Multi-variable test: 5 simple steps to improve conversion rate

Objective

Since Google appeared and changed the rules of the game, the user's attention to the Web page has been declining. For any current topic, there are thousands results to focus on, and the opportunity to capture the attention of the visitor is significantly reduced (2002, the BBC report points out about 9 seconds). Imagine yourself browsing the Web: Will you read all the text and pictures and try to get a complete picture of what the page is all about? The most likely answer is: "No." "With the barrage of information around us, we're like spoiled kids who don't devote enough attention to what a webpage is trying to tell us."

When we quickly decide whether to focus on a website, it depends on how many things we can figure out in milliseconds. Providing a good first impression is the responsibility of the designer and the site owner. The chances of convincing a visitor are very small, and most of the design (and probably you) take this as a secondary job because people think that designers are only about aesthetics. However, most sites are not designed to impress visitors, most of which exist for sale. Whether it's for a visitor to subscribe to a blog or download a beta app, the existence of each site is ultimately about selling something.

In this article, we talk about how to use a scientific approach to use a/b test and multivariable testing to create more sales, downloads, registrations (or any other business goal) for the site. Like all science-related things, this article explores and reproduces step-by-step ways to increase your conversion rate (the ratio of customers to visitors). Also, you may be interested in an article published here, the ultimate A/b Test Guide.

1. Defining challenges

How do you get site users to notice what you're offering and then let them take action? I want to answer that software download Gold page in my own personal blog. This page has all the correct elements: Product name, product description, certification, rewards, scoring, and an outstanding download link. However, only 40% of visitors downloaded the free software. Please note that almost all web traffic on this page is targeted, either through Google search or from the relevant reference site. So why are the remaining 60% visitors not going to download the software? Fixing this loophole is my challenge.

Keywords: clearly define the goals of your site (or individual pages)

As far as I'm concerned, the required operation is to have the visitor download the software, and the challenge is to increase the download rate from 40% to as high as possible. Some of the most common challenges you can use a/b test are:

Increase the registration rate, reduce the bounce rate, increase the number of newsletter subscriptions. Increase the number of leads collected from the landing page, and increase the amount of downloads for the white paper or software trial version. Optimize purchases and promotions, dramatically increasing the conversion rate from visitor to customer.

It is entirely possible that your site must be satisfied with a variety of purposes. For example, the challenge for a blog is to get more subscriptions and increase the audience's participation (depending on the number of comments). In that case, the best strategy is to solve one (clearly defined) challenge at a time.

Quick overview: A/b test. View Details

2. Suppose

The next step is to do a low conversion rate assumption list (percentage of visitors taking action), and low conversion rates are hard to produce accurate results (which is why we call them hypotheses), but there are still three good resources to help you:

(1) You: Yes, you! Although it's hard not to fall in love with your website, it's time for self-criticism. Try to follow your visitors, does your Web page provide enough to attract a visitor with no relevant knowledge? Remember, it's not like you, your visitors won't wake up in the morning and say, "Wow, this thing is awesome!" "Criticizing your site is a good way to improve it.

(2) Web Analytics data: Another resource for improved methods is your analysis tool. Specifically, the submitted data and search keywords provide valuable data. For example, many visitors come to your Web site and search for keywords that you may not notice. In this case, visitors may mistakenly think that the resources you provide are not what they are searching for and leave your site, and handling such cases can increase the conversion rate.

(3) Usability testing: Feedback obtained from usability testing always surprises you! Perhaps you will find that visitors do not even know what the Web page offers. In this case, it is a good idea to test the color and size of the action. If you do not have a large budget, you can try affordable services such as usability testing or bulk feedback.

Key point: Determine which factors affect the conversion rate.

Feedback from others does not accurately evaluate your site, and you can write down ideas that may affect the conversion. For my software downloads page, I assume the download rate is down mainly for two reasons: 1, most visitors do not notice the download link. 2, many visitors do not know that the software is free to download.

