DEV Community


Posted on

How does A/B testing work?

I wanted to have a much more implied answer to this question for the non-techy ones who are not much familiar with the topic but find it fascinating. This is my attempt to define A/B testing, as I know it. Feel free to correct me.

A/B test experiments are strictly based on a data-oriented approach. You use your already stored data to determine the areas lacking the mass appeal to convert the audience into customers. You apply the A/B test and receive the data about if the new version is bringing in some conversions or not.

What A/B testing does in all this is place triggers on the website, which activates the changes placed while building the experiment(with the help of codes and A/B testing tools and more), after a set action that could be a click on the page, while navigation to another page in the website, in form of pop up, etc.

That way the test loads on the screen of the random visitor (who fell in the variation’s bucket). A/B testing is a carrier as well as an agent between idea and conversion.

The basic ‘A/B testing’ works as a mechanism to enable testing of optimization efforts that to validate the ideas behind that change. To prevent the confusion it goes step-by-step.

Identifying, Ideation, and Hypothesis

To find the direction for your experiments, analyze your existing data to find the areas which could use some optimization. Within your analytics, you can find out where you have the most traffic and which pages convert the most by observing screen recordings, heatmaps, analytics, and hosting site data.

After checking the pages you finalized for experimentation, you’ll need to come up with ideas for the variations and what changes should be implemented. The term and process which is used for the idea generation is Conversion rate optimization.

Those ideas could use a hypothesis to convey the motive behind the change. That will help the developers understand your perspective while moving on with the creative part.

Experiment Building

Experiment building is done per the requirements of the presented idea to be mirrored, where the design files will be perfectly converted into HTML/CSS. Other necessary skills, such as JavaScript, coding, UI, and UX graphics, and more, are used to build a high-quality experiment. The developers ensure the experiments work well in all applicable devices and browsers.

Implementation and Analyzing

The test is implemented on the site per audience targeting. It’s only viewed by a specific set within your audience. Those who fall under the version’s bucket (the audience which fell in the variation’s bucket during traffic allocation) will be able to see it; others will see the original version only.

The performance of the tests has to be tracked until the experiment gains enough data for analysis. It takes some time to collect the sample size to evaluate the results.

The experiment records the results in the reports, those results are the KPIs like clicks, views, submissions, lead generation, sales, exits, and more. The winner between the original version and the Variation version (the one which includes the change) is determined on the basis of those results recorded in the reports.

If results are positive then the variation replaces the original and continues to increase the expected conversion rates.

The process may vary as per the experiment’s nature, type, and skills used. But overall it all depends on the data and how you make use of it to convert more users into customers.

via Brillmark

Discussion (0)