All advertisers need to rotate the different ways we have to impact and convince our audience in order to discover the most optimal way to become their Top Of Mind. This trial and error is known as testing and, in digital marketing, it can be done on any element that influences a potential consumer to make their final decision.
The problems we face in the online environment when choosing between two creatives or between two landings are many, and there are many factors that can bias the data of an experiment done on the Internet:
Seasonality
Demographics
Interests
Locations
Messages
and many more
Maybe an active ad one week might perform better than another ad the following week, not because it was better, but because there was a special offer for a few days or because the audience it impacted was more relevant, or because it appeared in different places than the other.
So, is there a way to counteract these factors, and is it possible to get two different ads to the same person without bias? Not really, but there are ways to minimize the chances that these factors can influence the results of our experiments within Google Ads.
Google Ads Experiments
For some time now, Google Ads has allowed advertisers to perform A/B tests with the messages shown in their ads, with the landing page where the user lands, or to compare performance between Smart Bidding strategies, using a tool within the platform.
The main added value is that, by creating an experiment, you are rotating two different elements (i.e. two different landings) during the same period of time, for a very similar audience. In parallel, we can also compare the data of the experimental element vs. the original element, indicating when there is sufficient statistical relevance to prove or disprove the hypothesis on which we based the test.
So... How does it work?
Well, there are several steps to follow in order to create an experiment in the platform and in some of them we will find some concepts that are worth taking into account when making decisions.
Within Google Ads, we go to the Drafts & Experiments section.
When creating an experiment, we will create a new campaign based on the structure and history of the original campaign. Before activating this campaign, we will have to work on the changes we want to make on a draft.
2. Click on the blue plus (+), choose the campaign we want to test and give it a name. In our example we choose the campaign ES_ESP_S_Product, we name the draft Product V2 and save.
3. Now we are working on the draft where we will make the changes we want. In our example we change the URL of the landing page to which we direct our users. If the conversion rate of the users that enter our ads improves, stays the same or gets worse.
4. Once the changes have been made, click on the APPLY button. It is important that we then give the option to Perform an Experiment, from which several options will be opened.
5. We must give a name to our campaign experiment, in this case we will give it the same name that we have given to the draft: Product V2.
We can also give it a description, such as the change we have made.
We set the start date to be Monday, May 18 and end one month later to get enough data.
The % that we attribute in the "Experiment Split" section will indicate the percentage of the original budget that we will pass to the experiment campaign. In our case it will be 50% for the experiment and 50% for the original campaign.
In advanced options we have two ways to segment the audience we impact:
Search Based: Google randomly assigns a variant to each search. This format allows us to get data faster at the cost that the same user who searches several times, could see both variations.
Cookie-based: Google randomly assigns a variant to each user. This format is slower to fetch data but a user who searches several times will always see the same ad at the cost of fetching data more slowly.
Once each box is completed, we save the experiment and the new campaign is created.
6. After a few days, we will want to see how our test performed compared to the results of the original campaign. As for a normal campaign, we simply look for it in the Campaigns section and click on it. As soon as we enter, we see that the interface is different from a normal campaign, we see some metrics in big size with percentages underneath:
These numbers correspond to the results of the experiment and the % underneath to the change from the original campaign by marking a minimum and a maximum.
But how do we know that the data are reliable?
In statistics there is the concept of "p-value". This tells us the margin of error we can afford in the experiment to prove or disprove the hypothesis we are working on.
In the case of Google experiments, we always work on a p < 5. When we see a blue asterisk () next to the percentages, it means that the change with respect to the original campaign will be, at least 95% of the time, between the % in the parentheses). Examples:
For conversions, we know with 95% reliability that the experimental campaign has increased the number of conversions by between 5% and 33% compared to the original campaign.
In Cost / Conversion on the other hand, we do not know with confidence whether the number can go down by 12% or up by 8%.
In global terms we see that we achieved an improvement in performance (higher CTR and more conversions), so the decision would be to apply the changes made in the experiment to the original campaign.
Conclusions
Testing is one of the great pillars of digital marketing. With it we can discover the most optimal ways to convince our audience to choose our products over our competitors.
Google offers this own tool with measurement included with which we can make all the changes we want from a campaign to your experiment to be able to test what we want.
