Amagi Media Labs Limited (AMAGI) Earnings Call Transcript & Summary

Apologies for some of these glitches. I'll just open again. Thank you, everyone, for joining us today. This is the first session in what we hope will become an ongoing educational webinar series for investors where we periodically discuss industry relevant themes and developments. Today's discussion is strictly focused on AI and media. So we request all participants to keep their questions limited to this topic. The session will run for 1 hour. Baskar will speak for the first 40 to 45 minutes, followed by Q&A in the last 10 minutes. [Operator Instructions] In the end, you are invited to share feedback through a Google form by scanning a QR code. Please take a moment to share your inputs and help us improve future sessions. Before we begin, please note the disclaimer. Questions relating to the company's financial performance, business outlook, operational metrics, strategy or other company-specific matters will not be addressed. Please note that certain comments may include forward-looking statements, industry observations or estimates based on management's current views and assumptions. Actual outcomes may differ materially. With that, I'll turn it over to Baskar to kick us off.

Yes. Thank you very much, Amoolya. Thank you, folks, for joining us. Good morning, good evening, good afternoon, wherever you are. So today's topic is an interesting and exciting topic, AI and media. I'm sure AI is the topic for all of us across many different industries that we're really working on today. And media is no different, and we're going to talk about it today, right? So if you look at it, the first thing I want to talk about is to really understand how a typical video value chain works, right? As you know, television and all the OTT video streaming that we see, for example, I want to start off with an example of how that actually works. If you look at it technically, if you take any sort of a sports match or a news or entertainment content, it starts from the camera and ends with a screen where we watch content. We watch content either on television sets or on our phones or on our tablets, wherever you're watching, that's where the content comes through. But it always starts with the camera today as we speak, right? So there are a set of sections of how the process works today, as an example. Let's look at it, for example, a game, right? So if you're watching a cricket match, for example, the action obviously starts in the stadium. At the stadium, the first thing that you see is that you have multiple cameras. All these cameras gets connected to a van called an OB Van, which is nothing but a van standing outside of a stadium typically with a satellite antenna on top through which it gets beamed to a studio. We've all seen it. I'm sure any of you have gone to a stadium, you would have seen this as a mechanism that you see. Obviously, there is commentary boxes from that commentary is getting done. And also, there is some sort of a production room where they're trying to look at what cameras -- camera content makes sense and they do the switching of the cameras. Now once that is done, it comes to a preparation stage where it comes to typically a studio. In studio, this cricket match gets converted with sports scorecards all the logos of the different teams, multiple audio tracks that you want. And of course, the advertising and ad breaks, all of that gets added in the preparation stage in a typical studio environment. Once that is done, it gets distributed either through cable or through streaming TV platforms that they are aware of through which we actually watch this content. This is typically a stage where, obviously, monetization could be advertising, could be as consumers subscribing to content. You kind of do multiple different monetization capabilities today. And obviously, viewership could be across the globe that people are starting to watch content. This sort of sustain. So starts from a camera, producing content, preparing the content, distributing the content and eventually getting to viewers, for example. Now if you look at it, I just want to give you an analogy of what this works in the world. It's also like a retail analogy, right? So I'm sure a lot of you have seen the whole retail workflows and chain. For example, like retail, there is a production phase where I'm sure that the consumer goods, you produce your goods, you take it to a warehouse and do all the packaging, then you send it through trucks and any sort of logistics to get to retailers. And eventually, consumers buy it either through directly or through the warehouse of the retailers through a quick commerce or an e-commerce or the mechanism that you see it. Very similar is what's happening in the content business as well. If I'm a content creator, the first thing I do is to really produce content, exactly like a factory, there is a content factory where hundreds of hours or thousands of hours of content gets created, right? So that's the first part of the value chain. Once it's created, it goes through a preparation stage where like a warehouse, you're actually preparing this whole media stuff and eventually using some sort of logistics, you're sending it to the platforms. And eventually, these platforms eventually deliver to the viewers, right? So exactly like a value chain of a retail, think about it in that particular analogy that I want to really talk about from a media distribution flow standpoint as well, right? Now if you look at it, AI is transforming every industry. So like I'm sure you folks are hearing it every day. Media industry is already changing. And it's found no big difference. It's exactly like every other industry, AI is transforming this industry as well. Now look at it, what we're going to do today is to kind of understand this transformation that's happening across 2 spectrum of things that we're going to talk about today, right? One is if I'm a content creator, a business of either building TV channels or OTT operations or any sort of studio or a content creator today, for example, I'm looking at either operating leverage through AI, and we talk about agentic and workflows