Bummer! This is just a preview. You need to be signed in with a Basic account to view the entire video.
A/B Testing6:00 with Anya Mezak
The first method of quantitative testing that we’ll be discussing is A/B testing. This method compares two or more versions of a website or app to find which one performs better on a specific conversion metric.
How to know when to run an A/B test and how to get started running one
- Something is off in your analytics data
- Find out why
- Propose a potential solution
Control group -- People who see the original website design. This is used as the baseline for your experiment.
Experimental group -- People who see the design being tested.
Multivariate testing -- Technique for testing multiple concepts at the same time.
Conversion rate -- Rate at which visitors to your site get to a particular goal that you have chosen (ex. making a purchase, signing up for a service or subscribing to a newsletter)
Guideline for helping you determine the sample size. (refer to the chart in Sample Sizes section)
[MUSIC] 0:00 Usability testing is a great form of quality 0:04 research designed to get to the root of behaviors, so 0:07 you can anticipate how people would use your product and why. 0:10 However, if you want to get into the data representing how people are using your 0:14 product today, that's where quantitative testing comes in. 0:19 To kick off our quantitative method deep dive, 0:23 let's discuss A/B testing starting with an example. 0:26 Ever notice that your Facebook feed has a different design than before? 0:30 But when you mention it to your other friends on Facebook, 0:34 they don't know what you're talking about. 0:37 That's likely because you're part of an A/B test that Facebook is running. 0:40 They are evaluating if the new design you're seeing 0:44 performs better than the original. 0:47 They may want to know if you'd spend more time on Facebook or 0:50 if you're more likely to click on an ad when using this new feed design. 0:53 During an A/B test, one group of visitors to your website sees the specific version 0:58 of it, while another group sees another version. 1:02 At the end of the study, you will need to determine which of the website versions 1:06 performed better based on a hypothesis such as, 1:11 version B will receive more ad clicks than version A. 1:14 So how do you know when to run an A/B test, and how to get started running one? 1:18 First, you may notice something is off in your analytics data. 1:24 Perhaps visitors are abandoning your site or app before reaching their goals. 1:29 In our Amazon example, 1:34 this was when people were adding air conditioners to their shopping carts but 1:35 a high number of them were leaving before making a purchase. 1:40 Second, find out why. 1:44 One of the ways to do this is something we've already discussed, 1:47 a usability study. 1:51 This may lead you to discover what is throwing the users off track. 1:53 Perhaps it's the need for more product information or 1:57 maybe the checkout button is hard to find. 1:59 Third, propose a potential solution. 2:02 What can address the problem you've identified? 2:05 Perhaps it's adding more product information or 2:08 maybe it's moving the checkout button. 2:10 Propose one design solution that you can test against the current site and 2:13 an A/B experiment. 2:17 If you follow the steps we just discussed, 2:19 you may be ready to start running an A/B test. 2:22 Let's take a look at the technical steps involved and 2:26 how to create a test that can produce statistically significant results. 2:29 When we talked about testing the current site versus a potential design solution, 2:34 this was in reference to the control versus experiment. 2:39 The control group is a group of people who will see the original website design. 2:43 It is used as the baseline for your experiment. 2:48 The new design should be shown to the experimental group 2:52 to see if it outperforms the control. 2:55 If you want to have additional experimental groups, that would be 2:58 referred to as multivariate testing, but we won't be going into that in this class. 3:02 For a website or app that receives few visitors, you should plan on assigning 3:07 roughly half the visitors into the control and half into the experiment. 3:12 If you have many visitors like in the Facebook example, 3:17 you may need to sign as little as 1% or even less. 3:21 Simply because even at 1%, 3:25 you will quickly have enough volume to draw meaningful conclusions. 3:27 The design alternatives that you're testing must be limited to contain 3:32 specific isolated changes. 3:36 Examples include, the color of the critical button, 3:39 placement of a specific element, text variations, or perhaps a header image. 3:43 This is important so at the end of the study you can say, with confidence, 3:51 what exactly caused the difference in performance. 3:56 For example, was it the color, the placement, the text, or the image. 4:00 Even small, iterative changes matter, and can add up to large 4:06 changes in the total revenue or lead generation from your product. 4:10 A/B testing will compare two versions of a product in order to find which 4:15 one is better at achieving a higher conversion rate. 4:20 What is a conversion rate? 4:25 It is the rate at which visitors to your site get to a particular goal that you 4:27 have chosen. 4:31 Examples of this include, making a purchase, signing up for 4:33 a service, or subscribing to a newsletter. 4:38 How long should one of these studies run for? 4:43 Unfortunately, there's no single answer since much of it will depend on how 4:47 much traffic your site gets. 4:52 Start with how big you want your sample size to be, and 4:54 stop your study when you've reached that size. 4:57 The sample size can be determined by the size of the conversion improvement 5:00 you're looking for and the confidence you'd like to have in that answer. 5:05 I've provided a link to a guideline for 5:09 helping you determine the sample size in the teacher's notes. 5:11 If your site traffic or user behavior varies according to the day of the week, 5:15 extend your test to about two weeks. 5:20 That way, you test each day of the week an equal amount of time. 5:23 It should also help account for 5:26 unique spikes do to things like in newsletter coming out. 5:28 A holiday, the consumers payday, or people just looking around your site, but 5:32 coming back much later to complete the actual purchase. 5:37 Once you've understood the basics behind A/B testing, 5:41 implementation should be the easy part. 5:45 There are many tools out there to help you. 5:48 Just make sure you know what you want to test and 5:50 what success metrics you're going to compare. 5:53 Purchases, sign ups, subscriptions, or something else. 5:56
You need to sign up for Treehouse in order to download course files.Sign up