Three ways to sell more software (or anything else). They’re all different flavours of the same idea, but with different degrees of complexity.
B-A test: the basic-but-better-than-nothing option.
Take an important page that gets a healthy volume of traffic.
Look at the last 7, 14, 21 or 28 days of data (see the points below) and note the basic page performance indicators. Eg: bounce rates, exit rates, time spent etc.
Create a new version of the page – but make it noticeably different.
Changing “Download our trial!” to “Download our trial!!” won’t blow your visitors or your results away.
Let the new page work for 7, 14, 21 or 28 days and compare before with after. Hence the B-A name.
One version of the page will almost certainly out-perform the other.
Use it and test it further.
Split test: the intermediate-and-feeling-good option.
Take an important page that gets a healthy volume of traffic.
Either create a script that divides traffic to one or more variations, or use a service (see below) that makes life easier.
Compare the various versions.
One version of the page will almost certainly out-perform the other.
Use it and test it further.
Multivariate test: the expert-and-don’t-I-know-it option.
The same as above, but instead of pages you create variations of individual content. Eg: different headlines, texts, buttons, offers etc.
These are displayed in conjunction with each other.
It’s more complicated to understand and easier to misinterpret, but in the right hands can produce better results.
Points that apply to all of the above.
– Only work in full units of 7 days. Every website demonstrates a clear seven day trend, so don’t contaminate the data by looking at mixed samples.
– Only compare like-with-like data as far as possible. A write-up on Hacker News will boost your traffic but pollute your data.
– You might be able to write your own system for the tests, but why reinvent the wheel? Services like Visual Website Optimizer, Optimizely or even Google Website Optimizer are easy to setup, use and understand.
– Anything is better than nothing. We’ve run hundreds of split and multivariate tests for our clients, and if I remember correctly I’ve only ever seen two experiments that didn’t result in an improvement.
Do it today – you can be up and running in less than 20 minutes.
