You’ve built an amazing website, the photos are amazing, the navigation is flawless, and you have a fantastic assortment of rings, earrings, and pendants, but you just don’t seem to be making as many sales as you think you should. What could be the problem? And more importantly, how do you figure out what the problem is?

This is where A/B split testing comes into play.

We all make assumptions about what customers want. We pick colors, designs, organization, even product selection, based on what we think customers want, or what we’ve seen other websites do. The problem is that very few people actually take the time to test the assumptions we have made, we just go after more and more “traffic” on the assumption that if we can just get more people to our website we can make more sales.

That may be true, but let’s look at a few numbers. I’ll keep them simple, I promise.

Let’s assume that it costs us $1 per visitor to get them to our website. This means our cost of acquisition is $1. Let’s also assume that 1% of our visitors actually buy something (in a lot of cases this number is far lower than that – in a few, higher). This means that for every 100 people that come to the site, 1 person buys, or converts.

If you have 3,000 visitors you can make 30 sales and the cost to make those sales is $3,000. To improve your sales numbers you can continue to put money into gaining more traffic, just like everyone else, or you can improve your website to turn more visitors into customers. By increasing your conversion rate from 1% to 1.5% through split testing you can make 45 sales from that same $3,000 per month.

How Do I Spit Test?

Fortunately there a few useful tools for helping you split test.

Optimizely is a split testing service that allows you to add a simple script to the page you want to test. Using this script you can then build variations in Optimizely’s page builder and the script on your page then serves either your original page or the variation you created in the page builder to your visitors. The page visits are counted and the one that has the most goal completions is the winner.

Google Analytics also offers a simple way to experiment with page designs. You create 2 pages, your original and a variation that you want to test. you then place a script provided by Google at the top of the pages you want to test and set up the test in Google Analytics. Google then sends traffic to both pages until a winner is declared. This method has the advantage of being built into your analytics and does not require an outside server. But you are limited in that you cannot do multi-variant testing.

What Should I Split Test?

You can, and should, split test nearly everything. But I like to start with a few large conversion blockers first.

  • Test your Calls To Action, test the colors, the text, and the placement on the page. This can include the location and color of your Add To Cart button or Contact Us button.
  • Test your checkout page process, are you losing customers after they already added something to your cart because your checkout is too hard or complex?
  • Test your product titles. Gold Ring may be accurate, but Engraved Victorian Ring will likely get more action.
  • Test your product descriptions. As with titles, product descriptions can make or break a sale. You are selling emotion and sizzle, not brake calipers, so see if your product descriptions can help you with that sale.

How Long Does it Take?

I’ve done a lot of split testing over the years. In every case, more data is better. And making a decision based on not enough data can send you down the wrong path. As an absolute minimum, I like to test at least 500 sales. Not visits, sales. The closer the results are the longer you need to let the test run. I’ve run tests that required 1,000 sales or more to declare a winner.

This chart can help guide you with your split testing goals. Once your test results fall within the margin of error for a given level you feel comfortable declaring a winner.

[table id=1 /]

Every website can be improved. The trick is to make improvements based on legitimate data collection rather than just guessing and assuming that you know what your customer wants.