Using NetInsight To Build Visitor Segmentation
Posted by Gary Angel
I’ve been working through one of the truly fundamental tasks for any web analyst – producing a rich descriptive profile of a visitor segment. I’m using Unica’s NetInsight for this just as a change of pace from all of the Omniture Discover and Data Warehouse examples that I’ve done in the past. In my last post, I started off building a profile of a particular group of product-interested users on a financial website.
The first variable I chose to profile was visit source – and right away I found a fairly significant difference between the target population and my chosen control group.
As you can see, my target population is much more likely to come Direct, from Yahoo and from key Financial sites than is my control group – and much less likely to come from Google. At the end of my last post, I made the point that the profile isn’t like a report – it’s more like a mini-analysis. This Yahoo discrepancy, in particular, begs explanation and so I wouldn’t leave it as is. I want my profile to explain how we’re getting these visitors and the numbers about Yahoo and Google are far from self-explanatory.
So my next step in building out this profile is to drill-down on the Yahoo numbers. In NetInsight, you can easily add a dimension to any report. I had this report which shows more Yahoo traffic than Google traffic:
By adding the Campaign Channel dimension, I get this report which shows almost no Yahoo campaign traffic:
NetInsight’s handling of missing values in sub-dimensions is confusing. When you add a dimension, missing values from the sub-dimension are simply eliminated from the whole report. So instead of seeing a line-item under Yahoo with no campaign channel, I see the Yahoo number reduced to the visits that have a campaign channel assigned. I sure don’t love that approach. But for our Profile, the important point is that it isn’t a known campaign driving the Yahoo traffic.
That’s puzzling.
So I went back to the original report and drilled-down to get the actual referring URLs. In NetInsight, I get this break-down just by left-clicking on a line-item.
I’ve blurred the actual lines here but what’s important is that 75% of the traffic came from a single page and the rest came from a long-tail of pages and Yahoo searches. So I clicked over to that page and lo and behold, I found a large featured ad for the product in question driving directly to the content I’m analyzing.
It’s a campaign after all!
It’s just not properly tagged so NetInsight isn’t recognizing it. Yes, it’s true: half (at least) of all the interesting findings in an analysis turn out to be data quality problems. But at least by drilling down, I’ve made sure my profile won’t reflect those issues.
My next step was to check those financial sites and see if a similar issue existed there:
It didn’t. These links from the #1 financial site were mostly long-tail news articles. There wasn’t one dominant source and there was no campaign.
So for my Profile, I classified the Yahoo traffic from the key page (only) as a sponsorship campaign and turned my attention to the Google numbers.
My first question about the Google numbers was simple – were they really lower than average or was the pie getting distorted by the Yahoo campaign. Look at it this way: suppose that for an average visitor in my control group, 35% are sourced by Google. Now suppose that my target group is completely identical in sourcing but I add in a bunch of additional people from the Yahoo campaign. My percentage from Google is going to fall in the target group and it will look as if visitors are less likely to source from Google. So I quickly re-calculated my percentages after taking out the Yahoo campaign:
Taking out the Yahoo campaign reduces the other Yahoo traffic so that the Target looks exactly like the Control. But Google is still way below average and Direct and Financial sites both well above average. We’ve learned that the Financial number is real and seems to be driven by long-tail news linkages to our product area. Score one for good PR. But what about the Google numbers?
To check Google, I looked at Google PPC sourcing for my control group. In that group, 14% of the traffic was sourced by paid ads. I did the same check for my target content visitor segment and found that only about 7% of the group was being sourced by paid ads. What about organic? Well, 25% of the control group visits were sourced by natural search on Google but less than 4% of our target group were sourced this way. A quick check of the relevant terms confirmed the problem – my key pages for this product are very poorly indexed on Google.
Here’s the final version:
At this point, I feel like I’ve built a pretty good profile of sourcing for my target population. This compact little section tells my audience how we’re getting visitors to the target area, how sourcing for these visitors compares to my broader control group, and even includes some built-in actionable directives. I’d have to eliminate those directives to fully automate this report. But if I’m preparing a profile for a presentation, this is what I’d like to have to describe where my target visitors come from.
There are a couple of points worth making about this exercise. First, building a profile takes work. I’ve only done a small part of it here but you can see that building a good profile of a visitor segment isn’t a cookie-cutter activity and it can’t be done well in five minutes. What I’ve done here is more valuable, more accurate, and more interesting than the original referring domain report. It’s also, and this is at least equally important, much less likely to be misunderstood or misused.
But to get to this point, I had to do a fair amount of slicing and dicing in NetInsight. That’s why those slicing and dicing capabilities are important in a web analytics tool!
Second, everything I’ve done is in comparison to the control group not the total site. That’s vital, and it makes the whole profile vastly more interesting.
Finally, when you do a profile well, it will be no surprise if you can start to suggest actions and insights right away. I picked this target area of our client’s site at random (truly). But in the course of building just the sourcing part of the profile I found one data integrity issue and came up with at least three reasonable improvement opportunities for the product manager. Pretty cool.
Done well, real analysis is a powerful thing. And good segment profiles are a powerful tool for good analysis.
About the Author: Gary Angel is the author of the "SEMAngel blog - Web Analytics and Search Engine Marketing practices and perspectives from a 10-year experienced guru.
Leave a Reply