Part 10 in a Series on Web Analytics and Search Engine Marketing Programs
Those who have known me for a while generally realize that I am not the most organized person in the world. I’d like to plead age, but I’ve pretty much always been a bit ragged around the edges. So it’s no surprise that in putting together the list of huddles I made a mistake here or there. But I have to admit, it was a pretty big DUH moment when I realized I had forgotten to list my own huddle! Oh well. I am, as I’ve said before, planning on doing one on SEM Analytics.
Which brings me back to the subject at hand – SEM Analytics. We’ve moved through a number of stages of analysis – many at quite a high-level. And you may find that much of the most valuable web analytics for Search Engine Marketing has already been covered. That’s especially true if your site has specific conversion points and you’ve already been optimizing conversion or revenue or lifetime value as appropriate. If that’s the case, then your Bid Management System is already handling most of the necessary day-to-day optimization, and web analytics will be primarily useful in understanding channel issues.
If that’s not the case, then you may well be using a Web Analytics package as your primary optimization tool. For the most part, this isn’t particularly challenging – it’s actually more of a reporting issue than an analysis issue.
However, here are few tips that might make this task quite a bit easier.
Most WA packages will let you track all of the following from a Campaign: responses, visitors, return visitors, conversions, eCommerce values. For SEM, these statistics are usually available at the campaign-level (which should correspond to Ad Group) and at the Keyword level. This is pretty good reporting – and if you have specific conversions and eCommerce data it will provide you pretty much everything you need for most the common optimization strategies.
However, if your site doesn’t have specific end-points and you are trying to measure the value of multiple events or of engagement then you’ll probably find the basic campaign numbers to be insufficient. You don’t typically get good measures of engagement for a campaign.
There are two strategies for handling this. The most obvious is to build a visitor segment based on the campaign. If you do this, you’ll have all the behavior for that segment and the engagement numbers (and the total value of events triggered) can usually be calculated quite easily. The drawback, of course, is that this requires a segment for every campaign. Not only is this a lot of work (and work that must be constantly maintained as new campaigns roll-out) but not all tools provide unlimited segmentation.
A better strategy in most cases is to build a segment for each level of engagement your interested in. By running the campaign report against all visitors you get the total of responses. Divide that by the number of campaign responders you get when you apply the engagement segment and you’ve got the percent of campaign responders that “engaged” at any level. So one segment can be used to calculate engagement for EVERY campaign. This technique preserves segments and saves lots of work – since the engagement segments need only be setup once and new campaigns will automatically be included in the campaign report.
For situations where you want to accumulate values for lots of actions (especially page views) and you want to dynamically optimize for them, this segment approach won’t work. Where every page view should record a certain value, it’s pretty much impossible to build an Engagement Segment that will measure total value. Every land has some value – and there is no single “cut-off” point that will measure engagement value.
In this situation, creating segments by campaign would work – but it still has the drawbacks mentioned above (lots of segments, lots of work), and are some alternative approaches. First, some tools (like WebTrends’ latest version) will let you build a score for a visitor. That score can capture discrete events like Page Views and can then be used to report average scores by campaign. That’s a beautiful approach.
In Omniture, you can use events and record a specific monetary value for every event. This works extremely well for sites that have a multiplicity of events they want to track – because you can track campaigns by the total and average value of all the events triggered. It’s a little clumsy for tracking page views, however, since it requires that you pass an event with every single view.
For other tools, you might choose to handle this situation with either custom variables or by passing eCommerce events. The former approach will work pretty well in some tools but not in others. The key is the degree of flexibility you have in reporting around a custom variable. The second approach will work for most tools, but has the same inherent clumsiness that I mentioned about Omniture events (in fact, it’s usually quite a bit worse) and, if you have real eCommerce events mixed in with your site will generally create too much confusion to be viable.
If you don’t mind the extra expense and some custom tag coding, you can use an extra report suite for doing this work. In the report suite, you send the event type (such as page view) and the campaign name concatenated together as the fully qualified page name. This approach is, I admit, quite weird. But it will give you straightforward reporting of the total value of engagement from each campaign! If you’re a media site and you can’t get reasonable reporting of engagement by campaign, a little weirdness may be worth enduring to get the answers you need.
Here’s one final tip about basic SEM optimization in web analytics tools. Regardless of whether you are using score based systems or levels of Engagement, it is often quite useful to report on multiple levels of engagement and – if scoring - on the distribution of engagement scores by visitors as well as the total and the average.
By multiple levels of engagement, I mean segments that are designed to capture different aspects of engagement or different tiers. At the simplest level, if you are measuring engagement by Page Views, you might have a segment of 2+ Page Views and a segment of 7+ Page Views. If you measure campaigns by the % of visits that are “engaged” at each level, you’ll usually see a fairly even relationship between the two. Most campaigns that score higher by the first measure will score higher by the second. And the rate of drop-off will be fairly consistent. But you’ll also find instances where specific campaigns don’t follow the site-typical pattern.
These campaigns almost always require special study. In many cases, they deviate from the expected pattern because of particular navigational issues on the site. If that looks to be the case, you’ll have to find the optimization rule you think is most applicable for that particular campaign. If you can’t find a navigational explanation, then the findings may indicate peculiarities in the distribution of visitors who are sourced from the campaign. In this case, it’s worth looking at the individual keywords that make up the campaign and also the common navigational paths to see if you can identify a sub-segment that is driving the unexpected performance.
In my next SEM Analytics post, I’ll cover some secondary optimization analysis techniques – looking at conversion/engagement by ad creative and analyzing entry pages.
Other Posts in this Series: Introduction, Searchnomics Issues, Getting Setup for SEM Analysis, SEM Data vs. Web Analytics Data, High-Level Search Engine Reporting, Analyzing Search Traffic in more Detail, Measuring Search Effectiveness for eCommerce Sites, Measuring Search Effectiveness without Conversions, Measuring Search Engine Marketing as a Channel, Measuring Search as a Channel Part II and Time-Based SEM Analysis.
[X Change is quite a limited venue – and with a lot of people signing up last week we are getting close to our limit – which is great! But there is space remaining for another ten to twelve people - so if you are a procrastinator, you can still register at http://www.semphonic.com/conf].
Comments