A/B testing is an iterative test method to verify and improve the effectiveness of an advertising campaign or website performance by testing different variations (A or B). The purpose is to retain only the variations that gives the best results. This method was initially used for mailing campaigns. However, it can be extended to advertisements inserted on other sites, banners, or even the sites themselves. The principle is always to present several variants to a large number of Internet users, and see which among those will trigger more or less page views.

In the case of web design, it will usually not be limited to a single change in the page. However, one must only change one major component of the page at a time, for example the position of the logo or banner, or the color of the titles. You must then perform the test itself: make your server give out  both versions A and B the same number of times during the same time of the same day, then analyze the results using statistic tools. Make sure your sampling test is representative. Which of the two generated the best performance. Also make sure to choose your performance measurement indices well: for two versions of a same webpage, the one which has had more access for example is not necessarily the same as the one that generated the most sales or the one on which the browsing time was longer. To refine your results, it is often necessary to make a second pass or more ...

A/B testing is a good method to know the actual performance of a change. In the case of a website and a web design, choosing the right performance measurement is key. On the other hand, the iterative nature of this method prevents its strict use on web sited (only change one thing at a time): a website consists of too many separate elements (content, images, layout, design elements, etc) to afford a unit test each time. It would take way too much time! You must therefore find significant variations and try them one at a time to use A/B testing to design a website and get a measure of the results.