Some websites feel like magic.You arrive at their homepage looking for something and within a few clicks you\u2019ve found what you were looking for\u2026 and a little more besides.They\u2019ve gotten you to click on a few other links, or add a few other items to your cart, almost before you knew it had happened.Then you sit down to work on your own site and you just know users aren\u2019t having that same smooth experience. But you want them to.Well, those websites didn\u2019t get that way on accident. And today we\u2019re going to dive into a practice that can make the difference.Split testing allows you to test two versions of a page, element, or site to determine which leads more visitors to achieve a goal (usually tied to an increase in revenue).Those two options — an \u201cA\u201d test and a \u201cB\u201d test — are randomly served up to visitors so that half of your visitors get one version, and half of your visitors get another. By measuring what they do after arriving on that page, you can figure out which version is most successful, and then make that your new default.For example, maybe you\u2019re trying to determine if a red button or a blue button will lead more people to buy a widget. You could run a split test, showing a red button to half your traffic and a blue button to the other half, to see which results in more sales.But tests are limited to buttons \u2014 you can also test copy, page design, images, testimonials or, in the case of a helicopter tour company, an entire site (though that\u2019s not an approach I\u2019d recommend\u2026 but more on that in a minute).Where Most People Go Wrong When Split TestingThere\u2019s a lot of information out there about split testing; after all, if there\u2019s one thing marketers like to do, it\u2019s create marketing content.But, as you may know if you\u2019ve dabbled with it in the past, a lot of the \u201cbest practices\u201d out there probably won\u2019t work for you.Why?Because most of that content was written by companies that have a million uniques a month or more\u2026 Which is actually a pretty small percentage of the sites out there on the internet.Even businesses that are making a few million dollars year often don\u2019t see that many visitors, and that means most of those best practices simply aren\u2019t relevant for most sites.So today I\u2019m going to distill down my experience from the last 9 years, during which I\u2019ve been involved in countless split tests for countless companies, and share best practices that work for \u201cthe rest of us\u201d \u2014 that is, those of us with sites that aren\u2019t Amazon or Google.What to Split Test: Make sure it\u2019s just rightA few years ago a helicopter tour company decided it was time to redesign its website. Normally, this wouldn\u2019t be a big deal. Companies decide to redesign their sites all the time, after all.The site hadn\u2019t been updated in years and they wanted something more modern \u2014 something with a bit more flash and pizzazz. They hired a web design firm and spent the better part of a year and somewhere in the range of $20-30K hammering out the details for the new site\u2019s design.Finally the big day came, and they launched the site.And, due to a mis-configuration (that is, totally on accident), half their traffic when to the old site and half their traffic was sent to the new site. Unfortunately, what they found was pretty tragic.The new site didn\u2019t convert nearly as well as the old site had. In fact, the old site was converting about 30% better.It wasn\u2019t pretty, but it worked.There\u2019s an important lesson to be learned here. Test too much, and you run the risk of investing too much time and money in a solution that ultimately doesn\u2019t bear out.But, it\u2019s important to note that the opposite is also true.Change too little and, without significant time or traffic, you won\u2019t be able to see a measurable difference based on those changes.It\u2019s a fine line \u2014 but you want to change enough so that user actions are measurably different, while also being able to iterate quickly on those tests so you can implement those that are successful and move on.For most businesses this comes down to two versions of a single page \u2014 not just an element on that page, but also not more than one page. Most companies can spin up a new page in relatively little time, and A\/B testing a page allows for enough of a difference in results that those results can actually be measured. Then, once you\u2019ve determined which version performs better, you can iterate on that page, testing smaller elements, if you\u2019d like, to see which part of the page\u2019s design is actually leading to better conversions.After all, once you discover that a change can improve your bottom line, you\u2019ll want to get it in place as soon as possible, or you\u2019re missing out (literally) on an increase in profits.What to Measure: Looking at the Results of a Split TestUltimately, when it came down to deciding whether or not the new helicopter tour company\u2019s website was a \u00a0success, it came down to looking at sales numbers.And that\u2019s important to recognize.All too often when looking at split testing marketers fall into the habit of measuring clicks or views instead of dollars. But sometimes that\u2019s misleading.For example, imagine that we have 1,000 visitors come to a sales page. Half of those visitors see version A of the page; half see version B.Of the 500 people who see version A, 300 of them click on a button at the top of the page that says \u201cLearn more.\u201dOf the 500 people who see version B, only 100 click on a button at the top of the page that says, \u201cGet your copy now.\u201dBut of the 300 who clicked on version A, only 10 people wound up actually buying the widget, while all 100 of those who clicked on version B made it all the way through the sales process.If we were just measuring clicks, version A would have seemed much more successful than version B \u2014 but when we look at dollars, it\u2019s clear that version B is the winner.What page of your website do you wish performed better? Share an idea for how you could use split testing to make it more successful in the comments.