today, for example, or I'm looking at market expansions where can AI really accelerate my expansion, be it in terms of reach and audiences. These are the 2 real levers if you look at it, that we really talk about when we talk about what's the opportunities in front of a content creator to leverage AI. The first step, if you look at it is from an operating leverage standpoint, agentic workflows. And we're going to talk about what does it mean and how operational leverage actually comes to the whole system. Now if you go back to the world of really factory production of consumer goods. What is the biggest change in the retail industry that's starting to happen is the whole new evolution of physical AI. We are talking about robots. We're talking about things which are starting to come in and starting to make the factories completely autonomous so that now cost of production will come down in the production phase. And this is a very similar scenario that's happening in the production space in the content business as well. Content business is getting disrupted with Gen AI. I'm sure a lot of us are building videos today with all the Google Omni and Google Nano Banana and obviously, with a bunch of tools like runway, a whole of opportunities there that are starting to happen, sedans. It's very early today, but now we are starting to see production really at high-quality production systems starting to evolve. In the next few years, you're going to see a dramatically fundamental shift in terms of how this particular technology is going to start stuff. What this essentially would do for the whole thing is that the cost of production is going to deflate. So let's look at it from an economic standpoint. If AI works and promises the way it kind of works today, what would happen is that the cost of content production will deflate. That essentially would mean that the whole economics of the business have to be rethought for every part of the value chain in terms of the production cost actually deflates in terms of stuff. And when I say deflate here, we are starting to see a new set of AI-first studios evolving -- these are studios who are trying to use AI as an important tool, but bring in the creative content capabilities and storytelling capabilities of humans and being able to fuse this to create very compelling stories. I'm sure we have seen some in India, but we're seeing this across the globe as a major shift that's starting to happen. It's early in the days of what it is, but that's a big shift that's starting to happen. I feel maybe live sports might be the only non-AI content to survive if you look at it in the larger scheme of things, right? Live sports is like gladiated sports for all of us. We -- I think from the Roman times, we always love to see humans competing against each other. And I think that is something that we'll continue to see even in the future, given the emotions and the human needs, for example. But all of the content technically will start to come to a point of where it will be very hard for us to humans to delineate whether those AI generated or human generated to some extent. So that's a big change that's starting to happen. But once you create the content, the other pieces which are important as to what we call the media preparation stage. This is a stage which is a lot of human cost today, for example. People have to take the content, encode it in different formats and create a lot of different, what we call metadata. Now what are all of this? If you look at it, I'm just giving you an example of what it means by metadata. Think about you're watching a content today. I'm sure when we open our OTT platforms, we see what we call all of the data that you see on the screen is what we call metadata. If you look at it, there is an image, which we call an artwork. Typically, it's a description, which is a storyline. There are program guides ratings. You see the actors, you see the episodic information. You see the content advisories, you see a lot of audio tracks, audio subtitles, for example, and who are the creators, copyrights, PG ratings. All of this forms what we call as metadata about each of the content that you create. Now think about in an environment where I have thousands of hours of content getting created, and I want to do it across 40 countries means I have to have multiple languages, not only with the metadata, but also the language of the content, subtitles and everything else. And think about I need to do it across multiple languages, multiple countries, regulations are different in different countries. This is the heart of what we call as a metadata explosion that happens in the business. Today, it's technically a human cost today. Lots of humans in this factory of content, lots of humans work, either trying to manually creating this metadata content, looking at compliances. So think about almost like -- and I think about if I'm a sports company, I would be doing thousands of hours of sports every month. I'm getting a lot of content coming in. News, for example, every day, I'm creating a minimum of 6 hours of unique news content that I'm creating. So you're seeing a whole slew of content coming into my system. I start to get inundated with the amount of work to be done. So there is clearly a big headcount crisis in most of these companies today in media companies because it's amount of work to be done. And if you look at it -- and when we talk to customers and others, what are the learnings that we're having is for every dollar that they've been spending on technology, there's $2 to $4 of human toil. This is a big limiting factor for the video businesses today because people have to see eyes and ears, people have to see it, control it, manage it. Everything was a human element of this whole system, right? And that is the biggest change that we are starting to see primarily through what I call agents to the rescue. So agentic infrastructure when we talk about -- and again, agent is a very overabused word in the industry. Everybody calls everything an agent. I'm going to kind of slightly talk about what agent means in this context. But largely, what we're seeing is that -- now the whole chore of how people