I still shake my head at this stuff sometimes.
You often read in the newspaper or talk with company reps about how they’re stoked because their business saw 20% growth in the previous year.
Of course, this is good stuff. However, these companies often have had hundreds or thousands of employees working their tails off to see this kind of growth.
It’s kind of funny when we, as online business owners, can tell people we grew our business by 20%. . . YESTERDAY.
No, I’m not saying this due to the fact that our business is so small and we just grew it from $100 to $120 a month.
I’m talking about how we can make simple changes to our websites and see this kind of increase in our conversion rates instantly.
One of the businesses I’m involved with (I’m a shareholder, but it’s not my company) runs some pretty decent numbers. As in 5 figures a day. I work with them very closely in a hands on way helping them to create and co-ordinate their marketing strategy.
We’re constantly testing to keep improving the conversion rates. In a recent test, we improved the conversion rate by 20.4%.
Think about that for a moment. In one simple swoop, we grew the company by 20.4%.
Now hold on for the knock-out blow. The 20.4% conversion rate improvement came about by adding a single graphic to the page in the right place.
Check out the graph from Google Website Optimizer that shows the result below.
You’ll notice that green bar means that it’s a significant improvement (and the test has finished).
You can also see the 20.4% figure (where the arrow is pointing).
Quite amazing isn’t it?
We’re often so worried about driving traffic that we forget to really target the conversion rates enough. As you can see, this isn’t an elaborate test. It’s just a simple A/B split test. However, the results are self evident. One graphic grew an already large company by 20%.
Are you testing your conversion rates enough? It’s worth it.
There is a free tool that Google offers that does the trick very nicely. It’s relatively simple to set up and the results are pretty easy to interpret.
In a nutshell, the tool allows you to put some code on your webpage where you want to test. This part of the page then becomes dynamic, showing some users one thing and other users another. You set a goal, for example a sale, and you see which content causes users to reach your goal more often.
If this is all new to you, I suggest you check out the details at Google Website Optimizer first and then come back to read the rest of this post. What I share will save you some time and allow you to get better results more quickly.
Originally I was going to give you a bit of a tutorial on how to set it up on your site, however, when I went to the Google Website Optimizer help files I found they were actually really good. So I would watch the videos on how to set up your first experiment there.
Instead, I’ll give you a bunch of things that I’ve personally found to be useful when you’re getting started with conversion optimization.
Best Practice with Google Website Optimizer
Here are a few things that I’ve learned while using Google Website Optimizer that will help reduce your own learning curve and get you better results more quickly.
1) Start very simple
The first thing I would say is to start simple. Don’t get over excited and go and try to test everything on the page. Just start with simple things like your page headlines and get used to the system. Keep in mind that the more variations you add, the longer the tests take. We’ve gone down the track of trying to test lots of variations and the tests take way too long for it to be of any use.
Remember, that the slower you get your results, the more time your site is going to be unoptimized and the more money it’s going to cost you as a result.
2) Don’t use multivariate unless you have lots of volume
In most cases, I would avoid using the multi-variate form of testing unless you have high volume of both visits and successful goals.
The idea of multi-variate is that you test two different parts of the page at the same time. The underlying assumption would be that the two factors could interact with each other, yielding a sum is greater than it’s parts type scenario. In my experience, this happens rarely and the tests take forever to complete.
Here’s what I mean.
Split Test (This can be more than 2 variations)
Let’s say you want to test 4 different hero shots on the page (so your main graphic). Google Optimizer will show each variation evenly across all your traffic. So if you receive 1000 visits a day to your site, each version will be viewed 250 times. The general rule is that the more data a test has (so the more visitors see each variation) the more confident you can be that the difference in the results can be attributed to the different image, not just a case of the normal random differences in the data (sampling error).
So in very simple terms, if it took 1000 views per image to get this kind of confidence, we could say it might take 4 days to get your results based on each image being seen 250 times a day as above. There are other factors at play, but lets ignore those for the simplification of this example.
After 4 days, you have your results. You can then run another test on say the page headline with 4 variations and you’ll have your results in another 4 days. So 8 days all up for both tests.
Now, let's say we wanted to do a multi-variate test because we believed that there might be some interaction effects between the image and the headline. In other words, a magical combination of headline and image acting together has the best results of all.
In this case, we can test both hero shot images and headlines at once. Now let’s look at how that might play out.
You have your 1000 visitors a day to your site.
You have 4 variations of images and 4 variations of headlines. So it will look like this.
Image 1 – Headline 1
Image 1 – Headline 2
Image 1 – Headline 3
Image 1 – Headline 4
Image 2 – Headline 1
Image 2 – Headline 2
Image 2 – Headline 3
Image 2 – Headline 4
Image 3 – Headline 1
Image 3 – Headline 2
Image 3 – Headline 3
Image 3 – Headline 4
Image 4 – Headline 1
Image 4 – Headline 2
Image 4 – Headline 3
Image 4 – Headline 4
So you actually have 16 variations to test instead of 8 in the 2 x split test option above.
So instead of taking 8 days for both tests, it takes twice as long, 16 days for each combination to see 1000 views.
We used to test like this until we realised that we almost never saw this interaction effect. It simply wasn’t worth taking twice as long to get the same results.
3) Don’t test small stuff
In the same way that you shouldn’t sweat the small stuff, you shouldn’t test the small stuff. Go for big impact things. Change obvious big blocks. Change headlines. Change things above the fold. The small stuff is often so small that your test would need a crazy amount of data going through it to pull any results you can rely on. In other words, any results that you could attribute to your experiment, as opposed to simply natural variation.
4) Use high volume goals for faster testing
To get tests done more quickly, especially for small sites, we use a rough and ready trick that helps to speed up our experiments. For example, we’ve got sites that make only a couple of sales a day. If we made our conversion goal to be the sale, it would take months to get any results.
So as a cheat, which yields rougher but way faster results, we’ll put our goal to be on the order page. We do this with the assumption that we are testing readiness to buy, as opposed to actually buying. We trust that readiness to buy (so actually getting to the order page) is highly correlated with actually buying.
The benefit naturally is that there will be far more people going to your order page than buying. This means you’ll see what is working on your test page much faster.
5) Have humility
This is important. Some of your “best” ideas will in fact be your worst. Deal with it and be stoked that you found out through testing rather than just running with it on your own judgement. The numbers have put me in my place more times than I care to remember.
6) Wait until the test is complete or close to completion
One thing I’ll say is to reserve your judgement until the test has lots of data. So many times I’ve thought we had a winner, and then it bombed 3 days later. For some reason, some sites see a lot of variation when experiments first get launched. Just hang tight and don’t go reporting to anyone early. Just ride it out for a while. If it runs for a long time and then it consistently steadies out, I’ve found you can finish the tests early before Optimizer tells you to.
Check out how turbulent some of the starts of our experiments have been.
You might find this test result below interesting. The fact that we prominently featured our Twitter and Facebook icons above the fold on our home page actually hurt our conversion rate by 6.24%. Think about it in this way. If we just thought that was a no-brainer and ran it, it would have cost the company hundreds of thousands of dollars. A sobering thought.
Is there anything I’ve missed that you would like to know? I’d love to hear from you to see if you would like any other info, perhaps in another post.