Metrics
Assuming you have your two populations (the target and control) what metrics do you use to build a profile?
It’s not always going to be the same metrics of course – and for many businesses it’s the dimensions that are specific to your customers and site that will matter most. But there are some common general behaviors that will certainly belong in any profile along with some techniques for showing how profiles compare.
For acquisition sites, key profile metrics will almost certainly include source, bounce rate, total behavior (views/visits), success events and content interest. Other variables like geography and device may also be interesting. For a logged-in site, source and bounce rate are much less likely to be interesting. You’ll probably be focused on total behavior, content interest, tool-usage distributions and success. And, of course, for logged-in visitors you’re much more likely to have deep offline or registration information.
Using NetInsight to Segment and then Profile
Let’s look at some different techniques for building a profile around one of these variables.In this case, my target group was a population who looked at particular type of investment product. Nearly all of the visitors who did this happened to fit into one of the company's pre-defined investor types (institutional/individual/advisor/etc.). So I used that investor type as my control group.
In NetInsight, you can create filters from any dimension or metric. You start by where in the interface you want them to go:
You pick your underlying dimension or metric:
Then you define your filter:
Once you’ve created them and added them to the interface, you can slice any report (standard or custom) by them.
For this analysis, I created two filters. One was for the broader investor type. The second was for the broader investor type AND users of the target tool (I purposely excluded the small percentage of tool users who weren’t in the broader category type to make sure they didn’t muddy the numbers).
Profiling a Single Metric or Dimension
For acquisition source, I started at the summary level:
It’s evident at a glance that our target population on the left is much more likely to come from Yahoo than from Google. But we can get a better sense of this by exporting the numbers into Excel and calculating each source’s simple share (simple to do in the tool as well of course):
There are any number of ways to present this type of information. Here’s a classic bar chart:
Two pie charts might look good, but there are too many values in the table to present well as is:
Instead, you’d probably collapse the rows to do something like this:
Collapsing columns (MorningStar, etc. into Financial) like this is essential for pie charts – but it would probably crisp up the bar chart as well. Chart types that support too much cardinality are often a bad choice because they encourage you to be lazy about refining the message.
Stacked bars do a nice of presenting the data in a more directly comparative fashion:
But with this style you’d need another visualization or amount showing how much of the total each type represents.
This looks pretty nice but in my opinion is actually a little harder to absorb. On the other hand, it does put the emphasis on the comparative aspect of Target vs. Control – and that’s usually where I want it for a profile.
If I wanted to balance the comparative view with the absolute importance to the Target group, I might be inclined to separate out this presentation into two discrete elements – the first capturing what’s important to the target group in a single pie chart and the second how this compares to the control group in the set of 100% stacked bars.
There is, of course, no one right way to present data. But the presentation you choose should capture the salient elements and emphasize the points you want to make.
These differences in referring site may well be driven by additional factors (like campaign) and may, in turn, influence additional factors (like Entry Page and Content Interest). Perhaps there is a Yahoo campaign running that is driving directly to our tool. If that’s the case, I want to make that clear in the profile.
So regardless of how I chose to present the sourcing data, I'm not done with this variable. A profile, by definition is a sort of custom analysis. You’re working to find what’s really important in describing a group. In my next post on this topic, I’ll use NetInsight to drill down into these Acquisition sources and show how you further refine the presentation of the profile.After that, I’ll put it all together and show a complete profile of our Target group.
[Side Note: I’m speaking at an industry conference (AIM) in LA this week on multi-source attribution. If you’re interested in the subject and would like a copy of the presentation just drop me a line! And - though I am sure he's not reading this - I have to add a note of congratulations to our own Jon Entwistle who tied the knot this weekend and is off to Costa Rica for his honeymoon!]