have been doing all of these content preparation stages, for example, across multiple continents, globalization of the whole thing will start to become what we call agent tech. And let's kind of give an example of what agent is so that we're all on the same page, right? So look at it, again, I'm just taking a coffee machine example, so that's easier for all of us to understand. Take a simple coffee machine, for example. A coffee machine, which is standard is a non-agent. And when I say non-agent, it has a preprogrammed set of things that you can do. You press a button, it does a particular sequence of things. You need an espresso, you need a capuccino, you need a latte, to click a button and it actually gets done. So this is not an agent. So this is just automation, a basic thing that the machine is supposed to do. And this is a normal world of what we live in today, for example. What is an agent? If you think about the coffee machine became an agent for a moment, it actually is AI-enabled, how will it look like, right? Think about the coffee machine actually talking to you. It's telling you, I've been monitoring you. And based on your facial expressions and the time of day, I predict that you might need an espresso. What did the system do? It understood your need proactively. It anticipated what you would like to have. And it actually is able to kind of act on that and be able to deliver something that was not really thought about when the agent was created or when the machine was created, the machine was not programmed specifically to react to what you're seeing, but it's trained itself to kind of react to this whole thing. This is what we call as an agentic infrastructure. So this is, again, not only for a coffee machine, for any task, anything in any job in any of the industries that you see, this is what's happening. The ability to action on -- first to be able to predict and eventually anticipate and act on it is what we call an agent. Obviously, this is not a stop, right? As you know, as AI technology is becoming better and better and better, we are starting to see what I call true AI agency. What this means is like this is the AGI world that everybody is talking about, right? In a true agency, it learns by itself, designs by itself and changes in ways that as creators, we don't anticipate this whole thing. This is the future of how AI technology and agentic infrastructure is going towards. Think about the coffee machine really saying, you know what, I've invented a new drink for you, and I think you like it because I know what it is. And I'm going to make a cup of that particular coffee for you, for example, which is interesting because it actually invented something new because it customized itself because it knew that you wanted it, for example. That is what we call true AI agency, right? So this is the future of how the progression of agentic infrastructure is going for all of us. Started with no agent, today, we are in the world of agentic era where the coffee machine equivalents are starting to really understand our needs and predict our needs and be able to do that. The third step is where it's able to pretty much invent and evolve itself is where I think the future lies in terms of how the AI technology is going. And we're seeing this in our own business and media business as well, where we're seeing that particular transformation that's starting to happen. Now if you look at it from a packaging and preparation standpoint, it's exactly like a warehouse. Now warehouses across the globe are starting to look at robotic agents, agents which are actually doing the job of being able to sort things, package things, getting things done, for example. As you know, we've seen [indiscernible] robots to multiple different robots starting to really come into this picture. Very similar is what we're seeing in the AI in the media business as well is where we're seeing an evolution of digital agents. These are agents, as I talked to you, these are -- think about it as very glorified coffee machines in the [indiscernible] similar is tasks and software, which are actually AI-enabled software, which understands the media, understands your video, understands audio and understands the storylines and is able to extract and create those metadata. What I showed you as an artwork there or what I showed you as a program guy as a rating of a particular content or what I showed you as a story summary, for example, today, machines can actually create those summaries automatically. We don't need a human in the middle to do that. By just providing an overall sense of philosophy of how you want the metadata to look like, systems can do that today automatically. That's the power of these systems. So what it essentially does is that these agents are starting to replace a lot of churn work which is very hard because we couldn't get our customers to scale because all of the media companies were really struggling with this problem of messy media. You would have seen our [indiscernible] report that went out a week back, and I would urge some of you to download that and see that as well, where we talked about metadata problems is one of the largest problem the industry faces, which is slowing this down. And we believe, like what you've seen here, for example, the agents will start to become a very, very important part of the whole equation to solve these problems, for example. And once you solve it, how should it really look like, again, this is an imagination of a daily experience. Think about you are a scheduler in a TV channel, for example. The way I think eventually, we need to -- in an agentic world, you need to enable it is 20 to 50 a.m., this is obviously early mornings. The system is actually working. You are sleeping for sure as a scheduler, but the system is actually building all of the schedules for you automatically. It understands your taste. It understands the genre of the content that you're working with. It understands new content coming to the system. It understands the business needs that you have. It's understood the analytics of what happened yesterday in terms of viewership. It's looked at social signals and figured out that