Netflix (NASDAQ: NFLX) wants viewers to quit browsing and start watching its massive catalog, and it's juicing up its content recommendation tech in order to convince them to do so. The first stop on its journey: figuring out which image associated with the title of a movie or TV series will convince a user to click into the content and watch it.
While changing the artwork on its titles sounds like a fast solution, according to a post on Netflix's tech blog, it was anything but. Approximately 82 percent of a subscriber's focus is on the title art while browsing for new content, but that subscriber considers a piece of content for only 1.8 seconds before moving on, the SVOD provider discovered in a 2014 study.
Gopal Krishnan, who coordinates product innovation at Netflix, said in a blog post that the provider built a system to conduct a "groundbreaking series of A/B tests" which increased user engagement on the site. With a title's artwork as the deciding factor, the study had three initial goals -- identifying artwork that helped users find content they wanted to watch faster; getting members to engage more with each title and watch more content in general; and making sure the artwork didn't misrepresent the title.
Netflix teams ran a series of tests with members, such as one in which slightly different artwork was shown with its related content to users in different "test cells." Figuring out which art best related to the title help increase take rates dramatically – for example, The Short Game, a documentary about grade school kids competing with each other in golf, saw as much as a 14 percent improved take rate over the film's default artwork when a different image was used, Krishnan explained in the post.
Nick Nelson, global manager of creative services for Netflix, said the provider has been working for years "to create a framework that allows us to effectively intersect big data with creative, ultimately helping members discover stories they will enjoy even faster. As a result of that work, we now have the unique ability to understand how to most effectively tell our members why a story is right for them -- all through a single image."
The testing presented some engineering challenges, however: Krishnan said Netflix had to invest in two major data areas: client side impression tracking – figuring out how many times a particular title came into a user's "viewport" – and stable, unique identifiers for each piece of artwork. The teams also had to build incrementally toward increasingly sophisticated and rigorous tests, and "fail fast" in order to move the process forward.
The technical blog post gives an interesting view into the workings behind Netflix's content recommendation and discovery efforts. While still in the works, the result will hopefully be better than Mashable's proposed solution, Netflix Settle.
Viewers find best value in Netflix, HBO Now, Hulu subscriptions, study says
Innovations in content discovery, quality key to improving users' streaming media experience
Consumer frustration just one factor leading to video content piracy
Netflix trims the size of its streams by 20%, cuts bandwidth use and reliance on cable