An Easy-To Follow Method For Ad Optimization & Testing

Most of us in the PPC world know that testing ads is a great idea – almost necessary optimization work. There are plenty of articles around the web expounding the benefits: increase QS, lower costs, boost conversion rate … all good stuff.

In this article, let’s tackle the basic block and tackling necessary to get ad testing up and working for you.

What Ad Is Better?

With ads, as with keywords, this is not always black-and-white. There are many metrics we can measure, but which should we focus on? What is better — Higher CTR or higher conversion rate? More Revenue or more margin? Higher QS or lower costs? It is possible to have really high CTR ad that costs you a ton of clicks without making sales, and vice-versa.

No simple answers here – every business is different. Let’s cover some basics and agree on some optimization parameters just so we have something interesting to talk about in the rest of the article.

Let’s consider an ad that reads something like: “Free dvd’s, porn downloads, and Myley Cyrus’ phone number.” You are definitely going to get a lot of clicks! Your widget sales at ACME company might not do so well though.

On the other hand, we could have an ad that reads something like: “Do not click here unless you want to pay $100 for a widget and have your credit card in your hand.”

You can imagine we will not get many clicks relative to the other ad. We would expect a higher conversion rate than normal for those who actually do click-through though. So we’ve got the “free dvd” ad that gets high CTR and low conversion rate, and the “don’t click” ad that gets low CTR and high conversion rates.

A third kind of ad might be the “what is that?” ad. It might read: “We are something-you-didn’t-search-for experts. Great shipping!” and it points to the homepage.

This ad might perform better than both of the above ads overall, but its notable characteristic is that it is untargeted and points to an untargeted landing page. So what do we optimize for? What is that “overall” metric?

For this article, let’s settle on one. We will use Margin. That is: (Advertising Revenue – Advertising Costs) / Advertising Revenue. At the end of the day, it is often the money in your pocket that matters.

By the way, the same techniques work regardless of what metric you choose to optimize on, but for the sake of clarity, we are going to stick with margin for this article.

AdWords “Optimize” Option

There is a Campaign setting called “Ad Rotation.” It is located near the bottom, in a section called “Advanced Settings,” then “Ad delivery: Ad rotation, frequency capping.” You may need to click “edit” to reveal the options. The options are:

• Optimize for clicks: Show ads expected to provide more clicks
• Optimize for conversions: Show ads expected to provide more conversions
• Rotate: Show ads more evenly

Let’s start with the first option, the default setting. This setting will optimize your ads according to the highest click-through rate. That is a good thing, although it might not necessarily correlate with improving your other metrics.

This option will favor the “free dvd” ad, from above. High CTR (lots of superfluous clicks), and low conversion rate combine to make bad margin. Let’s move past this setting quickly.

This gets us to the second option, the “Optimize for conversions” option. This option might favor the “don’t click” ad, from above. High conversion rate is not necessarily high margin (if we click once and sell once, we have still only sold one widget).

To clarify an important misconception: this option does factor in both CTR and Conversion Rate, which can be approximated with something like CTR * Conversion Rate. It is not the case that it considers Conversion Rate only, regardless of CTR. (Thanks to Jim Prosser at Google for the important clarification.)

We want something that will help us tune and maximize both parameters, as well as driving costs down and revenue up – the sweet spot for margin.

There is one more aspect of both of these optimization settings that is troublesome for our purposes. These settings create a dynamic traffic split to the highest CTR/Conversion-rate ads, thus pulling traffic away from the other ads. That is a good thing if that is what we want, but it makes data collection really challenging for our purposes. We can’t get a fair test if traffic is not split evenly amongst our ads.

For you contrarians out there at this point:  While it is theoretically possible to adjust for the varied traffic, the statistics alone are beyond most advertisers, let alone the traffic requirements, and even then, we are simply unable to control for some of the elements introduced by the optimizations. They favor particular keyword + ad combinations, for example, something we can’t adjust out to get a fair comparison of a challenger ad.

The first two options don’t quite meet our needs, so what’s left?

