Multivariate testing

Google
Website Optimizer

Much of my job involves supporting a marketing department in web stuff. A lot of the time this means finding and explaining new ways to help them market. I've recently been looking at multivariate testing, which is a big word to mean testing multiple variations of a web page to see how effective they are.

To play with it, I've been using Google Website Optimizer on my own site. If you go to my home page, you'll be presented with one of four versions of the page. I count someone clicking on a blog post as a "conversion", which is kinda contrived and you would normally use something like a sale.

In case you're interested, the variants are:

It's all pretty contrived and I'm using a site that doesn't really matter, but it's shown me how useful this tool is. If your site has some kind of goal (sale, enquiry, some kind of interaction you want to encourage) it's really incredible the effect some changes can have.

If you're marketing to mere mortals rather than geeks, seemingly meaningless changes can have a huge impact. Making the buy button bigger and brighter, putting in little bullet points to overcome their objections, changing the location of things.

These multivariate testing tools allow you to test a bunch of variations on a page. They could be radical changes, or really simple ones. Different users will get different versions, and you can quite happily track what's happened. The tools contain all sorts of statistical crunching to allow you to see real patterns in the noise, once enough data has been collected.

The example I'm using is pretty simplistic too. I've only changed the front page, not the whole site. These tools allow you to track users throughout a visit, so you could make a site-wide change and be able to give the user a consistent experience throughout the visit, then log conversions.

The great thing about this kind of tool is that, when the marketroids come up with their latest "brilliant" idea, rather than having to shoot it down with your mere logic, you can just say "sure, let's test that out". I've always thought of marketing as the place where failed salespeople go, because a salesperson is very easily measured and marketing is traditionally quite hard to measure. Not any more!

So my task in the near future is to shop this idea around our marketing geniuses and start playing. I think it'll be actually quite fun, because you can throw seemingly wild ideas at it, and see how it goes with real people. The testable nature is actually quite liberating, and should free you to try out your wacky ideas.

I'm quite looking forward to unleashing it on a really busy site, and taking a crack at long lists of changes to try.