Over the past couple weeks, we’ve covered using heatmaps and live user testing to diagnose potential problems with your small business website and how to correct them. This week is about turning those findings into positive action by using Google Website Optimizer to test your proposed solutions.
This quick, free, and effective tool helps you set up tests that distribute new visitors to your old and new pages and track how effective each is at getting your customers to complete an action on your website. Whether that’s filling out a lead form, downloading a report, subscribing to your service, or placing an order – you’ll know which version of your website does more for getting you new business.
Creating Your Website Optimizer Account
If you don’t already have a Google Account for services like AdWords, Analytics, or Webmaster Tools, you’ll need to create one. If you do already have a Google Account, you can log in here and answer a couple quick questions to add Website Optimizer to your account. Once you get signed in and have your alternate pages ready, it’s time to create your first test, or ”experiment,” as Google calls it.
Creating an A/B or “Split” Test
There are two options for testing models: A/B and multivariate testing. A/B testing involves testing two or more different pages (even if they’re mostly identical), whereas multivariate testing allows you to test a large number of elements on a given page. Multivariate testing requires a lot of data (hits to your website) to generate results, and is a little more challenging to implement, so beginners should start with A/B testing.
Setting up your test takes three steps: providing your page URLs, installing your JavaScript tags, and then verifying proper installation. In the first step, you name your experiment and provide the URLs of your test pages and the “conversion” page, or where visitors coming through any test page need to reach to be considered a success. Two things to remember: the more variations you try to test with A/B testing, the longer your experiment will take, and all variations are measured strictly against the original and not each other, like a game of Blackjack.
In the second step, you’ll need to add some JavaScript tags to the code of each page involved in the test. These tags are what cause your test pages to rotate between new visitors and flag successes when they occur. To install these tags, Google gives you two options: install them yourself if you know how, or get a link that you can send to someone else so they can do the installation. Finally, Google will verify that your tags are properly installed and you’re ready to begin your experiment.
Once launched, there are a couple more notable options at your disposal. First, you can choose your percentage of traffic to use for testing. The default is 100%, but you can set this lower to minimize the short-term risk of testing unproven pages. The trade-off is that your testing is going to take longer to produce results. Second, you can choose to auto-disable losing variations to speed up testing. You set a threshold below which losing variations are automatically removed from the test, allowing more website traffic to go to the remaining pages. Even if you don’t select this option, you’re not stuck sending visitors to clear losers – you can still manually disable losing variations at any time. Once you’re satisfied with your experiment setup, sit back and watch the data pour in day by day.
When Should I Stop My Test?
Google Website Optimizer uses a statistical method for determining each variation page’s chance to beat the original version. You’ll be able to see each page’s conversion rate, plus or minus a range of testing variability, and how that stacks up with the original. This range is shown also as a color-coded bar in a box graph. Any overlap is grey – it tells you nothing. If there is some overlapping and some not, the part that isn’t will be colored yellow - meaning you need more data for definitive results. When there is no overlap between between a variation’s conversion rate range and the original page’s conversion rate range, the bar will either turn red (loser) or green (winner).
If you’re testing several pages, red variations should be disabled as they appear in order to allow more traffic to go to the remaining pages. Once you’ve arrived at a green bar winner, it’s time to conclude your experiment. Click on “Stop” on the experiment report, and you’ll be prompted to either choose a winning variation and continue with it exclusively or set up a follow-up experiment if you had several potential winners that you’d like to run on a more narrow test.
The hardest part of knowing when to stop a test is in the case where you aren’t getting conclusive results. If you’ve received a high volume of visits (several thousand) and your variations are still neck-and-neck with the original, you may need to re-evaluate your testing. If there’s something more worthwhile for you to test, don’t be afraid to end an experiment for being inconclusive and move on.
The Next Step
Once you get the hang of testing, you can branch out to using multivariate design instead of A/B testing to get quicker and more thorough results. For further reading, my favorite book written on Google Website Optimizer is Always Be Testing by Bryan Esienberg and John Quarto-vonTivadar. The ideas and inspiration contained therein are sure to keep you busy with finding ways to fine-tune your website. Also, be sure to check out the Google Website Optimizer Blog for the latest news and tips on how to get the most out of your testing!
Next Week’s Website Improvement Topic: Landing Page Optimization!
This was helpful