Hi everyone. And thank you for being here today. So my name is Max, and I'm excited to talk about how data science and AI are transforming the music industry. So I spent over 10 years working in data science across telecom, FinTech and consulting. I also had the opportunity to lead research and analysis at Warren Music Group. And now I'm science lead at Metis. So the music industry is evolving rapidly and AI is playing crucial role is that transformation. So today we'll explore how cloud native AI is driving innovation in entertainment and music production. So now that you know a bit. More, about me and let me walk you through, what we will cover today. So before we dive into the details here, a quick overview of what we will cover today. So what is the problem? we'll start by discussing the key challenges the music industry faces today. so we'll talk about data and understand, what the source, we have. And, how they shape a decision making introduction, to AI in the music industry. the high level, we look at how AI is used in various aspects of the business. And, we'll make, two stops on two, case studies. So the first one, it's, uplift modeling. So how AI helps optimize marketing campaigns by identifying high value audiences. And the second one, it's AI driven planning and marketing, optimization. So we're using AI to predict demand and optimize a budget. And we'll talk about the impact of AI on the music industry from perspective of a cloud native. So before we get into, let's first break down what the music business actually looks like. So the industry operates across five key pillars. So the first one, it's, and our, artists and repertoire. So these departments, responsible for discovery of new talent, finding the next big start. The second one, it's a recording and, production, turning raw talent into high quality music. The next one, it's an artist promotion marketing. So the guys who are responsible for, making people, more knowledgeable and, actually hear the music. the next one, when we already promote our artists, our stars. So we need to think about how to release tracks and distribute that. So getting music onto streaming platforms and other. Channels are of course, the last one and, but not least. So it's a step of monetization and royalties. So we need to ensure that artists and labels as a warm is a group, as ourselves, we get paid for the. So each of these areas generates massive amounts of data, and this is where AI can play a huge role in making processes more efficient and data driven. But of course, this business is not without challenges. So let's talk about one of the biggest ones. we understand that AI in our days is everywhere, but, we have. To do something, but, sometimes, we don't know and labels and artists, know that, AI could be potential, but the biggest challenge that, we faced, we don't know the artists don't know and the labels as well, where and how to use it, more efficient. So it's not just about adopting AI, it's about making it work for real business. And if that was not challenging enough, there is another big hurdle. so too much data. So let's talk about it. So AI is everywhere, but there is, one problem more. It's, data is everywhere. in warm music group, so we have, over. 2. 5 trillion streaming data points, 15, 000, parameters tracked for analysis. hundreds of tables about the artists, about the track, about the album, about the audiences, 180 companies. Provide, different details, different data and share this data with our music group. So 10 billion transaction process daily and growing every day. So the problem is not having data. The problem is making. Sense of it. and that's where AI, comes in, but we need to know how to use it. So where does all this data actually come from? So the first, main source, it's a DSP. It's a digital streaming platform such as, Apple Music. Spotify, Amazon and YouTube. So they provide insights into user behavior, play counts, skips and playlist, additions. So of course, the most, one of the most popular source of data, it's a chart metric. So chart metric is a tool that tracks fun engagement across, different platforms, the technical and, from. Technician sites, we could talk about, Librosa, so it's a non obvious, but anyway, it's a one more source of, data. Librosa, it's a Python library for analyzing audio features, so it provides, extraction, useful information like tempo, pitch, and mood from tracks. And also we use, SATATON. SATATON is a data platform, which we use. To identify rising stars, it combines, streaming, social and radio data to predict, breakthrough artists. So now we actually understand the data landscape. So let's see how AI helps solve real business problems. So AI is solving two key types of problems in the music industry. It's, two different layers. the first one it's operations. And strategy, the second one, the operations are responsible for optimizing marketing and engagement. it's a problem such as, segmentation, A B testing, common channel marketing. If we talk about strategy, it's a separate division, which driving business decisions, such as, stream forecasting, churn prediction, LTV modeling, and uplift modeling. So each streaming, service. Spotify, Apple music, YouTube keeps its own, data and the main, one of the main problems, we faced, we can't track a listener across, platforms, which limits how well