yesterday was [indiscernible] birthday. It understands all of that. So once it understands, it's able to schedule it, do what a particular scheduler can do as a coworker, it actually works with you to do that job. Think about 7:00 in the morning when you wake up. The first thing it says, "Hey, I finished my whole schedule. Here is a summary of the schedule. I want you to take some decisions. Hey, human operator, please help me to take those decisions", and it actually reads it for you to wrap up. Once you do that and you see the whole grid, you are comfortable with the schedule, if you look at everything and say, okay, everything is done. And in 10 minutes, you finish your job, you confirm it and the whole system is scheduled and you're done with the job. Today, this takes 8 hours of a particular person to do this day in and day out, example. Now think about the value and the productivity that can happen is now this person can suddenly can do tens of channels. And all our customers want to do more and more channels, more because of content -- amount of content available and the number of global regions they want to reach is much higher. Now these are agenetic infrastructure starting to really help for people to be able to create this whole preparation step which is the biggest core job today, for example, will become agentic, and that's the biggest change from an AI standpoint and that we're going to see as we move forward. But once you do it, the question is also, when you have the content done, you want to be able to take that content and deliver it across multiple platforms. As I told you, content is going global. You want to go to tens of countries in different languages and different regulations and we go to different people. This is a logistics problem. This is exactly like a -- you're a toothpaste manufacturer, you want to take your toothpaste and be able to take it to every possible store in the world, be it a [indiscernible] store or e-commerce or a quick commerce or wherever you want to take it to a big retailer, for example, when you start doing that, the biggest challenge is a lot of back-and-forth interactions that happens today, for example. Studio has to transact today, all of those transactions are almost like a human transactions today. Talking to each other. It's such a very simple thing, right? So for gain OTT platform, I ask a simple question, hey, what is the show for Diwali that [indiscernible] now just think about it, this platform, OTT brand has to do this for hundreds of such content owners a very similar conversation. Either this happens on e-mail today or in phone transactions where this person is actually talking to hundreds of such content creators. The reverses grew as well, the content creators are talking to 100 platforms worldwide across 40-plus countries where they need to kind of get these capabilities actually defined as well for example. So what you see in the world today of a lot of chatter and conversations and communications and negotiations happen between the studio or the content creator and the brand, which actually need to do best example. Now -- so this is a very manually intensive job. It's not only about communication. It's about preparation, negotiations and everything else that starts to happen. Think about a world that we're actually able to create 2 agents which are talking to each other. An agent, which understands the platform's needs, is talking to hundreds of such content creators saying, "Hey, what are you planning for Diwali. Each one of you, can you tell me what you're doing?" And think about the agent able to aggregate it from hundreds of content creators find out the best Diwali shows that it wants and starts to pay for advertising or even creating content page to really promote Diwali content for all its viewers for example. Think about an imaginary situation on the media studio side, for example, the studio and now be able to talk to all its tens of different platforms they're going to worldwide, and actually talk about what they're going to plan for Diwali and be able to provide that and prepare it in multiple languages, multiple art work, different regulation environments, they can prepare all of those capabilities [indiscernible] example. Now all of this is a big challenge. Just by having a human system, we were able to kind of find it very hard to connect all these dots together. What we see is the future is what we call inter-company agent transactions. This is a very powerful new concept that's to happen in every industry, media so different is if you look at all the agents that we talked about earlier, they're all inside the company doing a particular job for that particular company. In the media companies as a content factory, as I told you, they define -- they kind of decide on the content, create your metadata, [indiscernible] content, encoding, all of the content relocation is what it did. But here is a possibility where -- now these agents are talking to other platforms, transacting, negotiating commercials, identifying what's the opportunities and eventually be able to transact it like a marketplace. Think about the agents starting to talk to each other. Think about every company had agent interfaces through which it happens. Your legal documents get done. Your portion negotiations get done. The technicalities of what content you're transacting gets done. And it's -- this is the step in the future that we see, where you're going to see this all interaction starting to hand across agent to agent that's going to happen, for example. That's on the -- just -- so what we started off with producing content, which I think the Gen AI and with a lot of ones capabilities is switching. It's changing a lot of things in terms of how people are going to creating content. Once you create the content, how do you prepare the content to be distributed. We saw this whole agent infrastructure coming in. Once you do that, now you come to the point where you actually prepare the content, you kind of transact that content across different globe. Now interagent transactions become the core feature of how we do this. Now the innovation doesn't hear. The moment you be able to do