How multivariate testing works
An A/B test compares two whole pages. A multivariate test breaks one page into elements — the headline, the image, the button label, the position of the form — gives each element a set of options, and serves visitors every workable combination of them. The tool then reports which combination performed best and, more usefully, how much each individual element contributed.
That second output is the real point. The test is not only looking for a winning page; it is looking for interactions, cases where two elements only work when they appear together. A softer headline might need the longer explanation beneath it, while the direct headline works better with the short one. An A/B test cannot see that, because it never separates the parts.
The cost is traffic. Every option added to an element multiplies the number of combinations, and each combination has to be shown to enough people to be judged on its own. Traffic per combination falls quickly as the design grows.
Why multivariate testing matters
On a page with heavy, steady traffic it answers a question A/B testing answers slowly. Rather than running a queue of sequential tests over months, each waiting for the last to finish, one design measures several changes together and tells you where the effect actually sat.
It also protects you from a common misreading. When a redesigned page wins an A/B test, nobody knows which of the changes did the work, so the lesson cannot be applied elsewhere. Multivariate testing produces knowledge you can reuse on the next page.
Where multivariate testing goes wrong
Almost always, it is chosen by a site that does not have the traffic to support it. The test never reaches statistical significance, somebody reads the leading combination anyway, and a decision gets made on noise. Adding elements makes this worse, not better.
The other failures are quieter. Testing trivial things — button colours, a comma in a subheading — consumes weeks of traffic to learn nothing. Running through a seasonal peak, a public holiday or a campaign launch mixes an outside change into the result. And stopping the moment a combination looks good is the fastest way to ship a difference that was never there.
How to act on it
Be honest about volume first. If your page receives a modest number of visits and conversions are rare, use A/B testing on big, bold differences instead, and accept that a subtle multivariate design will never resolve.
If the traffic is there, keep the test small: a few elements, a couple of options each, all of them things you have a reason to believe matter. Take those reasons from evidence — recordings, form data, support questions — rather than from a list of ideas. Decide the run length and the winning rule before you start, leave it alone until then, and treat the finding as a hypothesis to confirm on another page. Running this properly is specialist work; it sits at the deep end of conversion rate optimisation rather than at the start of it.