On the face of it, conversion optimization is a simple idea and it can be really effective. You simply test different versions of a page and see which one results in the most sales, leads or whatever else you want.
But what if your website is relatively new and you don’t get a lot of traffic? Can you still use split tests effective? Sure you can, it just takes a slightly different approach.
Here’s a few ways that you can make your CRO more effective even if you don’t have much traffic.
Rule 1 – Make Big Changes
In any given split test you need a certain number of actions to reach “statistical significance” which essentially means the likelihood that the results reflect the truth and not just a fluke.
In other words:
Version A of a page might have a conversion rate of 5%, which means that on average it makes five conversions for every hundred visitors. But over the next hundred visitors it might happen to make seven conversions, and hence would appear to have a conversion rate of 7% – that’s why you need more visitors, so that these figures become more accurate and flukey events like this become less significant.
If version A and version B have very similar conversion rates then you will need a higher degree of accuracy before you can be sure that either one is better than the other. But if you version A and version B have very different conversion rates, the winner will be clearer sooner.
Testing more dramatic changes between page versions will be more likely to produce bigger differences in conversion rate, which means that you need fewer page impressions before you can announce a winner.
Rule 2 – Test Goals With Better Conversion Rates
You have a choice of things to optimize for and generally, it’s best to optimize for things that actually produce revenue – so you might want to optimize for sales, or leads generated for instance.
Unfortunately these sorts of things generally have lower conversion rates, which means that they take longer to test, compared to say, testing for bounce rate, time on page or click through.
For example:
Let’s take a simple sales page. Your main goal is to make a sale, but a secondary goal is just for the visitor to click through to any other page on your site. For sales the page converts at 2.7% and for your secondary goal it converts at 34%.
Let’s say that after 100 visitors you have made three sales and 32 secondary goals. This would give you the following apparent stats:
- Goal 1 – actual rate 2.7% – apparent rate 3% – margin of error 11.1%
- Goal 2 – actual rate 34% – apparent rate 32% – margin of error 5.89%
So you can see how the apparent or measured conversion rate for goal two would be much more accurate much sooner in the test. This is a quirk of how the statistics work.
While it’s nice to optimize for sales, you might help your bottom line more by starting with a higher value test such as testing for bounce rate:
To reduce your bounce rates, simply run a test where a click on any link on the page is considered a conversion. So if your bounce rate is 40% then your “conversion rate” is 60% – and that’s the score to beat.
A small warning:
It’s true that reducing your bounce rate won’t necessarily increase your sales, so don’t do anything that is obviously likely to hurt your sales, but so long as you use common sense you are likely to find that reducing bounces increases other types of conversions.
Oh, and if improving your bounce rate does increase your sales conversion rate, then that will make a sales based split test more feasible in the future!
Rule 3 – Reduce Variations
This is a pretty simple concept, if you are running a test on a given page of your website, and that page receives 100 visitors per month, then you will have to divide those visitors between the existing page (variation A aka the control) and your other variations.
If you have little traffic then you should only test one new variation at a time
This way you can give your new variation 50 visitors per month. Whereas if you tried to add another variation, each variation would only get 33 visitors per month – increasing the time it would take to get a valid result.
Rule 4 – Test High Traffic Pages
Similar to the point about testing for bounces instead of sales… It would be nice if you could start to optimize your checkout page or your contact form, but whilst you have very little traffic this may not be feasible.
It is likely that your home page gets more individual visits than any other page on your website, so start there (or otherwise on whichever page does get most traffic).
Testing your home page will probably be much faster than testing your checkout page, which means you will see improvements faster and that will probably mean that more traffic funnels down to those other pages – making it easier to test them later.
Rule 5 – Test Lots of Things
Of course while you should only test two variations on each page, there is no reason why you can’t run simultaneous tests on different pages…
Doing this won’t speed up either test, but it does mean that you can get more results on average over a few weeks or months, and that is clearly a good thing.
So for instance, you might test the copy and headlines on your home page with the conversion goal being to reduce bounce rate.
And at the same time test for email sign up completions on your blog post pages.
A word of caution:
This is a risky strategy and you need to be careful not to test things that may interact without considering how that will impact your results.
For example, let’s say you have an ecommerce site. If you test your home page and a category page at the same time, one variation on the home page might send a different quality of traffic to the category page, which could affect which type of layout works best on that page.
Ideally, you would want to test all four combinations:
- Home page A with category page A
- Home page A with category page B
- Home page B with category page A
- Home page B with category page B
But while you don’t have much traffic this may not be feasible. So for the time being try to test things within separate conversion funnels, as per our example with the blog post test, which are less likely to cause unexpected interactions and skew results.
You could even test for bounce rate on several different pages at once, so for instance you might test the copy and layout of your home page, about page and blog pages simultaneously.
Rule 6 – Test Every Page At Once
Your final option of course is to test every page at once within the same test. You can do this by grouping pages and testing general changes of things that appear on every page.
The classic example of course is testing your main navigation but you could also test things like:
- Header and logo
- Color palette
- Look and feel of the website
- Entire layout changes
Obviously, it’s up to you how drastic you want to go but bigger changes get bigger results as you know… (see rule 1)
Tests like this are generally less scientific, but when you get very little traffic, they can be an effective way to get started because you will be catching every visitor that your website gets for the test.
Another nice benefit is that you can track several conversion goals at once:
- Bounce rate average across the site
- Contact page views
- Add to cart clicks
- Leads generated
- Sales completed
- Etc…
And you can see how your changes impact all of these metrics and how they interact. Just so long as you remain wary of whether or not each metric is statistically significant.
Rule 7 & 8 – Start Now & Be Patient
Even if you are getting almost zero traffic, it is still worth launching a test now, because you can leave it for six months if you have to before you draw any conclusions. The sooner you start testing the sooner you start collecting data and the sooner you start learning more about your website and your visitors, and that’s not a bad thing.
Just remember to be patient, if you manage to run four or five tests in the next 12 months and you manage to get two or three good wins, then this time next year your website will be performing better than it otherwise would have been.
Remember:
Any improvements you make now will continue to pay off indefinitely!
Editor’s note: the author has since written a related post at http://www.thinktraffic.co.uk/why-your-split-test-is-going-to-take-forever-produce-mediocre-results