Now we are at the recommended setting – “Rotate”. This setting will evenly rotate amongst all of our ads. We need that in order to do controlled data collection and testing. This option is like an A/B traffic splitter. It is the random selection control that is the basis for all of those psychology experiments we learned about in school.

This setting will allow us to “challenge” the “champion” ad and get a useful measurement of which is better. First step: switch to Rotate.

Let’s check in. If you are currently scratching your head thinking that Google’s optimization options above must be better, then that’s great. Feel free to use them.

However, if you follow so far and wish to continue, we will continue with the assumption we are optimizing for margin, and we are going to use Rotate to help us get there. Now, let’s get down to some of the dirty mechanics of ad testing based on Rotate.

The Mechanics Of Ad Testing On Rotate

Now that we have Rotate turned on, what next?

At this point, I am going to switch to a style of writing where I simply recommend a course of action which works. There are a lot of options here and we could write a book analyzing the different techniques. But you are reading this online, your time is limited, and I am going to recommend we do it this way.

Add some ads, and tag their URLs with today’s date. I use the format: adid=YYYYMMDD because it is easily sorted and decipherable. Today is April 20th, 2011, so my ads are tagged like this: “https://www.ACME.com?adid=20110420”.

You can leave up your other ads if you wish, or even make copies of them and tag them with today’s date. The point here is, get some ads running with tags on them, and make sure they have enough traffic so we can wrap this up before the cows come home.

After you get enough data, usually a few thousand clicks and tens of conversions for each AdGroup you want to optimize, download an ad report. Include the ad text, the major metrics, and of course the destination URL. Now, add a new calculated column in your spreadsheet of choice and call it margin.

Sort the spreadsheet by margin, descending, and filter for each AdGroup in turn. Each AdGroup might have a clear winner. Sometimes it will be the highest CTR, and sometimes the highest conversion-rate ad, but you may notice it is often a strong combination of both (but not necessarily the strongest of either), that wins. You may have multiple champions that perform similar to each other, and better than all the rest.

You may find that the results are not obvious. You may need to collect more data (leave them running longer) in order to get data you can be confident in.

Once you have enough data that you have ads that win, we begin a new cycle. Pause all the losers, and add some challenger ads tagged with today’s date.

There are many resources online for ideas for how to manipulate creative, and this article is not focused on creative ideas. I will just add the universal suggestion to have a healthy mix of systematic iterative testing and out-of-the-box, try something different testing. Now we have some challengers and the old champion(s).

Now, make copy of the champion(s) and tag them with today’s date. By adding a new copy of the ad, we are accomplishing two goals. First, we are giving the champion a fair shake against the challengers – it is starting as a new ad just like the challengers.

Secondly, we get to benefit from having found a champion – it keeps running. It also allows us to emphasize the amount of traffic going to the champion versions, while still getting even traffic for our test.

With our example, let us imagine we have 1 champion, 1 champion-copy, and 2 challengers. In this scenario, 50% of your traffic still going to a champion, and 75% is involved in a new test.

Likewise, you can make more copies of the champion to send more traffic there, but be careful about hampering your ability to collect enough data to make a timely and valid conclusion about the next champion.

Another brief tip: You may quickly discover that certain ad text formulas work well across numerous AdGroups. For example “Buy our {stuff}” might universally be a better headline than “Info about {stuff}”, assuming that you substitute {stuff} with whatever each AdGroup is targeting. Leverage that.

In my testing, I often add something to the adid tag that indicates what formula I am using.

Continuing the example above, we might have respectively:

• https://www.ACME.com?adid=20110306_buy
• https://www.ACME.com?adid=20110306_info

Here are some related articles you might enjoy:

• The Pitfalls Of A/B Ad Split Testing, Part 1
• CPA Vs. ROI Optimization: What’s The Best Practice?

Finally, here is a quick recap. The optimize options do a decent job of what they do, but the don’t work well with a well oiled ad optimization plan.

Use Rotate, add your ads all on the same day, wait, and then pick your champion(s). Pause the losers, add some challengers and a copy of the champion(s), wash, rinse, and repeat.