this, for example, untimely what is happening is think about delivery to consumers. Take what is happening in the world of e-commerce, of commerce that we'll see as despite. If you look at consumer delivery there in a real world of how we are buying products today, our content -- our delivery of today is happening because we're discovering content through an e-commerce or a quick commerce or any website or equals that we are actually going through, for example. But I think we had a very, very -- almost in a close phase of how [indiscernible] is happening, and that's what we call as a genetic discovery. What does it mean? Think about the world where if all of us have our personal conversational bots with us. That could be a ChatGPT, Claude or a Gemini or tomorrow, whatever the sort of conversational bot that you have. Think about what would happen. When we go to a website, and I'm sure already most of us are not searching on a Google search today, but you're actually going to your conversation bots today to talk 2 things. Think about the bot. The bot has become extremely powerful. It understand -- it can actually understand your nuances, you understand your preferences. And we would start to really transact our content, our goods through that. So we're going to really buy all our stuff through possibly agentic discovery [indiscernible], which is nothing but our own personalized conversation bot, which understands us, for example. That's the same thing that would potentially happen for the viewer of viewing [indiscernible] well, right? We're already seeing that, for example, should you go to a destination, should you go to an OTT brand website to actually look for contact. Yes, the content is locked behind all of the paywalls per day for example, but think about a world that could potentially happen is you have an genetic discovery platform. You as a consumer would own the surface all the interface, all the clients that you have, for example, where you are able to pretty much decide on the content that you want to watch and you are able to click patents from that and be able to watch the content. The discovery is personalized to an extent that it knows what you have watched earlier and not specifically on a particular surface, not about Netflix or a prime or a YouTube or any of the other platforms that you're watching but they're actually having a good understanding of what you -- what you're watching behavior is across the whole spectrum of things you do. It also understands your mood. It understands your emotions. It understands the things that you are actually dealing with on an everyday basis. It understands you better than anything else in the world. And that's the power of this whole agenetic infrastructure that's starting to happen on a personal level. So I believe that the discovery of content, discovery of commerce will start to change, where we're going to start to really transact content through an entirely different surface, which is where I feel the next step in the evolution of these technologies are going to happen. So if look at it, we've seen how production is changing. We saw preparation changing. We saw the whole transaction of platform distribution changing, but more importantly, how consumers consume, how consumers discover will also change. So the whole value chain is starting to really look at the complete change because of this whole AI technology is starting to really impact everybody in that particular equation, right? And it's a tremendous value big change that's starting to happen. But not only [indiscernible] that. If you look at it, AI not only supports change from an agent standpoint that we talked about, but it can also help you to really make your content better, extended reach, create entirely new sets of content. With Gen AI, obviously, we're going to start creating a lot more content, which we're going to create capital. But that's only not the first stage is only question. What we believe is that there's going to be a big transformation in how we create content or even repurpose an existing content, for example. And that's going to be the big next step in the equation that's starting to happen. What is the fundamental hypothesis then starting to look at this whole thing? Example is if you look at it today, and this is happening across all the different media businesses that you will see it -- there's a rich reality of storytelling that tests, be it sports, be it news, be it content. There's a lot of reality of we are capturing. Think about it in a sports game, for example, there is close to anywhere from 20 to 100 cameras today on a sports field, capturing a particular moment in time in sports. There are sensors, there are microphones, there are so many different gadgetry that actually support a very rich reality of the interface. Now unfortunately, because of the current bottleneck of both humans and the hardware that was kind of stuff, you see what I call a bottleneck in all of these capabilities today, for example. This bottleneck was because we could not scale as humans, and they cannot scale our hardware that existed to apple. And that -- what it did enable us to actually create a homogenized mechanism of how we watch content today. All of us watch a very similar content. It doesn't matter if we are watching on OTT or cable or any other platform today. We watch a very similar set of capabilities. The same set of cameras which is the same sort of commentaries, the same sort of news content that you were watching for example, on a TV, for example, on a live TV for example or eventually even entertainment content exactly not changed. What would AI do for this is a big change of being able to transform this capability dramatically, for example. And one example, for example, think about it, if you had a single match that you're watching. So if you're watching a cricket batch or a football match, for example, single football match now can be actually if you remove the human cost of somebody deciding to switch cameras, if you move the hardware costs out of the system, for example, now you have a world where you can create many, many stories out of the same match about you like analytically look at