The myth of SKAGS (Single Keyword Ad Groups)

Category: Notes on Digital Marketing | 4 min read

There are a bunch of articles flowing around right now talking about the magical amazingness of SKAGS or Single Keyword Ad GroupS.

I remember when I first learned of these: It was years ago (not sure how many!) and I was working at an agency. Back then CPA bidding was in it's infancy and we didn't trust it.

We implemented SKAGS as a proprietery system that always got our clients amazing results.

When we set them up we would see immediate increases in quality score, CTR, conversions and more. It was fantastic.

These are the same things that are being touted right now but to be honest I just don't see the value in them any more. I think they can do more harm than good. Read on to find out why...

First, what the heck are they?

Setting up SKAGS basically means instead of making ad groups based around themes such as "auto" "travel" and "brand" you set them up specific to the keyword. This could be something like "Compare Auto Insurance Rates." This ad group could then hold the different match types for that keyword.

The theory is that by doing this you increase the quality score because the ad will be super closely tied to the keyword (usually literally using the keyword within the ad) and you could theoretically also make the landing page tied to it super closely. You also get the bonus of people generally clicking the exact keyword they typed in more often than not, boosting the CTR, which then boosts QS.

A better quality score gets you a lower CPC and better CTR, meaning it's a win across the board. There are problems though:

They don't scale

If you have maybe 10 to 50 keywords in an account, SKAGS can be a huge benefit but once you get beyond that and get into the hundreds or thousands it just becomes an absolute mess to manage.

There is this issue with PPC I see all the time where marketers always come forward with big plans that they will do everything right. They will set up SKAGS, constantly A/B test their ad copy, always be trying new things, etc. Then once they get to actual day to day of running the campaign tons of stuff gets forgotten: ad copy gets ignored, keywords grow stale, etc. The focus generally falls back on pacing the budget and reporting above all else.

The bigger your campaign gets the more stuff is going to fall through the cracks in terms of opimizations. When you are setting up SKAGS you are setting up an obligation to constantly work on optimizing on a huge scale across the account. The bigger the account, the bigger this obligation. To be frank, this generally sets people up for failure.

The other issue is building that many landing pages, while possible, is annoying as hell. You could technically use dynamic headlines and the like but the QS algorythem wouldn't pick those up as well as static ones.

Even if you DO create that many landing pages, A/B testing them all is also a complete mess.

You have to keep in mind the reality of what you are setting up, not what could happen in a perfect world.

Lots of other little problems come up as well - it's not a structure that is easy to pass on to another employee (what if you get promoted?) or to switch agencies with.

CPA and machine learning hate it

This part might be a bit controversial but I'm a convert to CPA (and now ROAS) bidding models. I used to be with the people saying manual CPC is always better but after numerous experiments where I was put up against the machine, the machine always won.

The issue is that the machine needs lots of data.

CPA bidding works best when there are lots of conversions. It seems to do this at an ad group level. The more conversions there are in an ad group, the better the machine learning does.

SKAGS on the other hand split up these conversions across hundreds of ad groups, meaning the bidding model wont be anywhere near as good.

You might ask how you can make good ads even if you have lots of keywords in one ad group and the answer is two fold: first, having lots and lots of ads in the group helps as the machine will try to show the right ad to the right person and also by making the ads dynamic. My article on creeping people out at scale can help with this.

You might say: okay, but what about outside of Google? I honestly would reply with the same answer. Facebook's machine learning is doing insanely well, as are all the other channels I have tried it on.

QS has changed

Remember how I started this article out by saying I was doing this years ago? Back then, this was fantastic and really did improve QS. These days I'm not so sure it is as effective as it used to be.

QS used to be a pretty simple calculation but with Google's obsession with machine learning it now uses more signals than you can count. Just saying "ad relevance" is no longer meaningful.

These days I regularily get QS of 8 or higher on keywords within adgroups that have hundreds of others. I do this by having lots of great ads, fantastic landing pages and by being hardcore about my negatives.

In the end

Set up your campaigns in a way you realistically can stay on top of them and make sure you'll actually be able to optimize and get results. You don't have to have a single adgroup with all your keywords, but don't go the opposite and make a new ad group for every keyword you add.

Find a happy medium.

Other Articles