How Netflix's A/B testing drives service’s relentless subscriber growth

Mirabelli showed how Netflix uses simple A/B comparison tests on everything—all the way down to user interface thumbnails. 

HUNTINGTON BEACH, Calif.—Netflix is a prototype for using simple comparison tests to drive viewer engagement and build audience, according to an entertainment marketing consultant who revealed some of the streaming service’s data secrets here during a Streaming Media West presentation. 

Gabriella Mirabelli is CEO of New York-based marketing consultancy Anatomy, which gives Comcast/NBCUniversal, 21st Century Fox, AMC and other media companies advice on how to drive viewership and build and sustain their brands in an increasingly digital TV ecosystem. 

“I think Netflix does an amazing job with their product, and you can kind of reverse-engineer what they’ve done with their product,” Mirabelli said, explaining in a nutshell the theme of yesterday’s panel presentation. 

Mirabelli showed how Netflix uses simple A/B comparison tests on everything—all the way down to user interface thumbnails. 

For example, Netflix will show a test audience two versions of a thumbnail for original series “Orange Is the New Black”: one full of elements, the other only featuring star Tyler Schilling and a few text elements. The latter tested much better. 

From this, Netflix learned to “keep it simple—less is more,” Mirabelli said. 

Through testing reactions of two control groups, Netflix also learned that viewers are more likely to click on a thumbnail featuring a villain. 

“Nice guys finish last,” said Mirabelli, noting that humans are instinctively prone to remember dangerous people. “You remember the bad guys.”

According to the consultant, companies like Group Nine and Upworthy that specialize in generating traffic for content over social media are now using this kind of comparison testing to maximize their audiences. 

Group Nine began doing this last year, when it noticed that the number of its videos that went “viral” had decreased from around 37% to 19%. The content company began packaging stories in two different ways, then testing the audience reaction.

Mirabelli showed how Group Nine packaged a story about a Donald Trump speech in which the embattled president criticized NFL players for kneeling during the National Anthem. One version featured just Trump; the other was a split image, with Trump appearing next to a kneeling Colin Kaepernick, the former San Francisco 49ers quarterback whose protest of shootings of African-Americans by police kick-started the protests.

The identical story packaged with the split screen captured 36% more traffic. 

In another example, she showed how a video featuring two chimpanzees was published with two different headlines, one of which indicated conflict between the two animals. The version with the bland headline generated around 1,000 clicks. The one pitting the two chimps against each other generated more than 2 million views. 

“You’re not hacking an algorithm, you’re hacking human behavior,” Mirabelli said. “Very small changes can make a very big difference.