the game, you did it very differently with tactical fans, you could look at all of this or a home fan, maybe I'll give you a complete different story line of through home game equivalent, the cameras I switch. The commentary that I do for you, the players that actually show you, for example, the statistics that you see on the screen, the graphics that you see. Everything will be specifically tailored towards a particular for [indiscernible] and what we want to watch for example. Today, this is very expensive. So I want to do it through [indiscernible]. And if you want to do it with a lot of hardware, it's a very, very hard problem today, for example. But AI fundamentally shifts this whole thing from a cost standpoint to a third and enable in real time, the ability for us to kind of be able to change this, for example. So the future is where you can have the same match, but you can actually see multiple stories. If you're a Gen Z, maybe you have a partial screen on which you are actually watching content, you're actually chatting with your friends, and you're able to do this. And you're able to put at multiple camera positions that you're able to see simultaneously for example. So you are able to create native storytelling across different things. And AI is a very big enabler for these sort of capabilities that we can actually be able to enable us to move forward as a company. But -- so this is one big change that we've seen in the industry. The other thing, for example, is if look at it, I'm sure lots of you are starting to look at macro trauma. And today, if you look at it, thanks to Gen AI, there's the micro drama is something that's getting created today at micro drama, lots of teams and companies and production houses are creating micro drama from scratch. Now with AI, but there are so much of stories that we've told in the past. Think about we're able to take our stories in the past. And over the last 50 years, I'm sure, have been so many storytelling so many movies and so many soaps and episodes and CDOs that we've done, for example, think about we're able to bring it to the new world. Bring it to the new world, think about -- again, example is a 3.5-hour solid movie. Think about the solid movie is actually available as 3-minute micro dramas in 10 episodes. In 30 minutes, you were able to see the whole story for example. That's sort of an interesting sort of play that we can actually bring on board, where we are able to bring in an entirely new set of content and existing content to a new format that they are able to say. Same story but now you could do multiple story-telling formats. And again, this is not possible pre-AI because humans couldn't have taken million as of content, converted into vertical from a storytelling fan. Now with the, these are all becoming a possibility of transforming content. So not only it's becoming a genetic, but the transformation of the content itself is a reality that's starting to happen. Now what I would kind of leave with you is that we saw that not only cost operating leverage can be brought in because of all the agenetic automation that you can actually bring on board, the smartest connects, deal with things, but more importantly, the content can be transformed. The storytelling can be transformed, which I think is a very big step in terms of how we see this whole transformation starting to happen. And so all of this and look at the technologies that are evolving. It's a very fast evolving space like everything else, but the content business technology is slightly more complex. Why is it complex? Because it has audio, it has video, it has text. Unlike businesses which are just a typical enterprise business there, we're transferring everything through only a text or a dock. Video businesses actually have to deal with audio. And when I say audio, it's not only about speech perhaps. It's music, background audio, everything has to be understood. This is where humans is spending a lot of eyes and years to bet all of this and hear all of this, for example, right? You take a video, for example, I need to understand the background, the characters, the action, the emotions, and I'm sure if you look at the game, I need to be able to understand if it [indiscernible] a goal to an injury [indiscernible] to a yellow card, every possible thing that I would like to kind of understand for example. Similarly, if you look at text, I need to understand what's happening, how is the summary creating, what's the same descriptions or everything else in the target at. So what you see essentially is that -- as you start creating technologies for example, the [indiscernible], it is very complex to solve these problems because you have to solve all of this together. Now already, this is a field that's actually started evolving much faster, right? Obviously, large language models, great in text, you can definitely use it, that's your base of what you do from a text standpoint. But if you look at it, obviously, you build small language models because you're reasoning you may not be very expensive, so you create those capabilities as well. But more importantly, you are talking about video and audio, which are the rich in terms of the stories they tell, the emotions they carry and the information that's available, for example. Now we are in a very early phase of that transformation. I think they've done leaps and bounds over the last 2 years in terms of obviously the Gen AI video generation. But you also need video line vision language models, which are nothing but VLMs, what we call, is to extract that information as well. So I want to have an eyes in years model, right? So you want to be able to understand from eyes and years, what is happening on the scene and be able to extract that because that's the way you understand the multidimensional aspect of a particular movie or a content that you're actually watching for example. So what we call is VLMs are actually driving that particular part of the house, right? Very early, very few VLMs have come in, which are starting to engage. But they are very expensive and very slow today, right? What's happening? The next step that's happening is what we call world model. I'm