Editorial Curation vs. Automation: What’s the Most Successful Way to Increase Engagement?
Which is better at driving content engagement in streaming apps: a row of content selected by editors or a row selected by algorithms? It’s a trick question. In our experience, the answer to driving engagement is not picking one or the other--it’s using both! You need editorial and data science working in harmony to select and arrange the content in a given row in the way that is most likely to appeal to an individual consumer.
Successful though this approach may be, you may face a battle to make it happen. The data science team at 24i has worked with a lot of streaming services on personalization and all of them seem to go on a version of the same journey. Editorial and data science teams start off hating each other. The editorial teams think there’s no way a recommendation engine can understand the nuances of how humans categorize content. Data teams believe humans can’t possibly choose a single selection of content that will appeal equally to all users of a streaming service. And they’re both right.
There is--and always will be--an important role for content experts to curate VOD and live content libraries and present them in an attractive way for the unique audience of each streaming service. Editorial teams have a great sense for which upcoming shows will be hot. There’s also a strong commercial incentive to ensure masthead content is promoted in order to maximize the return on investment in rights. If they’re excited about Game of Thrones, there’s a temptation to assume every viewer will be too. But what if you layer data science on top of those editorial decisions?
Making the Hero Banner Work Smarter for You
Let’s start at the top. Your hero banner is the most prominent piece of on-screen real estate in each app. It’s typically a carousel of key content you want to surface to users--for example that exclusive TV series or sports franchise you’ve invested so heavily in. So, it makes sense to have your editorial team choose what appears in this location. But it can make an enormous difference to engagement if you then let algorithms refine that selection dynamically for each individual user. Let the data define which five of your editorial team’s top 10 items should be promoted to a given user, and in which order.
For instance, if a specific user has never watched a single reality TV show on your service, the algorithm can make sure you don’t promote the latest singing talent show to them as the first item they see on login. Conversely, someone that has watched related content in the past will likely be matched with the talent show promo straight away. Equally, if a user watched your flagship new drama series yesterday, the algorithm can ensure you’re promoting something else from the editorial list to them until the next episode drops in a few days’ time.
Many editorial teams also vary the hero banner by day parts such as “morning,” “daytime TV,” “early evening,” and “late evening.” This has been shown to increase engagement, but algorithms can vary the hero banner by the hour, as well as based on the typical viewing patterns for that household at that time. This is most obvious in family homes where kids' content dominates post-school and before that (hopefully) early bedtime. However, if a viewer or household does not have kids, promoting Peppa Pig at this time would be lost engagement.
This winning combination of editorial and algorithmic positioning can and should be applied throughout your apps. When we interviewed Dan Taylor-Watt, the former Director of Product for BBC iPlayer, he told us the BBC saw a 36% increase in “play completes” when they added personalization algorithms that adjusted the order of iPlayer’s “New and Trending” rail based on the user’s personal history.
Editorial and Data--A Dynamic Duo
Any row of content that is populated by humans can be optimized by making subtle, automated adjustments per user. Here’s an example. We performed A/B testing for one of our customers to demonstrate the value of personalizing curated rows. One group of visitors to the service was shown, just below the hero banner, a row of content that had been personalized in terms of order. Further down the page they saw a row that had just been curated by humans and had no external intervention to determine the order.
In the other testing group, these rows were reversed. The purely curated row was positioned higher, and the personalized row pushed down the page. The results showed a 50% higher rate of overall conversions (the number of content items played vs. the number of content items displayed) when the personalized row was displayed in the more prominent position.
Getting to Know You
So what about brand new users who haven’t had time to build up a viewing history? Or ad-supported services that don’t require a user to login? What can algorithms do in that scenario?
The key for this cohort is to show them a wide range of content to increase the chances of them finding something they like. Once your editors choose the “featured” content that best showcases the breadth of your library, let the algorithms ensure that the order of content items in a row is switched each time a user logs in, or for each and every unique page view. The curated selection of content will remain the same, but there’s less risk of the app appearing repetitive on a return visit. You’ll also increase the chance of users discovering something new to watch.
Is a Fully Dynamic App the Future?
The logical extension of this approach is to think vertically as well as horizontally. Just as the data can be used to define the order of content items within a given horizontal row, it can also be used to refine the vertical order in which rows are displayed on the page. If the data suggests an individual user loves comedy but has never shown any interest in costume drama, let your algorithms shift your entire row of stand-up shows towards the top of the page and drop the row of historical biopics down.
How far will this dynamic UI approach go? Theoretically, you could have a completely dynamic page with no two consumers seeing the same experience, much like Google search results. You could move the “Trending” rail down if the data tells you that this particular user has never clicked on an item of content in that category before.
However, I can’t see many streaming services taking their commitment to dynamic UI this far. If the whole idea of a “Trending” rail is to help users find their next favorite bingeable series, you’re going to want to keep it prominent just in case. As humans, we love a bit of familiarity. If you’ve ever relaunched the UI of a website you’ll know users can be very, very vocal in their anger at a change to the status quo. Those same users will then be equally angry two or three years later when your next update comes around, because once again it upsets their desire for the familiar. So I wouldn’t advise that any streaming service dives straight-in to a fully dynamic page layout.
Where Do You Start with this Strategy?
If you haven’t tried combining algorithms and editorial in your service yet, I’d recommend running some tests on your hero banner to begin with. I guarantee you’ll be impressed with the results.
This is the second in a series of articles for Streaming Media in which I’m breaking down five new and emerging personalization strategies that we’re seeing used to great effect by our customers and other leading streaming services. You can find an overview of the other four strategies here.
Next week, I’ll be looking at how different data science techniques and messages can be used to achieve different results for your business. If you can’t wait that long, you can check out our e-guide: Five engagement-boosting strategies every streaming service should adopt right now.
[Editor's note: This is a contributed article from 24i. Streaming Media accepts vendor bylines based solely on their value to our readers.]
Companies and Suppliers Mentioned