Today I’ve made it one of my goals to tackle using the “Google Website Optimizer” to split test something. Split testing is a way of experimenting of delivering alternative forms of the same page to assess which variant leads to more conversions. What a conversion is can vary: it could be clicking a specific link (even one that leads offsite), or the visitor landing on a thank you page for a product sale/ebook download/newsletter sign-up. The page that is responsible for registering the conversion gets the conversion script.
Choosing a test type: a/b versus multivariate
a/b versus multivariate
So there’s two types of split tests you can perform with website optimizer: A/B or multivariate. The A/B variety involves making up multiple version of the same page and saving them as a separate file. Then, by inserting a code in to the original it will sometimes redirect people to the other, alternative page. Multivariate, however, allows people to stay on the same page but instead uses javascript to switch out certain sections of the post with other variants. Multivariate certainly appears to be more powerful, and for the most part probably the only real tool I’ll need — one article I was reading suggested that they were able to effectively use multivariate exclusively for all of their purposes.
In fact, now several hours into tinkering with Google Website Optimizer I’d even suggest that the a/b split testing is a needless complication, and that multivariate can do everything you could possibly want to accomplish with a/b but better. Save yourself some trouble and stick with multivariate.
I decided a natural candidate to try this tool out would be a newsletter sign-up, in this case MailChimp autoresponders in particular since they offer a limited free trial. So I picked a popular page on my blog and decided I’d put insert points for my variations at the top and bottom of the post. At both the top and bottom I made a few different options available: no form at all, one with a persuasive call to action on the form, a LINK to a form (instead of the form itself, and perhaps a few others. In total I had about 20 different combinations — 5 different twists for the top, and 4 for the bottom. This is a bit excessive, and consequently I set it to aggressively disqualify non-performing ones to thin the stack a little sooner. As a consequence I will setup an additional experiment with fewer variations and easier to achieve goals. Twenty is too many combinations to expect results too rapidly. I will be using the information gleaned from this post to test the successfulness (read:clicks) of various social button styles in the near future. This should serve to speed up my results so I can get a grasp more quickly of the effectiveness of using this process.
Website Optimizer WordPress Plugin
Plugin adds places for code underneath wordpress posts
I was able to plug the appropriate code into the posts using a nifty wordpress plugin I found that’s specifically for adding website optimizer code to wordpress posts and pages. This plugin is very useful, but only if your split test only involves a single page or post per experiment. If your tests involve more pages and you happen to be a Thesis user, thesis openhook may accomplish the task sufficiently. (Unfortunately, I don’t know how you’d go about it with other themes.)
One issue I ran into repeatedly when trying out different code to use was formatting issues with wordpress. I kept noticing extra space between my form and surrounding content, and it turns out the WYSIWYG features was causing this in WordPress. I temporarily turned this off by visiting the users page for this administrative account, and unchecked it. The WordPress post editor is still surprisingly easy to use and effective with this feature off.
In the coming weeks or months (more likely for the newsletter split test) one combination will hopefully show itself to be the clear winner relative to the rest at getting people through that sign-up form and onto the thank you page which has been strapped with the “conversion code.” After this happened I will be able to end the experiment and run a follow up experiment, which allows me to run a new variation which I make on the spot against the winner of the last test. Additionally, data from the last experiment will remain available during the follow up experiment. This setup allows you to continually complete cycles of further refinement on your experiments.
While I didn’t think this was necessary for my current experiment… Anyone interested in google website optimizer might also benefit from these links:
Once you setup your OWN split tests you’ll have some waiting to do before the results roll in so you might as well check out this interesting book on Google Website Optimizer which I ran across on Amazon…
Expect a round 2 post with lessons learned from the results coming soon!
