We evaluate the sample of recipients

[billal hossen]

Let's also remember to set the numerical objective based on the history : before launching the A/B test, it is always a good idea to consult the previous results.

Let's create the variants
Once the hypothesis has been established, we must focus on the distinctive elements of the two variants , which can be: the image, the title or the text written in a paragraph rather than their formatting, the button of a form, the arrangement of an element on a page, but also two web pages in the desired versions, etc.

The original A and the variant B, whether pages or ads, will have to be created and published but above all managed in a precise and punctual way, to collect all the useful data deriving from the experiment.

Some tools help in this phase: Google Optimizer, VWO Virtual Website Optimizator or the services offered by advertising platforms, such as Google AdWords or Facebook Ads.

If we test one element at a time (position of a module), or a combination of elements that are closely connected to each other (position and button of a module), we will be able to draw more precise conclusions.

In general, the principle applies that the larger the sample size , the more reliable the results we obtain will be.
There is no exact number, it depends a lot on the improvement we want to achieve and the level of significance desired. There are several calculators online (e.g. Unbounce, Evan Miller and others) that in a few seconds are able to determine the right sample size and the minimum time to run the test.

The data we need to be able to use one of these cambodia telegram data calculators are the following:

average daily traffic
minimum desired improvement
current conversion rate
number of variations
The two groups on which to do the A/B test must be random, numerous and homogeneous. Normally the A/B test is performed on a sample ranging between 10% and 30% of your database, with some minimum limits.

If the database is not sufficiently large, we can still conduct it, knowing however that the results will be statistically less reliable.