CUSTOMER & PARTNER STORIES
Decoding Discoverability: How a Global Team Came Together to Solve the Streaming Media’s Biggest Crisis
Streaming titles and services have exploded in popularity, driving search time up and customer satisfaction down. Find out how AI-innovators used next-generation tools to make entertainment search and discovery fun again.
* For premium SVOD services on costs associated with churn.
The Problem
- Open a streaming app.
- Scroll scroll scroll.
- Take a break to respond to a text.
- Play Connections.
- Scroll the streaming app again.
- Search for that one movie a friend recommended a few weeks ago.
- It’s not on the app.
- Quit and spend 2.5 hours on TikTok.
It’s a scenario that’s happened to countless streaming viewers–they sit down to watch something, are quickly overwhelmed by an array of movies and shows of all genres, then abandon their plans entirely.
Search paralysis is such a problem, in fact, that Nielsen reports the average search for something to watch lasts an average of 10.5 minutes. Viewers that do resolve to watch something are forced to choose one show or movie from the 2.7 millions titles across nearly 40,000 different channels and streaming sources available in the US alone.
As a streaming service with one of the world’s largest portfolios of titles, Cineverse needed to address their customer’s pain point of finding the perfect show or movie to watch. As a technology provider, Cineverse also needed to create a product that would integrate seamlessly into partner offerings while helping them increase customer retention. And critically, Cineverse needed to engage some of the most innovative minds across the planet to build a never before seen solution that answers a nearly impossible question… “What do I want to watch?”
The Solution
The answer is cineSearch, an AI-powered entertainment search and discovery tool. Users chat with Ava, an artificial intelligence-based video adviser who acts as both a learned film critic and the ultimate movie buddy. Ava provides recommendations based on myriad film and television dimensions including genre, mood, setting, and creative talent then directs users to the perfect piece of media across any streaming service.
“Years ago, we asked the question, ‘What if you could chat with an AI-bot about movies the way we all used to get recommendations from the clerk at your favorite video store?’ With the latest advancements in Artificial Intelligence, that’s no longer a hypothetical situation.”- Tony Huidor, CTO & COO of Cineverse.
cineSearch service was developed by Cineverse using Google Cloud’s Vertex AI Search. The underlying core cineSearch technology was built as an adaptive layer on top of the PaLM 2 Large Language Model (LLM).
But that development process did not happen in a vacuum. In order to build a first-of-its-kind film and TV recommendation tool, Cineverse had to engage world class partners to innovate faster and better than everyone else experimenting with LLMs.
The Process
With a core technology in place, and a global engineering team hard at work, Cineverse engaged industry-leading experts in every facet of cineSearch’s development process to make the tool, and Ava, a reality.
First up, global data + AI consultancy, Datatonic, a nine-time Google Cloud partner of the year that has solved complex business challenges for a diverse range of companies from clothing retailer ASOS to satellite communications company Inmarsat.
“It was clear from the beginning that an agent-based approach, combined with a complex chain of LLMs was the only way to deliver a product that provides a rich, nuanced and satisfactory user experience. But we faced a big challenge when defining the user journey. Do people start a streaming search with genres? Does it start with a previous movie they like? What search terms do we repeatedly see?” – Valentin Cojocaru, Head of Data Science at Datatonic.
Together, Cineverse and Datatonic were able to institute safety measures and Google Cloud best practices to ensure that cineSearch is safe, robust, and ethical. Additional guardrails avoid unintended replies and keep the conversations on topic. But another partner was needed to ensure Ava had a comprehensive understanding of a century of film and TV history on a scene-by-scene basis.
Enter Vionlabs, an AI technology company renowned for its expertise in emotional metadata for the entertainment industry. Vionlab’s AI was built on years of human labeled entertainment information and can now scan for factors like pace, scene length, color and mood as well as emotional impact on the audience.
“It’s extremely tricky to teach an AI storytelling. It requires an immense amount of labels and training data that you cannot get off YouTube or Wikipedia or any other place an LLM might scrape. In order for us to transform the user experience we need to understand what people are going through when they watch content and how they feel in each situation.” – Marcus Bergström, CEO of Vionlabs.
Even with the combined engineering capabilities of Cineverse, the AI-expertise of Datatonic, and the mood indexing of Vionlabs working together, cineSearch still was not complete. A final, critical partner arrived to take Ava from movie buff to movie expert.
Gracenote, the world’s premier provider of TV and movie datasets, brought a comprehensive set of entertainment information including title, synopsis, cast, director and other production information.
“Our partnership with Gracenote increases the number of films and TV shows that are discoverable by users and allows us to offer cineSearch users the highest-quality title information with intensity rankings – when paired with a user’s viewing history, streaming service filters and content preferences – will help solve a major consumer issue and the leading cause of viewer frustration,” said Huidor when the partnership was announced.
After years of late nights, early mornings, and truly global collaboration, Ava used the combined knowledge of Cineverse, Datatonic, Vionlabs, and Gracenote to deliver entertainment recommendations in a unique, articulate, and affable manner.
“It was kind of surreal,” said Cojucaru. “You spend all this time trying to create this complicated, fun and witty persona, slowly creating its personality and making sure it stays true to it. Then one day it all clicks, and you’ve managed to create something you’re proud to share with the world.”
The Results
cineSearch is currently available to use in public beta. Initial results show users receiving high quality recommendations in under a minute, a 90% reduction in search time.
Based on these early tests, Cineverse projects cineSearch could lower annual churn by 16% for premium SVOD services. This would produce at 24% monthly savings on costs associated with churn.
And, most importantly, Ava has something to recommend for everyone.
“I’m completely blown away by the massive contributions each team made to cineSearch,” said Erick Opeka, President and Chief Strategy Officer of Cineverse. “It’s unbelievable to see a blue sky idea from a few years ago become an actual, viable tool that everyone can use. And I’m especially thrilled that Ava has made it easier to pick something to watch on family movie night.”