sure some of you have heard of this whole world models. World model is trying to predict from the current situation that you see, what is the next step that particular real-world environment would have done. Think about it -- obviously, they started in the autonomous car scenarios. If I know that I am, what is the next moment in time and what would happen, I don't know, a car in front of me would move or should I actually put my brakes you are actually predicting what's the next step in the equation that you do because you understand the real world and you simulate the real world in front of you, for example. That's a big use case for [indiscernible] as well. If you look at content business, for example, you can create, for example, sports scenarios, for example. Take a sport scenario. If I know one example of a particular location or time of a person kicking a ball, I start to predict how the ball is going to really flow because the system starts to understand the physics of the ground, the physics of our planet, it can actually start to understand the ad drag, it can actually start to kind of predict all of that, for example. Exciting new possibilities if world models start to work the way it is in the next couple of years, I think this will be a big change for the industry. You can create much more immersive experiences that was really impossible today. Think about the camera positions, and I'm sure a lot of you [indiscernible] today, I would love to see it from the globes of [indiscernible] catching a ball -- protecting the ball for example. I would love to see it from a shoes of a player when Messi kicks that goal, for example. I want to see how fast the ball goes, for example. I want to be in the ball to kind of see how the ball is actually transferred. So the immersiveness of the experience is now driven by the cameras that we have. But in the future, with world models, you're going to see the immersiveness of the experience is going to be driven by the ability for us to track every part of a game or a sport or a scene or whatever you would see, for example. So immersive video truly is dependent on how world [indiscernible] will evolve. And this is a big area of research that's happening worldwide. And we're truly excited because this is something that I can dramatically change how video and AI can be a big change for us for example. But if you look at it, because of the 3 years also, as I told you, very hard problems of video and you want sometimes very low latency, you want your cost to be reduced. You want your performance to be very high. If you're talking about -- I talked about games where we're talking about a football game and you want to create multi different productions, different cameras to be switched to different people, for example. Now a human can see that and quickly in their mind, decide cameras to switch example. Think about if I put a machine in front of it to do that, the machines, if I put a, I don't know, a language [indiscernible] of any fashion, the vision language model, it will take a few seconds to even understand what the scene is all about. That's not acceptable in the content business, for example. So we -- to see that this is an important aspect is there is going to be hundreds of custom audio video models, models which are very specific to [indiscernible], some things can understand. If we are just a basketball game or a soccer game, something you understand just the environment. Some things can actually only do a modeling a great ocean scene for example, for you. So the world models and the custom models are a big important part of the technology evolution that's starting to happen in this particular stuff, right? So audio, video drives much, much larger needs than just LLMs. So it's not a Claude or a GPT or ChatGPT or Gemini that gets all problems in the video and audio business because it needs a lot more not only for generating of video, but also for extracting information on the video to take any sort of reasoning decisions. It's a fairly large and complex problems. We are in the very early phase of the transformation of those technologies that are starting to happen. But I think it's truly an exciting phase because audience videos are next because like humans and how we kind of taken and reason our systems, through our eyes and years that are very, very important part of the whole phase of how we access content and react to content. I see the same thing starting to happen in the systems as well as we go forward. So in a sense, if you look at it, I just want to kind of stuff AI transformation is happening across the value this content business. So it's not one is another. We're seeing this change foundational transformation like every other business that I'm sure you folks have been kind of interacting with multiple different companies and multiple different businesses media business is going through that transformation. It's truly an exciting time because if this is a new year that's starting foundationally. And I think this is going to change the players, change the who creates -- changes the personas, changes the way you consume content changes the whole story telling economics completely. And that's exciting because this sort of a transformation happens once in a lifetime. And that's where I think the opportunity lies in terms of being able to build not only technology but entirely new order of businesses and the value chain getting reconfigured completely. So with that, I would like to kind of stop this presentation for you to tell you that, hey, this is very -- this is an extremely exciting maturity. AI and media is one of the most compelling challenging opportunities that we are seeing. And obviously, I'm biased because I'm part of a media company, but truly, this is a big [indiscernible] folks. And the change is extremely palpable. The technology is complex, but that makes it the most fun in terms of doing the transformation in front of us, the experiences that you and me as audiences will see is going to be dramatically different. And with that, thank you for this time that you all spent with us and we'll go to questioning. Amoolya, what's the next step that you want to do?