My guess is that normal access is probably like this: visitors come to this site, see a bunch of text, look around for a download link, for some reason not found (perhaps because of the color of the title No difference), and finally leave the site. Others who notice downloading links may not want to have trouble reading the text, where there is a hint to say "... It's a free ", maybe they think the software is a trial version or a demo.

You may have the following steps to assume:

Perhaps your registration form is too long, and a short form will help increase the number of registrations? Perhaps your "Free trial" button is not obvious, the large size of the download button to help increase the amount of download? Perhaps your headline contains a lot of industry abbreviations, or is it too common? Perhaps the landing page where you reach the target has no obvious next step leading to a large loss rate?

3. A/b or multivariable test?

Once the list of reasons for the low conversion rate is listed, you need to think about the reasons with a different perspective. What you need to do in this step is to think about the factors listed in the previous step in a different version. Take "registration" for example, different versions will be:

form difference: A simplified form with only two blocks, a form that does not require an e-mail address, a multi-step form, a long Form submission button difference: "Submit" or "Register for free" or "register Now" or "Sign up!" ”

If you suspect that these small differences do not have any significant effect on the conversion rate, you should read the 37signal registration 30%, which simply tests the changes in the headline news. Also, you should read the next Dustin Curtis has increased his Twitter followers by 173%, simply by changing the text of the link to "You should follow me on Twitter."

A/b Test
In A/b test (also known as a detach test), you only compare one factor on the page at a time, which may be the key to the page's impact on conversion (such as button color, size, ad copy title). In contrast, multivariate testing is a test of many factors. However, A/b test is simpler and easier to complete than a multidimensional test.

Multivariable Testing
In multivariable testing, you need to identify the different blocks/factors that affect the conversion rate in the page. These factors produce different changes that together lead to different versions of the site. Multi-variable testing is longer than A/b test, but it is more likely to produce better results.

Keywords: Creating changes

Derivative test
In addition to the problem of increasing the load on the Software page, I use my own tools to visualize the site optimizer, which provides a visual interface for the resulting changes, but you can also use other software. An obvious way to make it easier for visitors to notice the download link is to make the downloaded area the most obvious part of the page. In the design of the Web page, the title size and color of the "download" are dissolved together with the other parts of the page, causing people to not notice the download link.

For multivariable testing, I've chosen two factors on the page to make the difference: the "download" title to the sidebar and the download link to the PDF manufacturer below it. The test focuses on the effect of the word "free" and the effect of the highlighted download area. Here's what's changed after this test:

For the download header

"Download" with red "free download" in red "Download" with the default color, but larger font size

For the PDF manufacturer link

"PDF Manufacturer" with default color, but larger font size "PDF Manufacturer" with red

In multivariable testing, different changes in the synthesis result in different versions of the Web page. In this case, combined with the above changes, there is a different version of the total (4x3), each version has a "download" title and a "PDF Manufacturer" link (change 1 is controlled or the default change)

Different versions of the download area are used in multidimensional tests

Because of the definition, I've grouped two different areas of change together, so this test is called a variable test. If I'm only making changes in a single area, such as the "download" title, then this test should be called a "A/b test."

Keywords: defining test objectives

Each test has a goal to measure the effect of different versions. In this test, the goal is the number of downloads. Other types of goals may be the number of registrations, the purchase statement, the number of clicks, exposure opportunities, browsing volume, or loss rate. It is important to define the test objectives that are relevant to your business goals, for example, if an E-commerce store is to optimize its sales, it should not target the click "Add to Cart" definition, but should target the access definition of the "Thank You" page after the purchase is completed.

4. Test and analyze the results

What is a/b or multivariable test, very simple: When your Web page has visitors, randomly display a version of the page. In other words, your traffic is evenly distributed across different versions. The function of each version is to track changes to the specified target for the test. For example, in my case, the goal is to increase the number of downloads, and every time a visitor downloads the software, the visual site tracks the pages that are displayed to visitors as much as possible. Setting up a test using this tool here can help me make a selection and see what the resulting editor changes and immediately browse to the specified target on how to move on this page.

