AI's Streaming Stack: Engagement Agents
The three companies covered in this issue’s column are all involved in bringing more monetization or engagement to today’s streaming content. The key differentiator is using AI to delve deep into content meaning and adjacent data to create new opportunities for content formats that previously did not exist.
While we sometimes cover tools that are suitable for smaller publishers, these AI tools require a certain scale of content to … well, work at scale. Two of these companies have been around for 10-plus years. The other one is much newer but already has an impressive connection with a media company.
Moments Lab
Moments Lab is helping media and entertainment IP owners create new revenue streams from their existing IP by identifying content within assets in their video libraries that can be repurposed to attract new audiences. The company describes this new product category as an “AI MAM.”
One customer it’s working with is Banijay Entertainment to help the Paris-based producer and distributor find micro stories within existing content. Banijay could use Moments Lab’s Discovery Agent to search for the “top 10 best fish dishes ever from MasterChef Australia, for instance,” says Moments Lab co-founder and CEO Phil Petitpont. “Our agent has a capacity to find all these fish dishes, take the top 10, and then publish them on social media to generate new audience streams.”
Using Moments Lab’s AI chatbot-based Discovery Agent for natural language content library queries (Source: Moments Lab)
Petitpont goes on to explain how this capability increases profitability or reduces costs for publishers like Banijay: “Most of the time, what our customers see is this capacity to really scale the revenue line of the business.” Banijay has told Moments Lab that repurposing a 1- or 2-minute video currently costs roughly $1,000. With help from Moments Lab, that cost drops to under $100.
“It’s a totally new way to generate a revenue stream that is very different from just reselling the rights of Season 10 of MasterChef,” Petitpont explains. “It’s like creating a new narrative angle that will then generate a new audience on social media.” This short-form content is served on YouTube, Meta, Instagram, or TikTok.
Petitpont says there are three revenue streams: mid-roll and post-roll ads, sponsorships, and directly inserting ads into the video. “Lots of marketing teams are using this tool not only to generate revenue streams, but also to promote their shows,” he notes.
Like most companies, Moments Lab picks and chooses models to deliver a multimodal solution. “We are using 20 different models,” Petitpont says. “If I’m searching for some moments about fish dishes, I can find not only all of the 2- and 3-minute sequences about each fish dish, but also a very specific step of cooking or putting it in the fridge or the oven.” These moments are identified in time-based metadata. Users can manually create clips with their clipping tool or use automated clip generation. The metadata breaks out frame-by-frame shot descriptions, complete with a full description of the visual details. Moments Lab also creates content sequences and summaries that can be used by an editor in their outside editing tool.
“We separate the production flow into two steps,” Petitpont explains. “The first is what we call the ‘upstream’ step: the original show, the 22-minute episode, the 30-minute newscast, the 52-minute episode, whatever. The second part is what we call ‘downstream,’ which is everything related to monetizing the show, whether it’s publishing the show, receiving the rights, or repurposing the show.”
While Moments Lab works with news, episodic, and reality content, Petitpont believes that reality content is especially ripe for splicing and dicing. “One of the things we observe that is very interesting for TV reality shows like Big Brother,” he says, is that publishers “want to start the downstream workflow during the show. Many things are happening at the same time. But as a director, you choose one thing at a time; you can’t cover two stories during the show. On social media, you don’t have this issue because you can choose to cover different parallel stories. You don’t have this issue of producing a linear show.”
The company’s product demo showed an easy-to-understand interface, featuring an AI agent customers can talk to. It also works by API. In one sample clip generated by AI in the demo I saw, there was administrative metadata such as title, content type, program category, season, episode, EPG summary, social media descriptions, chapter markers, and hashtags.
Moments Lab-generated administrative metadata (Source: Moments Lab)
“We are storing three kinds of metadata: visual elements, transcription, and summarization,” Petitpont says. He told me that customers can extract this metadata for other uses as well. “One important thing we’ve worked on is to avoid any vendor lock-in on the metadata. When a customer is indexing content with us into our platform, this metadata is globally available. You don’t need to pay to access the metadata to use it in other systems.”
