Read on to learn about A/B testing, including examples of tests you can run on your own ecommerce store.
What is A/B testing?
A/B testing, or split testing, is the practice of serving different versions of a web page, ad, or email to an audience to see what performs better. It is the most common testing method for online stores.
In a proper A/B test, the two (or more) versions are served over the same timeframe to randomly selected members of the target audience, as opposed to a before/after test in which the versions are shown sequentially. Similarly, in an A/B test, only one variable is tested, even if there are multiple versions. For example, testing three different button texts would be an A/B test. But testing multiple variables, like button texts and banner images, would be considered a multivariate test.
A/B testing requires code to serve the different versions simultaneously to different users. In advertising A/B tests, Meta and Google do this automatically. Similarly, some email platforms like Klaviyo have built-in A/B testing functionality. Website A/B testing requires custom or third-party code, such as Google Optimize.
Benefits of A/B testing
All digital marketing efforts create data. Advertising campaigns provide data on click-through rates, websites on conversion rates, and so on. The main benefit of A/B testing is being able to gather the data you need to make specific decisions.
A/B tests always start with a specific hypothesis. For example, “I think our sign-up form will convert better if we offer a 10% discount.” The data allows the business to make a conclusion about the hypothesis. This helps it answer the immediate hypothesis, but also gives it more information about the business as a whole. In this example, it helps it understand how price-sensitive its customer is.
Although A/B testing is primarily a marketing activity, the insights it yields can be used for a variety of other business decisions, including UX, product development, branding, and sales.
7 A/B testing examples for ecommerce businesses
There are many types of A/B tests an online store can run. Below are some of the most common, impactful types of tests.
Header copy
This refers to the header at the top of a page, typically a landing page. Since this is the first, largest text a visitor sees, it’s a great way to test the most valuable first impression for your site.
For example, Gymshark might test different versions of the line “Power. Made to fail in,” below.
Subject line
In email marketing, your subject line is your most important lever. It drives an email’s open rate, and if your audience doesn’t open your email, the rest of the email won’t have any impact. It’s a great way to test what catches your existing audience’s eyes and improve email performance.
For example, DUER might test another variant of the email subject line below, “Introducing: The Premium Dura Soft Midweight Tee.”
Ad tagline
Ad platforms allow for quick and easy A/B testing of lots of different copy variations. This creates a self-reinforcing loop—insights about ad taglines help inform future taglines and future versions to test.
For example, in this ad, BN3TH could test either the primary text (“Numb crotch? No thanks.”) or headline (“Ride Longer & Comfier, Save 25%🚵♂️”).
Call-to-action text
Call-to-action text can be tested on a website, ad, or email. Great calls to action help your audience finish the sentence, “I want to … ,” so testing your CTA can help you understand the intent of the user on your page.
For example, this exit intent pop-up form from Vahdam could test different copy on the button to see what gets more form submissions.
Product image type
Testing your product image type can help you understand the drivers of your product’s conversions. Some products are more practical, and benefit from simple images that highlight features, whereas others are more lifestyle-driven and benefit from first showing products in the context of use.
For example, Blender Bottle could experiment with a landing page that shows a lifestyle photo, such as a person in professional attire leaving the gym on their way to the office, before showing the straightforward bottle photos with features.
Pricing and discounts
Pricing can be difficult to test from a technical and customer perspective. Most website A/B testing tools don’t offer the ability to test prices. And these tests run the risk of annoying customers who purchase it at the higher-tested price and then learn someone else purchased it for less. However, Shopify apps like Intelligems do allow for price testing.
Alternatively, testing discount codes can be an effective way to get similar learnings. For example, a brand could roll out two marketing campaigns targeting the same audience, with the same ads, but a different discount offer, such as 25% versus $25 off. They could see which performs better based on both click-through rate and conversion rate.
Element removal
An A/B test can be addition by subtraction. If a website has lots of different options for shopping or navigation, marketers will sometimes test hiding an option and seeing the effect on conversion.
For example, LOLA might test the effect on purchase conversion rate of removing links to its blog, The Spot, from its navigation:
How to conduct an A/B test
- Form a hypothesis
- Create test variations
- Select an audience
- Run the test
- Analyze the results
Running A/B tests is a systematic process. Each test follows these five steps:
1. Form a hypothesis
A good A/B test starts with a theory about how to improve performance. This theory can be based on existing data or it can be based on your opinion. To turn your theory into a hypothesis, state it in the form “I believe (making X change) will lead to improved (performance in Y metric).”
For example, “I believe enlarging the main product image on our product landing pages will lead to improved conversion rate.”
Hypotheses don’t need to specify the amount of improvement—they only need to be stated directionally.
2. Create test variations
This can be done in an ads manager, email platform, or website A/B testing tool. Variations should be labeled descriptively for easy analysis later. For example, instead of titling a new ad variant “Variant B,” title it “Variant B—Emotional CTA.”
3. Select an audience
A/B tests can be served to an entire audience or to a subset of your audience. On a website, for example, serving to the entire audience would mean half your website visitors see the original site and half see the new version you’re testing. However, you might choose to only serve the new version to 25% of your audience, or only target visitors from Canada (in which case, half of Canadian visitors would see the original and half would see the new variant).
The right audience for you depends on what group you believe your hypothesis applies to and how quickly you’d like to gather enough data to make a conclusion.
4. Run the test
Typically, marketers will run any A/B tests for at least two weeks to ensure a test’s success. This allows for enough time to account for any coincidences or fluctuations, such as customers acting differently on a weekend. Make sure not to run multiple A/B tests on the same webpage or audience at the same time or you could cloud your test results.
5. Analyze the results
When analyzing an A/B test, you are looking for a statistically significant result. This is a data formula that indicates that the result you’re seeing is reliable and not the result of a small sample size or coincidence. The concept behind it is that if you have a small audience, you need to see a big difference in performance to make a conclusion. But if you have a large audience, even a small difference in performance can be conclusive.
Tools like Google Optimize will calculate statistical significance for you. For other results, you can use a statistical significance calculator.
Once you’ve analyzed the results and shared them with your team, you’re ready to prepare your next test.
A/B testing examples FAQ
Why is A/B testing important for ecommerce businesses?
How long does an A/B test typically last?
What are some best practices for conducting A/B testing?
Create a clear hypothesis
Test one element at a time
Make sure your results are statistically significant