After a large number of visitors on different versions of the test to compare, to see which one is the best performance, and how much improvement.

Key point: Analysis results.

After running around 4 weeks of testing, I tested a result on my software downloads. Can you guess what kind of change is the biggest download? How much improvement? Am I able to achieve more than the existing 40% conversion rate?

Holding your breath, the result is:

# Details conversion rate improved% confidence *1 default combination (control group) 39.4%--10 "free download" red, default "PDF Producer" link 63.2%60%99%9 "Download" Large Font ", PDF Producer" Red link 56.5%43.3%98%12 "free download" Red ", PDF Producer" red link 54.2%37.7%95% ... ..... 2 "Download" for Default, "PDF Producer" large font 41.3%4.76%56%

Note:% default Perfect calculation formula is 100 * (change%-control%)/(control%)
#: Refers to the number of combinations described in the screenshot above.

Credibility *: The confidence level of the statistical results (the probability of not making mistakes).

You can observe the headline red "free download" conversion rate of 39% liters to 63%, 60% of the astonishing growth. There are "downloads" of large font sizes (with links, red combinations) that also produce positive improvements (43%) over default combinations. All the results of the first three were statistically up to 95% or more confidence levels. Seeing a steady increase in downloads means that I can safely perform these transformations on this page. Also note that even the worst performing combinations have about 4% improvement, although it is not significant.

It is worth noting that the test results may be unreliable and that the improvement may be due to opportunities. Therefore, you must understand the reliability impact of different parameters:

Browse Number: The number of visitors, the test results will be more reliable. You can estimate how many visitors your test needs by using a split-test method such as continuous computing. Conversion rate: In general, compared to Web pages with high conversion rates such as 40–50%. Low conversion (such as 1–2%) pages take longer to obtain statistically significant results. Differences in performance: a test with a significant difference in transition behavior (such as "10%") is much more reliable than a test with a very small difference (around 0.5%).

Whether you're using a tool that automatically obtains reliability results or using an online calculator to measure the credibility of the results, this is an important tool. The use and execution of unreliable results can actually result in performance degradation. A/b test Reliability analysis can be calculated by reading the article statistical analysis and A/b test, or my blog article A/b test mathematical operation.

5. Learn from test results

Regardless of whether the optimized version of the page is perceived, every attempt will bring a lot of gains. Here are some points from my attempt:

"Free" is a very eye-catching word. If you offer something free, it may be doing a suboptimal thing, so don't be too obvious on the page. Free ads are best displayed near the action link, for example, "Free download" ads are displayed around the download link. Why not set the word "free" to be clickable? The question brought to our mind an important point. I'm sure if I've analyzed the link clicks on the page, I should have noticed that a lot of users accidentally find this not a link when they click on the "Free download" title. I should test a title clickable version. Red is only a combination of other elements, such as "free" (or other text with the action), to attract the attention of visitors and let them take action. But if your words are not convincing, the visitor probably won't take any action. The size of the text that causes the action is also an influential factor. Larger fonts tell visitors that this is the part that needs special attention. For example. Download request is more important than anything else on the page.

Even if you don't remember the above points, be sure to remember a key point: do not replicate these suggestions when your site does not test them!

Each site is unique, and the goal of each conversion is different. In general, the influence of the "free" opinion on the red of the word and the size of the text that caused the action is logical. But the smartest thing to do is to create a quick test to determine its effectiveness.

A/b test has great potential to stimulate the company's revenue and profits. Still, it's strange that there aren't many people using a/b test. If you have not done A/b difference Test, why is this so? If you have previously done A/b or multivariable test, please share your experience below so that others can understand the real examples.

Original address: Smashing Magazine
Address: http://isd.tencent.com/?p=2488
Electric Trader Chinese