A / B testing: how to set up an experiment in Google Analytics

1. Tatyana Bondar
2. We call webmasters.
3. Customize the system.
4. Fill in the data about our experiment.
5. Choose a goal.
6. We indicate the percentage of traffic for the test.
7. We enable even traffic distribution between all options.
8. Set the minimum experiment time.
9. We set the confidence threshold.
10. Set the experiment settings.
11. Set up the experiment code.
12. The code can be sent to the webmaster.
13. Paste the code manually.
14. Go to the "View and Start" section. We start the experiment.

At Inweb we practice training. These are short meetings with colleagues, where experienced specialists teach newcomers. The results of such meetings are summaries for the training of beginners.

The last topic that we talked about is devoted to A / B testing and tuning the experiment. We must admit that on the net we did not find material on this topic that would completely suit us. Therefore, catch a brief summary on setting up testing in Google Analytics!

Remember how they taught at school: only the axiom does not need proof.
Hypotheses and theorems must be proved. Today we will talk about the first. It is its presence that leads us to the need for A / B testing of a page on a site.

A / B testing is a study that helps determine which kind of page is best suited to our goals. Google Analytics distributes traffic to several variations of the same page and, analyzing the statistics, we conclude which content is more effective for the current task.

To conduct A / B testing, as you already understood, there must be a hypothesis - an assumption that you are testing. Consider the example of an order form on the website of our agency.

We had an application form. She looked like this:

Order Form - Original

Remembering gender policy, we decided to test the hypothesis: the woman in the photo converts better than the man :). We created a similar application form, but with a photo of Oleg, the head of the sales department. Can this somehow affect the behavior of users and improve / worsen the performance of filling out the form?

The test version of the form looks like this:

Order Form — Option 1

Tatyana Bondar

Inweb Blog Editor

So, there are a few steps to do this.

1. We call webmasters.

   We ask the programmer to create exactly such a page with the order form, only add a photo of Oleg. This will be our test page. An important point - the test page must be closed from indexing.
   You can do this using the meta tag - <meta name = ”robots” content = ”noindex, nofollow” />. Just copy it and paste it inside.

2. Customize the system.

   Go to Google Analytics. We pass to the section "Behavior", further "Experiments".

   

   Google Analytics → Behavior → Experiments

   Create an experiment.

   

   Create an experiment

3. Fill in the data about our experiment.

   • We call the experiment.

     The name is best chosen so that anyone who goes into this section of the presentation understands what is at stake.

   • Choose a goal.

     You can select any parameter from the proposed ones. Or create a new one. You need to indicate the goal that you are focused on and what you want to measure as a result of the test. The system, to achieve this goal, will choose the option that will be most effective. We want to test the page with the application form, respectively, in our case, the goal is the successful execution of the application.

   • We indicate the percentage of traffic for the test.

     Percentage of traffic participating in the experiment is an indicator that helps determine what percentage of the total traffic you are ready to use in testing. For results to be cleaner and more objective, you need a lot of data. In our case, we select 100%. Go to the advanced settings.

   • We enable even traffic distribution between all options.

     The system will evenly distribute users to 2 pages - the original and the version that we are testing. Otherwise, the system will give more visits to the option, which at the moment will be considered more advantageous.

   • Set the minimum experiment time.

     If you assume that there is enough traffic to your page, it may take 2 weeks for the test. Please note that you set a minimum period.

   • We set the confidence threshold.

     Remember, the higher the threshold, the more accurate Google Analytics will produce the results, but spend more time on such an experiment. If you want to receive notifications of changes in your experiment by mail, turn on such notifications and specify the address of the mail to which they should arrive.

     

     Create an experiment - step 1

4. Set the experiment settings.

   The original page is an existing page. In our case, an application form with a photograph of Julia.
   Option 1 - an alternative version of the existing page (application form with a photograph of the head of the sales department).

   

   Create an experiment - step 2

   There may be several options for the original page. If the traffic to the site, in your opinion, is enough, you can use more than 2 test page options. Making a decision on your own, without enough data, is risky: the system will not be able to draw conclusions.

5. Set up the experiment code.

   Using this code, the system sends the user who entered the application form, to the source page or to the test page. There are 2 ways:

   • The code can be sent to the webmaster.

     So he inserted it into the site. Moreover, the system is so smart that she writes a letter. You just need to correctly enter the email programmer who will be involved in the implementation of the code on the site.

   • Paste the code manually.

     Copy the experiment code that the system offers and paste after the head tag that opens at the top of the original page.

     

     Create an experiment - step 3

6. Go to the "View and Start" section. We start the experiment.

   After the experiment is launched, the system will need time to collect data. Please note: the time you set is the minimum, the maximum is 90 days.
   You can go into the experiment, see the statistics and draw conclusions, which page works best for your purposes. If the results are clear to you, and the deadline has not yet reached the end of the experiment, you can stop it manually. But we recommend not to rush into this matter and wait until the system itself lets you know that the testing is completed. Well it is, for the purity of the experiment.

   By the way, what do you think, what application form did we win? Leave your suggestions in the comments. And experiment! “After all, when the theory coincides with experiment, it is no longer a discovery, but a closure.”