123 Reg logo Blog

Multivariate testing: how to prove your website changes work

By John - June 19, 2012

For many big internet companies – Amazon, say, or Google – the idea that there’s a single version of their website is laughable. That’s because these firms are constantly experimenting with different designs, layouts, colours, processes or text.

By using website analysis and optimisation tools, these companies can create multiple variants of a page and then measure which version of the page delivers the best results.

In practice, this means that you might see a different version of Amazon’s checkout screen to everyone else. Or you might get a sneak preview of a new Google feature. These sites are being constantly fine-tuned, as testing tools provide an insight into what page design sparks the best results.

A/B and multivariate testing

The simplest form of this sort of testing is called A/B testing or split testing. It’s where you compare your current version of a page with an alternative version which you think might product different results.

For instance, you might change the layout of your order form to see if doing so results in a greater proportion of people completing their orders.

A more complex option is multivariate testing. When you perform multivariate testing, you can experiment with many different website elements at once in order to determine which combination of elements performs best.

The beauty of A/B and multivariate testing lies in its accuracy. Everything you try is tested in a live environment, on real website visitors. You can measure the results exactly and be sure they’re representative. It’s not like doing a survey or asking for people’s opinions, where people can lie or their answers are open to misinterpretation.

In short, you can prove which version of a page produces the best results, with statistical accuracy.

Getting started with multivariate testing

It could be easier than you think to implement some A/B or multivariate testing on your own website. If you’re willing to pay someone to do it for you, there are lots of companies who can advise you on what to test and set tests up for you.

However, if you’d rather give it a go yourself, there are some excellent free tools built into Google Analytics (Google’s free web analytics service) that can help you. If you’re interested in trying out some A/B or multivariate testing then this is an excellent place to start – but it does mean you’ll need to be using Google Analytics to measure your website’s traffic too.

Assuming you are, you’ll find all the testing tools under the ‘Content Experiments’ option. To find this, log in to Google Analytics, click ‘Standard Reporting’, and then click the ‘Content’ menu.

There’s not enough space here to explain how to use Content Experiments, but there’s heaps of helpful stuff on Google’s pages, plus a good tutorial here.

(Please note that Content Experiments is a new service from Google which is replacing the company’s existing Website Optimizer. If you can’t see it in your Analytics settings straight away then you it should appear soon.)

Deciding what to test

What you decide to test on your website will depend on its aims. For instance, you might experiment with:

  • The size, color or position of your ‘Buy now’ buttons (e.g. are blue buttons better than green ones?)
  • The fields displayed on your registration form (e.g. with fewer fields, do more people complete the form?)
  • How you communicate special offers (e.g. is ‘save 20%’ or ‘save £5’ more effective?)
  • The length and format of product descriptions (e.g. does a list of bullets perform better than a paragraph of text?)

You can test virtually any aspect of your site, so to narrow down the possibilities, think about your website’s aims, then consider which elements might have the greatest impact.

Getting accurate results

One thing to remember when it comes to testing on your site is that it takes time to achieve meaningful results. You can’t count on the first ten, 50 or even 100 visitors to be representative of how you users behave overall.

Look for the ‘statistical confidence’ figure in your testing tool. As more people visit your website this should creep closer to the testing standard of 95% confidence, and it’ll give you a really good idea of how reliable the results are.

You also need to think about how many visitors your website receives overall. High-traffic websites can test many different variations because they have enough visitors to show them all to. But if your website only receives a few thousand visitors each month, it’ll take you much, much longer to see any reliable results.

A/B tests generally achieve high confidence scores much quicker than multivariate tests, so they’re the best way to start if your website doesn’t receive masses of traffic.  However, A/B tests usually only tells you about a specific change to a certain page. Multivariate testing can give you insight that can be used across a site.

Dangers of multivariate testing

Website testing – and multivariate testing in particular – can be strangely addictive. It’s amazing that you can come with an idea, test it, and see scientifically what’s best for your website.

But beware: a devotion to website testing could have you focusing on tiny incremental changes while your competitors make a giant leap ahead. A/B and multivariate testing can be a really valuable way to improve your website and boost your bottom line. But it’s only one weapon in your website-building armoury.

Learn more about testing on your website: