The biggest online video services want to ensure their subscribers always know what to watch next: 75 percent of Netflix and Hulu video consumption is attributed to video personalization. Also, 23 percent of viewers choose a service based on the quality of its search and discovery features.
Those stats came from Lakshmi Shivalingaiah and Rick Wisher of Adobe Primetime, who led an Adobe Summit 2017 session on the era of personalized TV.
Video personalization uses a variety of data to determine what shows or movies are the best fit for each viewer. It considers behavioral data, content metadata, user data, third-party data (perhaps on genres, casts, or awards received), and contextual data (such as which browser or device the viewer is using).
The speakers identified three steps for creating a successful personalization system:
- Define the audience targeted for personalization.
- Choose the right algorithm and use A/B testing to learn if efforts show increases.
- Select the right engagement metric. What’s the goal? Driving longer overall viewing time? Getting users to watch more videos?
Getting there isn’t always simple, and the speakers identified three challenges to creating a personalization system.
- Cold starts: When services don’t have enough information about their audience or their content libraries, offering personalized content is a guessing game. It’s especially a challenge for new or growing services. Solve this by creating audience segments and serving recommendations based on the limited data available, such as the viewer’s city or the device they’re using.
- Data scarcity: This is a problem with services have some data about their content or users, but don’t see clear patterns across the data sets. This occurs when services aren’t consistent in their data collection practices. Don’t try to solve this by looking at what content viewers click on and making conclusions. Just because viewers click on something doesn’t mean they enjoyed it. Instead, focus on how much of the content they watched. Session progress is a much better measure of viewer satisfaction, and that leads to solid recommendations.
- Demonstrating ROI: In this problem, someone at an organization believes personalization is a smart strategy, but has difficulty proving the value to top executives. Solve this by first choosing the right engagement metric for proving value (subscription-based services shouldn’t focus on video starts, while ad-supported services shouldn’t focus on completion rates). Once that’s selected, use A/B testing to measure the results of different personalization algorithms.