One of the main problems with A/B testing is velocity. A single change takes weeks or months to test and, since only one in seven tests is actually a winner, that means most of your time is spent patiently monitoring losing ideas.
That’s not ideal.
But what if instead of waiting for a 14% chance of success, you could test all your ideas at once? Because for most testing pros, it’s not that they don’t have great ideas, it’s just that they aren’t sure which will really move the needle. Advances in AI–namely evolutionary computation–have made it possible to test dozens of ideas in thousands of possible permutations so you can surface your smart ideas and prune away your bad ones quickly and efficiently. Our AI-powered conversion rate optimization tool (CRO), Sentient Ascend, allows you to do just that.
That said? We’ve been doing A/B for so long that some people don’t think a CRO can work like this. We know it can.
Here’s how it works:
Imagine you have nine different changes you want to test. For ease and simplicity, we’ll say of each of these ideas is a different design change, like a new button color, a different header image, or a snappy new message you want to try. Representing each as a random letter, here’s a visualization of all of those changes:
Now, say you put each of these into Ascend and pressed “go.” What happens next? Ascend starts testing each individual change against the control. That means it tests the change you’ve made for “A” against the control, “B” against the control, etc., seeing whether that new button color or header image or snappy new message moves the needle.
In our image above, Ascend knows that A, D, F, H, and I all had positive impact on conversion. They have, in other words, good genetic material. They’re the ideas that Ascend will start using to evolve your site.
Which is what’s happening in the picture above. To continue our analogy, say that “A” is that new button color, “D” is that new header image, and “H” is that new snappy messaging. What Ascend does in the second generation is combine promising genes to see which ones perform. So if it’s testing “AD,” it’s looking at the new button color and the new image and seeing how they do in tandem. Then, as before, it selects the winners from that population.
Here’s where the evolution analogy really hits home. Because not only does Ascend combine those good genes from the first generation, but it actually starts “mutating” your page by re-introducing some of the less promising genes from the first generation. After all, a successful design is about combinations, the interplay between elements, not just the elements themselves. Button copy that didn’t resonate originally might be perfect with that new snappy header or the new image that’s converting well.
And, really, on a high level, that’s how it works. Ascend keeps testing and combining the best genes, trying out mutations, until it eventually arrives at a winning design.
Evolutionary algorithms are a different approach to testing. Creative still matters, of course, but because you can test more, more quickly, you don’t have to prioritize what’s next or worry about acquiescing to the HiPPO for the next test or wait around for a small change to reach statistical significance. Instead, you can plug in all your ideas and let Ascend find the best combination. This approach allows marketers and testers to simply try more stuff, increasing their shots on goal and the likelihood that they find a combination of variables that creates a massive uplift.