Allowing developers to conduct A/B testing through their own custom interfacesGoogle is continually updating its products through small updates and features but, in order to perfect those features and decide which ones to launch, the company uses a lot of testing. Virtually all of its updates are tested with a small subset of users, sometime several variations of a feature at a time, in order to get some real-world feedback before release. But the company also offers the Google Website Optimizer tool, which allows webmasters to make the same kind of testing, and the tool just got a brand new API to extend its usefulness.
“Today at the eMetrics conference in Washington DC we announced the new Website Optimizer Experiment Management API. The API allows for the creation and management of experiments outside of the Website Optimizer interface,” Gary Kačmarčík and Erika Rice Scherpelz from the Google Website Optimizer team wrote. “If you're not familiar with Google Website Optimizer, it's a free tool for running A/B and multivariate experiments on a website. Website Optimizer handles splitting a website's traffic, serving different variations, and crunching the numbers to find statistical significance.”
The tool, part of the Google Analytics suite, is designed to allow webmasters to test out changes and new features and tweaks before committing to them. The advantages are more than obvious and having real, measurable data to gauge the effectiveness of a change can make development a much faster and less risky affair. Developers can also afford to be more experimental in their implementations knowing that if anything goes wrong it will only affect a small part of the site's users.
Until now using the tool wasn't as convenient as possible especially when doing more extensive tests as it involved switching from the Website Optimizer dashboard to the site and back. With the API tools developers can create their own controls and even integrate them with the site or with the editor they use to write the code. However, developers can't access the experiment results with the API so they'll have to view them the old-fashioned way for now.