we understand individual behavior. So that's why uplift modeling and stream forecasting are one of our key focus areas. So it helps us measure and, real campaign, despite. These, limitations. So let's look at how AI supports the listener journey. So AI supports every step of the listener journey, started from, acquisition. So acquisition. It's a step which, led us to understand who are the new listeners. it's more about behavior monitoring, onboarding, ideas and the boarding strategies. So the second one, it's a development, the development. Part, it's about how do we keep them engaged. it's a problem, which we're going to address such as next best offer or cross sell or up sell model. The third one, it's a retention. So when we already have the user is, using our service. we need to think about who's at risk of living and responsible for, churn and think about, LTV prediction. And the last one, it's a reactivation. how do we bring them back? for example, if we found that the user had a high churn, probability, and it's a problem like a reactivation modeling. or incentives. So to tackle these challenges, we focus on two key AI applications. Uplift modeling for smarter marketing and sales forecasting to predict demand. Because these, two main problems. Cover the main, part of our business. So let's see how app uplift modeling helps optimize marketing expense. So basically, app uplift modeling help us, measure the real impact of, advertising and the optimize budget allocation. So we analyze two groups, the user who saw an ad and those, who did not. So two A models. Predict engagement for each group and uplift is the difference between these predictions. It tell us where ads actually drive more listens or clicks instead of spending on user who would engage. Anyway, we focus on those, where ads make a real difference. so maximizing impact and reducing wasted spend actually what happens when we apply uplift modeling in real campaigns. So AI driven segmentation was used to launch ads on TikTok and Spotify. So the results showed a 30 percent reduction in the cost of adding a track to a user library compared to traditional segmentation. So this proves that AI, especially uplift modeling could help us optimize ad spend by targeting. The right, listeners and improved, efficiency and reducing, costs. That's actually our main purposes. So now let's take a look, at another key AI, use case in our agenda. It must be, and it will be stream forecasting. So stream forecasting, it's about, Sales forecasting, actually. So we need to predict how many times a song, will be streamed as a critical for plan because it's critical for planning and marketing. We built a three months forecasting model using historical streaming trends, track metadata and the process, which we already mentioned today. And additionally, we used, around 700 features in total. So the result forecasting accuracy improved by 60%. helping labels, allocate budgets more effectively because, it's becoming more crucial and more important for warrior because warrior consists of different, independent label, such as Atlantic as well. So now let's see how this forecasting translates into a smarter marketing investments. So forecasting doesn't just help predict streams. It's also optimize marketing spend because, when we build, model, we need to understand that we need to not only implement our solution. We need to interpret. the decision interpret, the model. And, for that to answer on that question, we built partial dependence plot and partial dependence plot, provide us key insight. So more budget does not always mean more streams because, after. built in partial dependence, we found that, oftentimes we're faced with a plateau effect and plot effect shows where additional spending stop being effective. So by reallocating budgets based on these, insights. So we saved, 27 percent of, marketing costs without losing performance. And, we've built based on that model. real cloud native AI platform that integrates forecasting seamlessly into business process. So the first step, of our implementation, it's a front end around on Streamlit, which making predictions. And, provide for, our business units and business users more easy to access, mathematical model. And the next step, a fast API backend connects the user interface with our ML model and handling request efficiently. So user. Type specific, scenario in Streamlit. So Streamlit send this request to FastAPI. FastAPI launch ML model, ML model score and predict all forecasts. Send this forecast, in FastAPI back. And, the model process input features and return. stream forecasts, in the real time and show the real forecasts in the stream lead. So this architecture allows for fast scalable AI driven decision making and bringing real value to the music industry. And, at the end of the day, it all comes down to one thing. So music has to win. AI is not here to replace creativity, but to amplify it, helping artists and businesses. Make smarter and data driven decisions. And, our purpose and our main objective, is build cloud native AI solution that optimize marketing, predict trends and maximize impact. And we did it, but the real goal, of course, it's connecting music with the right audience in the most effective way. So thanks for your time and feel free to scan the QR code to connect. And I'm happy to take any questions.