TPE #16: better RSAs with 3 high-impact test ideas

Responsive Search Ads (RSAs) have been around since 2018 - time flies! However, since July this year RSAs are the only available ad type in Search campaigns. Some advertisers may still be using ETAs, but in the end, everyone has to flip the switch.

Optimizing RSAs is quite different in comparison to ETAs, and a lot of advertisers are not aware of the testing possibilities. Or even worse - putting everything in the hands of Google Ads, and blindly trusting the algos, with unpinned ads only. 

In this newsletter, we’ll discuss why it’s important to take matters into your own hands and what metrics to optimize for. We will also give you 3 concrete experiment ideas, so you can start experimenting right away.

TL;DR

• The goal of the algorithm doesn’t match yours as Google optimizes for CTR predominately.

• Conversion per Impression (CPI) or Revenue per Impression (RPI) is a great testing metric if you want to maximize your conversions or conversion value within your CPA or ROAS target.

• There are many ways to set up RSA experiments. We’ll give you 3 high-impact test ideas so you can start experimenting right away.

Let’s dive in.

There are 2 fundamental problems with RSAs.

1) The goal of the algorithm doesn’t match yours.

Google optimizes for CTR. By serving the ‘best performing’ headline & description combinations they maximize their ad revenue. However, for most advertisers, a high CTR isn’t the metric of choice.

This means your goal doesn’t align with Google’s. There is a reason why Google only reports impressions for all RSA combinations (and not on additional metrics such as conversions). 😉.

2) The learning period is insanely long.

RSAs can show up to 3 headline combinations. If you were to add 15 unpinned headlines that would result in 2940 (!) possible headline combinations. 

If you assume that each combination requires a minimum of 100 impressions to test, you need more than 294000 impressions to identify the combinations that perform best. And we didn’t even include unpinned description lines.

Google might be able to increase the learning phase and rely on less impressions per combination.

Let's say Google’s AI is able to advance the learning phase by 75% - which is a lot - you would still need 73500 impressions per RSA before you can conclude anything. And it will still be based on CTR only.

Metrics to optimize for 

Before you start testing, you need to determine how to pick winning ads. When we are running ad tests, we generally focus on two different metrics depending on if you’re in Lead Gen or Ecommerce. These are both impression-based metrics. Let me tell you why.

Every time your ad is shown (impression) you have a chance of generating a conversion. The user must first click on your ad and then convert. Impression-based metrics basically show you which ad will receive the most conversions or highest revenue possible from the impression (within your actual target CPA or ROAS). 

You will need clear CPA or ROAS targets when using these metrics. Before you take any conclusions based on one of the two metrics below, you want to first separate the wheat from the chaff by filtering out unprofitable ads that do not meet your CPA or ROAS targets.

Now we’ve got that cleared, let’s zoom in on the actual KPIs below.

1) Conversion Per Impression a.k.a. CPI

The conversion per impression (CPI) is the ratio between conversions and impressions. This KPI is perfect for Lead Gen accounts. It ensures you get the most conversions possible for the impressions you receive. Here are the pros and cons according to Adalysis:

Pros:

• It takes into consideration both clicks (CTR) and conversions (Conversion Rate).
• When you want to base winners off of both CTR and Conversion rate; it’s the best metric to use.
  

Cons:

• It doesn’t care about the cost per conversion (CPA).
• It doesn’t take actual revenue into account (ROAS).

2) Revenue Per Impression (RPI)

The revenue per impression (RPI) is the ratio of revenue to impressions. This KPI is perfect for Ecommerce accounts. It ensures you receive the most revenue possible for the impressions you receive. Here are the pros and cons according to Adalysis:

Pros:

• It takes into consideration revenue (ROAS) and volume (CTR).
• This is a great metric if you want to maximize the amount of revenue (within your ROAS boundaries). 

Cons:

• It doesn’t care about the volume of conversions (conversion rate).
• It doesn’t always lead to the highest ROAS possible.

If you’re using POAS tracking this KPI will in fact ensures you receive the most profit for the impressions you receive. Basically you’ll be switching to profit per impression (PPI). 

Pro Tip - Our preferred ad testing tool Adalysis has created an excellent guide to scientific ad testing. We’ve been using their methodologies and ad testing KPIs since the very beginning. We highly suggest studying their free resources. You will have no ragretz.

RSA testing templates

Testing with ETAs was fairly straightforward. You’d run two or three one-variable-only ETAs with fixed headlines and descriptions, compare their performances and optimize for your metric of choice. RSA testing is a bit different since you have more options. We’ll discuss the 3 most impactful tests below.

1) The Mix & Matcher

If you have never run RSA tests before, we’d highly suggest experimenting with unpinned RSAs vs partially/fully pinned RSAs first. 

By default, RSAs mix and match items such as headlines and descriptions to optimize towards combinations that people respond to best. Pinning certain items lets you tell Google which headline and description positions should appear where you pin them.

This test is quite easy to setup. You should create 2 identical RSAs, preferably with as many items as possible: 15 headlines and 4 description lines. Once you’re done, pin certain headlines and/or descriptions in the second ad, while leaving the other one unpinned. Don’t just randomly start pinning, but think it through as if you were a skilled copywriter (triggered).

2) The Copywritooor

Good copywriters are often better than AI at organizing items, such as headlines, to persuade the user to click and convert. If you got lorem ipsum dolor sit amet tattooed on your forearm this is the one for you.

For this test, you want specific headlines or description lines to test against each other, and you do not want to rely on any machine learning algorithm at all. This test is also known as the fully pinned vs. fully pinned ad test.

In this test, you’re essentially creating an ETA vs ETA test. You want to create 2 or 3 RSAs while pinning all of the headlines in each RSA. Aim for 2 or 3 pinned items for each headline. We personally prefer testing with headlines first, since these are the most impactful. You can set up the same experiment for description lines later on.

We suggest using a framework for the pinned headlines. You can customize this for your own needs, but let us give you two examples.

Example #1

• Headline 1: Headlines related to the ad group and underlying keywords
• Headline 2: Benefits of your product or service
• Headline 3: Call to actions

Example #2

• Headline 1: Headlines related to the ad group and underlying keywords
• Headline 2: Benefits of your product or service
• Headline 3: Social proof or authority

This way you can still let the algo create some variations, but you are making sure the ads are as persuasive as possible by making sure all combinations are in a logical order and contain the correct type of copy.

Pro tip - Leverage Keyword Insertion for headline 1, so your keywords can be included dynamically to increase your AdRank.

3) The Micromanager

If you still want to leverage Google’s AI, and have lots of ad variations, while still micromanaging some items, this is the template for you. 

This is basically the same experiment as experiment #2, but instead of pinning 2-3 headlines per available headline location, you’d only want to pin 2-3 headlines for just one headline location. This test is also known as the partially pinned vs. partially pinned ad test.

Maybe you’d want to test one of the following:

• Which benefit performs better in headline 2?
• Do I communicate the promo in percentages or absolutes?
• Which CTA works best?

In conclusion

As Google Ads continues to automate things, it will level out the playing field. Everyone will be running the same strategies and techniques eventually. So how do you stand out? Right - with your ads. So, write those hypotheses, set up those experiments, and get creative!

Cheers,

Bob & Miles