Imagine what Bach would do with an electric guitar... it would probably sound a lot like Van Halen.
Ninety years ago a book was being written by a man named Claude Hopkins. That book would reveal an advertising secret that revolutionized the industry. It was called Scientific Advertising and it’s been a “must read” ever since.
And, it’s even more useful today, online.
There are two components to this method, and I have a pair of acronyms to help you remember them.
ART – Always Run Tests.
3D – Data Driven Decisions.
Let’s dive in to these old concepts and see how they can drastically improve the performance of your marketing campaigns today.
Always Run Tests
What do I mean by “run tests?” I’m talking about showing two different ads at the same time, or running two versions of your website.
People tend to put something up on their website, or write an online ad and just let it sit there. Then they’ll try to improve it by making some changes and letting those sit for awhile. This is a tough way to learn what really works — you’ll never know if your sales changed because of the changes you made or the change in season.
Split testing and multivariate testing are the solution to this problem. You test multiple messages at the same time. You can do this on the web and it’s not too hard. In fact, if you’re advertising with Google Adwords or on Facebook, it’s really easy. All you have to do is make two ads.
In Facebook you’ll make two Campaigns. Keep the targeting the same and change the ad text or image. Or you can keep the ad the same but change the targeting. Either way, only make one change between the campaigns. Watch the data, and when you have a winner just retire the other campaign and make a new test.
The same goes for Adwords, but you’ll make two ads within the same Ad Group. Change something in one of the ads and the data will tell you what works best.
In online ads you might try testing:
- the wording of your headline
- your call to action
- different kinds of punctuation
- capitalizing the first letter of each word
You can run tests on your site, too. The easiest way is to use the Google Website Optimizer. You’ll make two versions of a page and the tool will give you some code to paste into each one. Then you’ll add a bit of code to the page visitors land on after conversion (e.g. the “Thank You” page or Receipt page.)
The tool will show your visitors one version or the other and track the conversion rate for each one. When a winner has been discovered you’ll be notified and it’s time to make a new test.
By Always Running Tests you’ll ensure that your online marketing is constantly improving, step by step, bit by bit. That’s what matters. Whenever you’re running tests you’re improving your return on marketing investment. If you want to always improve your ROI, you should Always Run Tests — A.R.T.
Data Driven Decisions
A test will give you empirical data and it’s up to you to make sure your decisions are based on good data. In the case of advertisement and website tests, you’ll have click rate data for your ads and Website Optimizer will track conversions on your site.
You should also have Google Analytics running.
The best data to make decisions on is full conversions: which version got more sales / donations / sign ups? If you don’t have a good way to measure that, you should get to work fixing that. In the mean time, make testing decisions on what data you do have.
It’s not just finding the right metric that counts, but gathering enough data points. This is easy to illustrate.
If I flip a coin 5 times I might get heads 3 times and tails twice. Based on that data, I could say that a coin will land on heads 60% of the time. That’d be wrong, of course. I’d never make a profit spending my time placing bets on coins landing heads up.
So, don’t go running an ad just because it made 2 more sales than another.
Statisticians have a formula for figuring out how much data is enough. You don’t need to know the equation, though. You can use a calculator or one of Google’s tools that takes care of the math for you. You can do a search for [split test calculator] or check out this one.
Aim for a confidence of 95% or better. If you aren’t getting a ton of traffic you might have to settle for less, but the higher the confidence score the more likely your data is meaningful. If you aren’t getting a good confidence score, drop the newer version and test something else. Don’t retire a winner without good data to back up that decision.
As you may have guessed by now, more traffic means better testing. The amount of data you gather is like a multiplier for this whole process.
More traffic will:
- make each improvement even more profitable
- let you discover winners faster
- allow you to run more tests at the same time
- give you an advantage over competitors who can’t test as efficiently as you
And your traffic will grow as long as you make Data Driven Decisions — 3D.
So get to testing! If it’s all too overwhelming, just start small. Run one test in your pay per click ads. Expand from there to all your campaigns. Then you can work with Google Website Optimizer to start running tests on your site.
OK, Hopkins probably didn't use an abacus. He would have had access to a mechanical calculator like this one from 1914. But, you've got the power of millions of those in your phone. From http://en.wikipedia.org/wiki/Comptometer#Model_E
It’s going to be way easier for you than it was for Claude Hopkins’ readers back in the 1920’s. They had to print up two mailings and divide up their list of addresses by hand. After paying for postage they had to count responses for each version using an abacus or some such ancient tool.
You’ve got it easy: calculators, spreadsheets, ads that can be deployed in seconds, and analytics that would make Hopkins faint.
Of course, you might prefer to hire someone to help with all this. In fact, tests are a big part of what I do for my clients with a monthly marketing plan.
Leave a comment below with your tips, experiences or questions about split testing online. What kind of results have you gotten, or what obstacles have you faced?