A/B and Multivariate testing
A/B testing enables you to compare two versions of a web page and identify the one that generates the higher conversion of desired outcome (like purchase). Typically, one version is the original and the other is the test.
Multivariate testing is a more elaborate version of testing in which multiple components of a web page (examples: headline, image, and CTA button) have different versions. Then each permutation of the components and the corresponding versions are tested to find the combination that generates the best outcome. However, in most cases, this elaborate testing is not practical because each combination requires significant traffic for statistically relevant comparisons. “Sitegainer” is among the tools that help with this type of testing.
Instead of full-fledged multi-variate testing, a simplified (“managed”) multi-variate testing is more popular. In this testing, multiple versions of the page are created with input from professionals and then compared against each other to find the best. (i.e. Random permutation of components is not used in this methodology). Tools such as Optimizely and Google Optimize do a great job in facilitating such testing.
The process for testing with the help of these tools is straight forward – give the website url to the tool and it “captures” the website to their system. The tool creates a code snippet which is inserted to the original website. Then multiple versions of the website can be easily created using the tool (by modifying the original version using the embedded editing tool). Then specify what proportion of audience can see test versions and voila! – testing starts.
Let us look at the simpler case of doing A/B testing without testing tools like Optimizely – with just Google analytics. This methodology requires creation of various versions of the website externally (without help from tools embedded in Optimizely or Google Optimize).
To set up the experiment, go to Google Analytics > Behavior > Experiments
- Click the button “create experiment”
- Go through steps including specifying experiment duration and confidence interval
- Then provide the url of the original version and the test variation(s). As mentioned earlier, the variations need to be created manually.
- Generate the code and insert in all the pages (in original and test variations)
- Launch the experiment
Google analytics will provide up-to-the-minute stats and will announce when the experiment is complete. You can use the winning version going forward to gain the maximum number of future conversions.