Yes. Yes. Just before we go to questions, there's a QR code on the page here for [indiscernible] feedback to us so that we can improve. I'll start with [indiscernible] some of the questions that are on the chat here. But I want to reiterate that the discussion is going to be strictly focused on AI and media. So all questions, please be limited to this topic. We will not be taking company-related questions in this session, right? So the first question, Baskar, is from [indiscernible] given that we have strict SLAs up to 99.9999% with clients, agentic AI can be prone to hallucination and be at times unpredictable. What are the safeguards we have in place to mitigate such risks?

Specifically, I wouldn't want to [indiscernible] make the systems not alternate and be able to deliver value to all the [indiscernible] so a lot of work that our company and actually a lot of companies are doing is to bring [indiscernible] our India restate problem, right? You need to put a lot more [indiscernible], you build what's called [indiscernible] systems and when you're talking about a factory of content, it is thousand [indiscernible]. The predictability to bring in is the most important. And I'm sure if you talk to any companies which are in the cutting edge of building AI solutions for customers, the challenge is not to demos and POCs are the easiest. The hardest is to get them into production and make it work realistically in all these lines. So that as part and parcel of the key value of building, technology and engineering of these companies to make it happen. So that's what it is today.

Got it. From [indiscernible] regarding the prep work, are we seeing any improvements in the speed in work due to AI? Does it in turn reduce pricing because of AI-led deflation?

So partly to look at it is this is a human cost reduction, we mentioned, right? If you look at it, the fundamental in the core technology remains exactly the same secret. You need to be [indiscernible] content. But I think what was a human element, take an example for that. I'm sure you folks have seen that today, for example, speed to test that a point that all our subpar needs are going away. [indiscernible] of any artwork creation, a promotion for them, which is a human job. Somebody has to do is changing. So this is actually a net impact in terms of -- we are seeing this is a an incremental that's starting to happen because this is a cost of human cost that was there, which is either not possible or would it be expanded and that's where I think newer business and possibilities are starting to happen for us. That's what we see.

Another question is from [ Shankar ], can AI become a lever for getting production and preproduction workflows to be hosted in Claude?

Absolutely. One of the trend lines that we are seeing across the whole customer spectrum is that AI by definition, is driving a progression towards more scalable cloud platforms. [indiscernible] is fairly straightforward because I don't have actual [indiscernible] in my own facilities exactly. Clearly, I need more from [indiscernible] non-cloud. And that's progression is happening because of acceleration of AI workloads that we applied in market format, and that's a big driver. Absolutely. We see a lot more progress [indiscernible]

Question coming in from [indiscernible] from DAM Capital. Have you seen increase in OTT cable-based content because of AI? Any real-life examples on cost that has been reduced?

Because in bits and pieces today, [indiscernible], I don't think it's not a big factory approach yet the whole system. I'm already seeing a lot of content creators telling us that, for example, the VFX cost has come down. So that's the first step. So they're not going and changing the whole movie content or storytelling completely, but parts of the story is now getting told through an AI, either a background creations or in specific place locations where they're actually starting to use it for storytelling. So that's where we're seeing it. It's early days, but I'm sure given the direction and what we're seeing in micro drama, I see a lot more activity. But in the OTT platforms and the content creation platforms, albeit small today, but definitely, the direction is very clear that it's going to be a big tsunami of sorts in terms of how content is going to be created in the future.

Got it. A similar question, Baskar, on cost saving metrics from Ayusha, are there any -- is there any evidence on cost savings for production prep, et cetera, especially given the higher token costs?

No, because, again, going back -- see, token cost is only a very small part of the cost of AI today. As I told you, it is all video. And videos are primarily where a lot of custom models are doing the jobs. Yes, there is GPU cost. So I wouldn't call it token cost in that context of the currency because token costs are very specific to the language models and how you do things. But in the work of multiple GPU cost and expenses, it is there. But what we're definitely seeing, for example, is in some of the capabilities that we're seeing, humans wouldn't have done this job. Just the volume of jobs is so high that we couldn't have done it. That's number one. So clearly, we're seeing possibilities for our customers where they can create new revenue possibilities. So that's one place that we see. The second one we're seeing is really cost savings because humans are very inefficient in doing some of these capabilities, for example. And we're seeing tremendous savings in those capabilities. And the cost of GPU is lower comparing to the cost of the human cost that has to be applied if you go to do that, right? Most of our customers are looking at more of an expansionary aspect, not as a cost-saving capability as it stands today.