A Roadmap To Becoming An A/B Testing Expert

About The Author

Kevin runs an A/B testing service to help business owners improve their conversions. He also created an app to track your iPhone use called Moment. More about Kevin ↬

Email Newsletter

Weekly tips on front-end & UX.
Trusted by 200,000+ folks.

The method of pitting two versions of a landing page against each other in a battle of conversion is called A/B testing, and it lets you test two entirely different designs for a landing page or you can test small tweaks, like changes to a few words in your copy. Running A/B tests on your website can help you improve your communication with visitors and back up important design decisions with real data from real users. With the multitude of tools available, split testing has become easy for even non-technical people to design and manage.

A/B testing, also known as split testing, is the method of pitting two versions of a landing page against each other in a battle of conversion. You test to see which version does a better job of leading visitors to one of your goals, like signing up or subscribing to a newsletter. You can test two entirely different designs for a landing page or you can test small tweaks, like changes to a few words in your copy.

Running A/B tests on your website can help you improve your communication with visitors and back up important design decisions with real data from real users. With the multitude of tools available (detailed later), split testing has become easy for even non-technical people to design and manage.

ab-illustration-opt-500

When To Start Testing

Start testing only when you have enough visitors and enough conversions to run the test in a timely manner (a conversion happens when someone completes one of your goals). What does that mean?

The ideal number of visitors will vary according to your typical conversion rate. Plan on at least 1,000 visitors for each variant, and 150 conversions for each variant. For some websites this might take four hours to complete, for others an entire month.

To find out exactly how many visitors you’d need to run a test, plug a few basic metrics into Evan Miller’s sample-size calculator.

You could run a successful business with 80 visitors a month, but you wouldn’t be able to run a statistically significant A/B test. Don’t start A/B testing before you’ve done any marketing. Get a steady flow of people to your website before doing any optimization.

Keep in mind that you don’t have to build your product before starting A/B tests. You can test a splash page and find out how future customers respond to planned features and pricing tiers.

Your First Test

For your first A/B test, keep it simple. Tweak your existing landing page to get your toes wet. Focus on low-hanging fruit:

  • copy in h1 and h3 headings;
  • copy in call-to-action buttons (for example, “Try it free” versus “Sign up” versus “Get started”);
  • the color, size and position of call-to-action buttons;
  • a short page versus a long page (by hiding long sections).

You can run multiple variations at one time, so dream beyond just two tweaks. You can also run multiple A/B tests at one time, but focus on one at first to get the hang of it.

Run the tests anywhere from a couple of days to a month. Your A/B testing tool will declare the statistically significant winner. Once a winner has been declared, make the winning variant part of your permanent website by updating the underlying code.

Then, clear the way for more A/B tests.

Where To Go From Here

Low-hanging fruit is a perfect place to start, but A/B testing isn’t all about that. Sure, testing button colors and heading copy will improve your conversion rate, but think beyond how a page looks. Think outside the box:

  • Highlight features versus benefits. Are you pitching features of your product in a checklist? Try illustrating the benefits of the product by describing a dream scenario for the customer.
  • Accelerate page-loading. Keep the landing page super-simple, and make it load in under a second.
  • Show a big video with someone talking. Try removing the big screenshot and dropping in a big video that demonstrates your product, perhaps one in which you’re talking to the camera. Make a personal connection.
  • Survey new customers. Talk to new customers to see what made them sign up and what has been most valuable to them so far. Highlight these in a test.
  • Find out what confuses new customers. Ask new customers what confused them about the website or what questions they had that were left unanswered. Incorporate your answers into an A/B test.
  • Add testimonials. Use social proof in the form of testimonials. Try putting an avatar or company logo next to each name.
  • Change the writing style of headings and main content. Change the style of headings and content on your blog to see how it affects newsletter subscriptions. Try writing two versions of the same article.
  • Experiment with pricing tiers and your business model. Change your pricing structure, even if only on your public-facing page, to see how potential customers respond.
  • Make the sign-up form super-short. Remove any unnecessary steps in your sign-up form.
  • Radically change the design. Try a different approach with your landing page, like what Campaign Monitor did with its big modal for new visitors.

Remember that conversion isn’t a one-time deal. When you say that you want more registrations, what you’re really saying is that you want more lifetime customers. When you say that you want more podcast listeners, you’re really saying that you want a larger dedicated audience that won’t stop telling their friends about you.

Monitor how your A/B tests affect your long-term goals.

Tools To Use

To run A/B tests, I recommend using VWO (Visual Website Optimizer). You won’t need to edit your website’s code for each test and redeploy. Instead, you would use the tool’s WYSIWYG editor.

vwo-insights-opt-500
(View large version)

With VWO, you can do split URL testing, which tests two completely different pages, as well as multivariate testing, which tests more than two variants at a time — think of it like A/B/C/D/E/F/G testing. VWO uses statistical significance to declare the winner. A new version of the software is coming out soon, too.

Optimizely is another simple WYSIWYG tool, and Google Analytics Content Experiments is a solid free option. If you’re looking to A/B test an email campaign, use Campaign Monitor.

To set up your first test with VWO, install the code above the closing <head> tag in each page that you’re going to test.

The only other step is to define your goals. Complete this sentence: “I want more…” Perhaps you want more people to sign up for your free trial, subscribe to your newsletter, download your podcast or buy something from your store. Your goal will determine which variant in the test is the winner.

Taking It Too Far

A/B testing is not a silver bullet. Optimizing the conversion rate will make a good landing page better, but it won’t fix a product or company that has fundamental problems.

Narrowly focusing on A/B testing will turn your customers into mere data points. Your customers are not conversions that you push down a funnel. They are real people with real problems, and they are looking to your website for a solution.

Don’t sell out your visitors for short-term gain. Putting a giant red “Try it free!” button will increase conversions, but your business won’t be any better off in 12 months. Keep the long game in mind.

The goal of A/B testing isn’t to mislead potential customers into buying something they don’t want, but rather to clarify how you should communicate a product’s benefits and to make sure that customers understand the language you’re using.

As long as you’re confident that the product is great, then just use A/B testing to tweak how you present it to customers. If you know that the product is helpful, then don’t try to manipulate visitors through your funnel. Always put the product first.

Finally, don’t ignore your gut. You can use data to back up your instinct, but rely on your experience in the industry. Don’t become 100% data-driven.

Other Resources

Further Reading

Smashing Editorial (ml, al, il, mrn)