Unbounce has been helping marketing teams develop and generate better performance from landing pages since its inception, but now the firm is using comparative data to show customers how machine learning could improve their results even further.
In a series of event appearances starting last Fall, Vancouver-based Unbounce has been staging “Machine vs. Marketer,” where it challenges business professionals to pick what they feel is the most effective landing page from a section of paired options. The choices made by humans are put up next to those run by a machine algorithm that has analyzed more than 60,000 English language pages for lead gen purposes that Unbounce has gathered in its database.
Guess who tends to win?
“Once we get enough data, turns out machines can outperform us,” Carl Schmidt, co-founder and CTO at Unbounce, told B2B News Network. “We see the Machine vs. Marketer game as part of this ongoing conversation (about machine learning). By introducing it as a game, we make it fun and non-threatening, where it’s not about machines taking your job.”
Beyond the game, Unbounce has developed a Landing Page Analyzer tool based on the data. It has also started to provide some benchmark information to marketers about what works on landing pages and what doesn’t. In sectors like health-care and legal, for instance, fear-based words like “risk” and “difficult” proved highly effective in terms of conversions. As you might expect, pages that were 500 words or fewer tend to work better, as do those where language is kept simple. Six out of the ten industries the study examined showed better conversion rates when the reading level was appropriate for a 9th-grade reading level or lower.
According to Schmidt, Unbounce is still working on more formally integrating machine learning into its core offerings. The end result has to be more than an added feature, he said, but almost a way of offering a consultative capability based on AI.
“We want to be able to look at that page and say wow, you’re in the B2B SaaS segment, so these are the general conversion rates, this should be the specific text on your download button, and this should be the text in your headline,” he said. “We see it as feasible in the next year or so.”
Over time, Schmidt suggested it’s possible that natural language processing (NLP) tools would actually be able to assist marketers with writing, but there are still some psychological issues to overcome.
“It’s going to be a long process of acclimatization, honestly,” he said. “Machines aren’t going to be doing a lot of the more creative aspects. If we’re positioning it as more of an assistance to marketers, that certainly helps. We just find as we’ve been running more concept tests, the bigger hurdles is that folks believe they simply know better.”
Latest posts by Shane Schick (see all)
- FPX research shows 58% of B2B firms pursuing digital transformation want to improve buying processes - January 16, 2019
- Ethics guidelines from INFORMS focus on analytics and operations research - January 15, 2019
- CES 2019 experts identify enterprise XR opportunities based on virtual and augmented reality - January 11, 2019