THE CHALLENGE
A Massive Catalog With Inconsistent Metadata
Plex operates one of the largest free streaming libraries in the world, with over 50,000 on-demand titles from partners including Warner Bros., MGM, Lionsgate, and Legendary Entertainment. The platform serves 25 million monthly active users across 240+ countries.
But as co-founder and Chief Product Officer Scott Olechowski explained, the deeper you go into the catalog, the worse the metadata gets.
"One of the biggest problems we have at Plex is just the sheer size of our on-demand catalog. As you get a little deeper in that catalog, the quality and the consistency of the metadata that's available to us is all over the board. It's a little bit wild west out there."
Scott Olechowski | Co-Founder & CPO, Plex
The result: thousands of titles that were essentially invisible to viewers. The recommendation algorithms couldn't surface content when the underlying metadata was too sparse or inconsistent to make accurate matches. Great content was sitting in the catalog, unwatched.
WHAT PLEX NEEDED
- Consistent mood, genre, and keyword metadata across the entire 50,000+ title catalog
- A way to surface deep catalog content that existing metadata couldn't reach
- Better inputs for their machine learning recommendation algorithms
- Preview clips for older titles that never had trailers produced
- Smarter ad break positioning for their AVOD content
THE SOLUTION
Vionlabs AI Metadata Enrichment
Vionlabs deployed its AI across Plex's entire catalog, analyzing every title at the scene level to extract rich, structured metadata that Plex's existing sources couldn't provide.
Mood & Emotion Intelligence
Vionlabs AI identifies 700+ distinct mood tags per title, categorized into 20+ mood categories. This goes far beyond genre. Two films can both be "thrillers" but feel completely different. Vionlabs captures that difference, giving Plex's recommendation engine a new dimension to work with.
"Our research has shown that people understand the mood that they're trying to achieve or know what kind of mood they're in and what content might work with that. So, we think that mood vector is really valuable and will help them make decisions faster."
Scott Olechowski | Co-Founder & CPO, Plex
Genre & Keyword Classification
Vionlabs generated 30+ predicted genres and 1,600+ descriptive keywords per title, weighted to reflect the actual story from multiple perspectives. This gave Plex the consistent foundation it needed for precise personalization, micro-taste grouping, and improved search and discovery.
Preview Clip Generation
For older movies and titles without official trailers, Vionlabs' Creative Lab automatically generates preview clips using scene-level intelligence, selecting the most compelling moments based on mood, character, and visual quality.
"Sometimes when you don't have a trailer for an older movie or a movie that just never had a trailer made for it, you need to kind of get a sense for what the movie is like. We think the preview clips will give us a chance to build a trailer-like experience for these movies that don't have a bespoke trailer."
Scott Olechowski | Co-Founder & CPO, Plex
Ad Break Optimization
As an AVOD platform, Plex's revenue depends on advertising. Vionlabs' Operations Lab identifies natural story moments for ad insertion, improving the viewing experience while protecting engagement. When ads feel like a natural pause rather than an interruption, viewers stay longer.
"A lot of our content is AVOD, advertising video on demand. Figuring out where to put the best ad breaks is something that we're always looking into. We could provide a better viewing experience, which the ads are a huge part of."
Scott Olechowski | Co-Founder & CPO, Plex
THE RESULTS
Unlocking the Hidden Catalog
With Vionlabs metadata powering its recommendations, Plex can now surface content that was previously invisible to viewers, content where the original metadata was too sparse or inconsistent to make a match.
"If we can consistently find similar content based on mood, based on genre across the whole catalog, we can start to surface content that never would have been found otherwise. The metadata was too sparse or inconsistent and we could never have made the match for that user."
Scott Olechowski | Co-Founder & CPO, Plex
What Vionlabs Delivered
- 700+ mood tags and 20+ mood categories per title, enabling mood-based discovery across the entire catalog
- 30+ predicted genres per title for consistent, reliable categorization
- 1,600+ weighted story keywords per title for deep personalization and improved search
- Automated preview clips for titles without official trailers
- Optimized ad break positioning for AVOD content
- Consistent metadata quality across the full 50,000+ title library
The integration was built to scale. Vionlabs worked with the Plex team to ensure the two products came together in a way that was efficient, cost-effective, and production-ready.
"The Vionlabs team has been absolutely tremendous through the whole process. We've been able to work together to find a way to make the two products come together in a way that's scalable. It's been really efficient, cost-effective, and it's been fun working with the team."
Scott Olechowski | Co-Founder & CPO, Plex
WHY IT MATTERS
User Experience Is the Competitive Advantage
Plex operates in one of the most competitive spaces in media. As an AVOD platform, they don't have the subscription lock-in of Netflix or Disney+. Viewers choose Plex because the experience is better, and they leave the moment it isn't.
"We are at heart a user experience company. There's so many different services out there to choose from. We spend a lot of our time investing in the user experience and really trying to separate it from just average apps, but really get in tune with what the user is trying to achieve, what they're trying to do."
Scott Olechowski | Co-Founder & CPO, Plex
Vionlabs gives Plex the content intelligence to deliver that experience at scale: better recommendations, smarter discovery, better ad placement, and visual assets that help viewers decide faster. All powered by a single AI metadata layer across the entire catalog.