Moments Lab stores the following metadata types:
- Text-based metadata for dates, rights, or anything that needs to be stored in text, like object data
- Vector/visual embeddings or text embeddings that make the output really semantic
Moments Lab was the only company I spoke to for this article that provided a demo, and it delivered a very professional-looking one. The company has been in business for 10 years and prices its services either by the hour or by the show. It will export metadata for customers in whatever format is required.
Versos AI
For the past couple of years, AI labs have been actively sourcing archival video from studios, streamers, and distributors to acquire the rights to train on video content. Curiosity Stream has been one of the leaders in licensing content for AI training, and it is doing this with help from Versos AI.
Video content licensor CuriosityStream is an early adopter of Versos AI’s AI Video Library Intelligence Platform (Source: Versos AI)
Versos AI’s AI Video Library Intelligence Platform is a tool that takes video content and prepares content libraries for AI training. By definition, video content consists of unstructured datasets. “At best, a library has a filename and maybe some other file-level metadata attached, but very little else,” says Versos AI CEO and co-founder Chris Keevill. “We look inside the video and describe scenes and frames as well as objects within frames. We can create almost the full, near-infinite understanding of what’s happening inside a video. That video understanding helps large-library owners manage, understand, find, discover, and retrieve things out of this unstructured data library.”
As recently as a year ago, for the most part, “labs were satisfied if a rightsholder like Curiosity Stream sent them 10,000 or 20,000 hours of nature and wildlife documentary assets,” says Keevill. “Increasingly, the model builders at these AI labs are looking for something more specific within that category because they’re training their models in more specific ways.”
Finished content is more desirable, Keevill explains, and the model trainers in many cases are looking for a narrative arc. For example, trainers want to improve their understanding of character development. “They’re going to want to look at full scenes and character development throughout a series. That’s one use case, but other, more specific use cases are where the models are looking to understand the physics of the world. They want to understand gravity,” Keevill says. “They ask for 10,000 30-second scenes of objects falling through the air and landing on the ground to understand gravity.”
Throughout this process, the video stays in the rightsholder’s storage. Versos AI can integrate and process from all the major cloud platforms, such as Azure, Google, and Amazon. “We process it,” says Keevill, “but it stays put.”
The company also emphasizes C2PA compliance. “Within Versos, we create a C2PA manifest,” Keevill attests. “Think of it as a stamp on a video asset that establishes effectively who owns it, where it has come from through the supply chain, and then where it is going from our platform next.”
Versos AI is primarily working with larger library owners with 10,000 hours of content. It also works with aggregators in the market that will do business with a small studio that has closer to 500 hours of content. Versos AI is a multimodal platform and uses a range of open source models. Data is delivered as a JSON file. The company does not have public pricing.
The Versos AI platform launched at the end of February 2026. Curiosity Stream is not only a customer, but has also invested in the company.
Sportradar
Have you ever wondered about all of the data that sporting events are now serving to consumers in a world of more and more personalized viewing experiences? The ability to take and analyze data and be able to provide context to what’s happening in the event is the business and technology of Sportradar.
Today’s sports broadcasts, to a large degree, are no longer linear-only. “Media companies have had to shift to creating this whole content ecosystem and trying to match the content, depending upon the platform, to the type of viewer and where they are,” says Mark Holland, SVP of media products at Sportradar. “People are viewing sporting events across every platform that they’re on. What we’re seeing is that there’s a lot more interactivity and desire for personalization.”
Holland envisions enormous potential for personalization in the content that sports programmers are capturing for every event, even if most of it goes unseen in a linear broadcast. “We’ve got amazing, insightful content for the home team, for the away team, for every single player on the team,” he says. “You do see a lot more of it in streaming. Connected TVs are starting to move into enabling different layers on the screen, and we’re seeing even different broadcasts.”
