Machine learning could soon determine everything about creating webpages, says Unbounce

Artificial intelligence has improved beyond recognition. With breakthroughs in machine learning happening every day, more and more column inches are being devoted to the prospect that super-intelligent computers are right around the corner—and that they’ll soon take over people’s jobs.

While that future is still on the horizon, though, there are multiple ways that A.I. has already started to transform our experiences, in industries that most might not have predicted. Take, for instance, the webpages you see every day on the Internet.

Unbounce is one of the first companies to use machine learning to determine what creates the most attractive pages on the net. A business that allows organizations to easily design landing pages and test which variants perform best, Unbounce has recently created a system that asks computers to grade a site’s homepage on how likely visitors are to engage with it.

Needless to say, the results have been surprising.

“We’ve taught our machines to look at the words on a webpage, and see what messages have correlated with the highest conversion rates,” Carl Schmidt, chief technology officer and co-founder of Unbounce, tells the Straight. “Just by asking computers to look at the language that’s being used, it can make a pretty accurate judgement on how successful that webpage is going to be. That was surprising to us—that we could pick the performance of the landing pages with just the words.”

Deciding to put the A.I. technology to the test at Unbounce’s annual Call to Action conference, the company devised a game where it asked hundreds of professional marketers to rank how good a particular webpage would be at enticing visitors to interact. Asking players to predict whether a landing page had a conversion rate that was higher or lower than average, the app saw users assess nearly 18,000 webpage screenshots over two days. Pitting man against machine, the goal was to see whether humans or computers could better guess which pages would spark more engagement.

The machines won by a landslide.

While the marketers scored an average of 50 per cent—the probability of randomly guessing the answer—Unbounce’s AI-powered technology was correct 79.9 per cent of the time.

Why? Schmidt has a hunch.

“I think, anecdotally, we put a lot of weight on layout and look,” he says. “Sometimes a little bit too much. The machine doesn’t consider any of that—it’s not observing at a lot of things that we think are important. By being able to correctly predict how well a page will do just by analysing the language, the A.I. seems to suggest that the words you’re using to tell your audience what to do are key.

“We have to be a little bit careful about how we position that messaging, of course,” he continues. “Obviously in a marketing setting you actually need to be good at layout, form positioning, pictures, and all that stuff. We’re not saying that people should just focus on copywriting—clearly the visual element is important. But at the same time, our findings with the machine learning experiment highlight the importance of carefully choosing your words on a webpage.

“We’re working now on bringing some design elements into it by using machine vision, but that’s super early days,” he says. “That effort is just getting started. But we wanted to share the initial results as soon as we could because it helps folks.”

Although the research is still nascent, Schmidt hopes his team’s investigation into A.I. will help Unbounce build a product that will allow them to recommend how people should best tweak their websites. Aiming to determine a definitive causal link between a landing page’s language and its conversion rate, the company will contribute to the growing revolution of machine learning. That’s something that has been on Unbounce’s radar for years.

“It was a no-brainer to go in the A.I. direction,” Schmidt says. “Twenty years ago, there was a lot of hype around machine learning, but it really never went anywhere. We didn’t have the hardware that we have now, and we didn’t have the algorithms. Unsurprisingly, a lot has been learned in the last 20 years—and three years ago, we started seeing the momentum start up again. Machine learning began experiencing a renaissance, and we knew that if we didn’t get into it, we would be absolutely irrelevant.

“The kinds of things that machines will be able to do in the near future will be so much more sophisticated,” he continues. “Without that, people will be wondering why they’re wasting so much time doing things by hand, just to keep competitive. A Boston Consulting Group study showed that public companies—large companies—are dying much quicker in the modern tech age. They’re being displaced by new entrants using technological innovation, and their inability to harness those advances was a big factor that’s driving the higher extinction rate. I don’t want to be one of those dinosaurs. Unbounce is eight years old already, and a lot has happened in that time frame. We won’t be looking back in eight years’ time and realize that we missed the boat on A.I.”

Follow Kate Wilson on Twitter @KateWilsonSays