I believe in simple campaign structures. From a grouping/theming standpoint, I tend to view keywords as 1’s and 0’s, and I don’t put too much stock in thematic groups at the campaign level. Given that PPC Associates uses a single-keyword ad group methodology, theming, as it related to ad copy, is irrelevant. What I do care about on the campaign side is whether my campaign structure accomplishes the following: best use of campaign-level settings and enabling of efficient optimization and scale.

The first piece is relatively simple. If there is something unique about the campaign settings you need (relative to the rest of your account), you need to break it out. This runs the gamut from geos, sitelinks, day-part settings, budgets, and demo settings to networks selection – splitting search and content is the most basic, and most important, campaign-settings-driven segmentation. While combining campaigns simplifies things, would you really want to run the same sitelinks for [ninja swords] and [cheap automobile]? Well, assuming you sell swords and cars…but that’s another matter altogether.

The second piece is where it gets a little hairy. Different people have different takes on what it means to scale your SEM campaigns efficiently. In many cases, people view thematic campaign structures as a means to scale faster. Don’t get me wrong, I’m all for people creating a campaign structure that fits with how they process data. If you have a hard time discerning between unrelated KWs if they are part of a single campaign, then maybe theming is what’s best for you. Just be mindful that what seems more manageable at the outset can quickly spiral out of control once we get to advanced optimization.

What I’m driving at is that, if you’re using advanced segmentation techniques, you’ll want to start with as few campaigns as possible. Imagine you start with a tightly themed set of 10 campaigns. Not a big deal initially, but what if you discover three distinct sets of behaviors at the geo level? And what if weekdays outperform weekends? And nights do worse than days? Granted, there are some bid-dampening rules you can put in place, but if you want total control, the segmentation I outlined here takes your 10 campaigns and could easily turn them into 120. (Start with 10×3 geo segmentsx2day of weekx2time of day.) Granted, the example I list is rather extreme, but you get the idea. Every new campaign creates another line in reporting, another budget to manage, and just more headaches.

Ultimately, it all goes to the concept of “keyword sculpting.” Think of keywords as blocks of clay. In order to create art (profitable KWs), you need to trim away the excess clay and shape the block. There are many ways to accomplish this with KWs – things that go beyond ad and LP testing. Segmentation according to a series of different dimensions is how to get the job done.

The first step is figuring out whether it’s worth segmenting or not. Like in my previous example, it is easy to run wild if you aren’t careful. First and foremost, PPC optimization should be data-driven. Remember, if you have some pre-existing business reason for only targeting certain geos and times of day, you should already be doing it. What I’m addressing here is what to do next with your simple campaign structure.

In a nutshell, justifying a new segment comes down to two metrics: CPCs and CVR. These two metrics are meant to address two things – difference in the makeup of the auction (i.e. more or fewer competitors) and differences in the makeup of your audience. These metrics should be evaluated relative to the average of the data set you’re exploring. Regardless of what dimension you’re evaluating, you should be looking for deltas large and consistent enough to warrant breaking out a segment – good or bad. Secondly, you need to factor in whether there’s enough volume to make an impact. Even a 50% delta in CVR isn’t really worth fretting over if it’s for a segment that makes up .5% of your total traffic. You’ll find more efficient ways of optimizing without adding unwanted complexity to your campaign setup. Now, in the event you find multiple small volume segments (think of a chunk of states or metros) with similar performance patterns, then group them and optimize them accordingly.

Bottom line – if the publisher settings allow you to break it out, you should be reviewing the data regularly to see if you can take advantage. I’ll leave you with a list of segments you should review. It’s gotten remarkably simple now with the advent of the dimensions tab in Google, so there’s no excuse. You can do this!

Geographic – countries, states, metros, etc.

Time – day of week, time of day

Network – search vs. search partners (not ideal on Google yet, but if MSN’s letting you do it, Google will catch up)

Device – computers, mobile phones, and tablets (there are 2x more tablets in homes than there were last year)

GDN targeting – text vs. image, contextual vs. audiences vs. topics vs. retargeting

Have a few more?  Post them in the comments section.

- Sean Marshall, Director of SEM

This entry was posted in Featured, Google. Bookmark the permalink.

6 thoughts on “Simple campaign structures + segmentation = optimal

  1. Hi Sean,

    You mention that theming is irrelevant to ad copy since PPC Associates uses a single-keyword ad group methodology. But I imagine you don’t adhere to one ad group per keyword for each and every keyword (match types excluded) in your account. At some point – possibly for tail or low-impression keywords – you do rely on groups of thematically chosen terms, no?

  2. Thanks for the comment Terry.

    As of right now, we SKAG everything so there are no tail term groups.
    Our approach is focused on winning terms so if it’s driving results, it gets isolated.

    For the low imp terms that convert, but don’t have enough volume to get SkAG’d, we just rely on broad match modified. The effect is that of clustering tail terms, buts it’s still a single keyword.

    Hope that made sense

    Sean

  3. That does make sense, thanks. You know what would be cool? If AdWords ACE (experiments) would allow users to set up split-tests between 2 campaigns, and we’d have the ability to test different approaches – like the SKAG approach – on campaign performance.

Leave a Reply

Your email address will not be published.


*

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>