With all of this content also comes a wealth of data, much of it of great interest to many fans, bettors in particular. “We collect sports statistics, information, and data and provide that to thousands of clients across the globe in a variety of different ways, from data feeds to visualizations, to content products, integrity services, and pretty much everything in between,” Holland says. “We really sit at that intersection of media betting and technology and enabling the storytelling of sports in the world.”
Holland explains Sportradar’s service as follows: “Our core business is providing the sports data to our media clients, and we do that through [offering] 50 or 60 different sports APIs. That’s the raw building blocks for our clients to build any sports experience they ultimately want, be it a match tracker, some sort of a ticker, or graphics within a broadcast.”
Sportradar’s offerings include a number of products. SR On-Air pulls real-time statistical updates and play-by-play data from Sportradar databases to populate broadcast graphics systems like Vizrt and Chyron. Ten years ago, Sportradar developed the analytics tool Radar360 to provide deep research, which is used during games as well as in pre- and post-game. Holland says, “During a live broadcast, our research team is asked questions like, ‘When’s the last time that this happened?’ or, ‘Has this ever happened before?’ or, ‘What’s the importance of this particular event in a sport?’ These answers can really enhance that storytelling in the course of a live broadcast.”
As recently as 6 or 7 years ago, Holland recalls, “broadcasters would have to set up very specific templates, and if something occurred during a match or a live event that deviated from them,” they had to describe it verbally, on air. “It wasn’t something they could show graphically.” Now these graphics templates can be manipulated in real time to capture these stories during a live broadcast using tools like those available from Sportradar.
“We have a product called 4Sight Streaming (built more recently for broadcasts, particularly the NBA), which essentially overlays statistics and probabilities over the live broadcast,” Holland says. “As you’re watching, you can see the probability of a particular player’s shot as they have the ball. Then, as they pass, those probabilities change. It really allows not only the broadcasters to start to tell more interesting stories, but a viewer to be able to turn that on and just have that much more immersive experience.”
Custom interactive NBA broadcasts with Sportradar 4Sight (Source: Sportradar)
How has AI changed the data you get and the data you can pass? Holland believes “the biggest thing around AI is the ability to essentially analyze the depth of data that we’ve got in real time and identify different anomalies, come up with different insights, look for patterns, look for milestones, all in real time very, very quickly and across a very wide and deep dataset and across a lot of different elements. Not just your top players, but every single player that might be on the field or on the pitch at any given moment in time.”
Sportradar GameFrame (Source: Sportradar)
Another Sportradar product, GameFrame (currently available only for the NBA), “takes a standard shot chart, which used to just be 2D Xs and Os,” says Holland. “This is a 3D animation with player likenesses, and you can see their various shots and the movements that they’re taking in order to make that shot.”
Related Articles
This is a tale of a few different AI use cases that involve research, localization, advertising, and UX. The first is a public broadcaster in Austria. The second is a TV OS. The third is a well-known vendor. The last is a major media company. What all of these have in common is that their AI applications have moved from the proof-of-concept stage to full commercial implementation.
26 Mar 2026
As with all streaming workflows, AI has steadily crept into the live streaming technology stack. In some cases, the impact is incremental, in others, profound. From production to monetization, here's a quick overview of where AI has become relevant for live event producers and engineers, and some areas where, surprisingly, it hasn't.
26 Mar 2026
How has AI entered the media workflow? For this new column, we'll look at different applications used in the media industry. For this issue, we'll start with asset management, asset storefronts, and localization. While some of this functionality—speech-to-text transcription, translation, voice synthesis, natural language processing, logo detection, facial recognition, and object detection—has been around for a while, the biggest improvement is that much of it is now available on workflows with live content.
15 Dec 2025
In a candid interview at Streaming Media 2025, EZDRM COO and co-founder Olga Kornienko and Future Frames producer Doug Daulton trade helpful analogies in discussing the growing importance of content provenance in the age of generative AI and deepfakes, as well as the role that the C2PA standard and DRM providers like EZDRM can play in protecting content authenticity and content creators' rights in a changing media landscape.
24 Oct 2025