A/B Testing

A/B testing shows two versions of the same page to different visitors at the same time, so you can count which version got more people to act.

Also known as split testing A/B test bucket testing

Definition

A/B testing means showing two versions of the same page to two groups of visitors at the same time. Half of them see version A and the other half see version B, and since nothing else differs, any gap between the two groups comes from that one change. Then you count which group did the thing you wanted, whether that is buying, booking or sending the form.

The idea is older than the web. A shop has always been able to try one window display for two weeks, then swap it for another and compare the takings. Online you do not wait for that second fortnight, because both versions run together, which means the weather and the season hit them the same way.

The part people get wrong is the number of visitors. A page with two hundred visits a month cannot tell you anything about a new button, because small differences only show up in a lot of traffic. As a rough guide you want a few thousand visits per version, and at least two hundred sales or sign-ups, before you believe the result. In the first two days one version will look far ahead, but that gap is almost always noise and it fades. That is why stopping the test the moment the numbers please you is the most common mistake of all.

Change one thing at a time. If you swap the headline, the photo and the button in the same test and version B wins, you still do not know which of the three did it. So change one element, keep the winner, then start the next test, which looks slower on paper and turns out much faster in practice. Run every test over whole weeks as well, because a Tuesday visitor and a Saturday visitor rarely behave the same way.

A test answers what happened, never why, so you still have to watch real people use the page. When Linkysoft rebuilds a website we usually test the order form first, since that is where the money is lost. Inside a web application the same method works just as well on a sign-up screen or a search filter. Linkysoft also warns clients about a quiet cost, because every week you spend testing is a week you are not fixing something you already know is broken.

Questions about A/B Testing

How long should an A/B test run?
Long enough to gather a few thousand visits per version, and always in whole weeks. Two weeks is a common minimum. Stopping on day three usually picks the wrong winner.
How many visitors do I need for an A/B test?
More than most people expect. A rough floor is a few thousand visits and about two hundred sales or sign-ups for each version. Below that, the difference you see is luck.
Can I test two changes at the same time?
You can, but then you only learn that something worked, not what. Keep it to one change unless you have very heavy traffic and a tool built for that job.
Does A/B testing hurt my Google ranking?
Not if both versions show visitors and search engines the same real content, and the test ends after a while. Showing a search engine a different page than people is what causes trouble.

Still not sure how this applies to your project?

Tell us what you are building and we will answer in plain language.