AWS re:Invent 2024 - Revolutionizing sports with generative AI: The Bundesliga trailblazer (SPT207)

good all right great well welcome everyone thank you so much for joining us today uh I know it's the last session of the day so uh hopefully you all will learn something uh and have some takeaways that you can walk out tonight so uh today we are here to talk about bu's generative AI Journey so for those who doesn't know what Bundesliga is Bundesliga is Germany's premier football league and when I say football for this session we are actually referring to soccer for our American uh audience so we will be calling football uh throughout the session with that let me introduce myself my name is El dogen I am principal engagement manager at AWS Professional Services team I'm responsible for our Bundesliga AWS Tech road map I'm here uh with two of my colleagues U Lucas rosovich who is the head of uh Partners strategic corporations at dfl or DOA fball Liga and Bin Zuber who is the um CEO of thef digital Sports so here's our agenda for today before we even get talk about technology we want to talk about the culture of innovation that we have at Amazon some of you as are you are partners and customers you're already familiar how we innovate but I'd like to talk about the culture first because this is the approach that we are taking with our Sports customers but also specifically with Bundesliga then we will be talking about some of the work that we already achieved together with bundes Lia and then Lucas will come to Stage talking about bundesliga's Innovation story and their strategy how they're approaching to continuously innovate and then uh Bastian will come and talk about how they are embracing the generative Ai and kind of what is next for bundes Lia all right so you've probably seen this before this is our mission statement and our mission is to be Earth's most customer Centric company this is very broad meaning it doesn't L really have restrictions or limitations about the markets that we enter domains that we target for it's not defined by industry geography we put customer at the Centric of our whatever everything that we do to continuously innovate while this Miss mission is all broad but it's also very specific because often tiebreaking question when we are trying to make decision is this really customer obsessed and what is it for the our customers so why we are putting customer on the center of our business in 2016 Jeff Bezos um in the share the the shareholder rather he wrote a lot of benefits about being customer obsessed but this highlighted one is one of the really good summary of customers are always beautifully wonderfully dissatisfied even when they are report being happy and business is great putting customer at the center of everything you do um we have endless inspiration and Innovation to continuously innovate on behalf of our customers one of the Hallmark mechanism of uh being customer obsessed is working backwards so this mechanism allows us to really start thinking what we want to achieve and work backwards from that and meaning we start with an idea and then really look at the vision what is it that we are trying to achieve and then start working backwards to start implementing those ideas and another one that I like to share is um oneway two-way door decision uh this is a concept that we at Amazon really adopt it and we see our customers also adopting it as you all know speed matters in business we have to make decisions that are um fast and we can take to the market so what we call those are oneway door decision they are reversible uh so it's like less risky and there are two-way door decisions that are have potentially longer term consequences so when we think about sports we are thinking about team sports clubs you know a lot of action and it's very fast so it's important to start small do a lot of experimentation and really take the speed being Speedy to take the uh products to the market and this is another quote from our um uh CEO and president Andy jassi uh we often say there no uh compression algorithm for experience through the experimentation starting slow and taking to uh larger steps and experimenting a lot you're actually learning a lot throughout the process and that's been the approach that we are taking also with Bundesliga and you will hear from them um later all right so let's talk about how we at AWS are transforming the sports industry using AWS Cloud we are looking in different areas where we talk about sports at AWS so the first area probably something that you can relate to there are so many data out there and that's the case for sports as well imagine where every player every ball position every fan uh preference and their basically behavior is being captured as a data we use this data to build a data storytelling um actually also the performance metrics that can inform the clubs or the teams that to make a better data driven decision on their strategy or improving on field performance uh and also providing a customized personalized content for their fans based on their need we also with National Football League we also have a prediction of injuries so we collect 500 million data points per week uh with for the their um player injury health and safety product and the second area is the live Cloud production so producing high quality low latency sustainable uh cloud-based production enables us to create a personalized interactive and experience and the third area now let's imagine um you are going walking into the stadium you don't have to bring a bag wallet anything you're just walking in and you immediately are recognized through our Amazon one identity management system and then uh you're going to concession stand with just walkout technology everything is basically at your fingertips while iot internet of things and sensors are providing on the background a sustainable experience for fans so that you don't have to wait on the line so that's what smart venue uh targets and the last but not least least is the fan data and analytics um to Super serve the fan we really need to identify and understand them we need to create personalized individualized experience and that start with the data collection and creating unified 360 fan profiles where we can actually know about fans where they come from to be able to provide those individual experience for fans of course generative AI we see applications across the whole uh for area many of Bo's leading organizations are working with AWS AWS Professional Services and AWS Partners to migrate deploy and in uh innovative solutions we are reshaping the sports industry and there are a lot of use cases but today we are going to be focusing on bundes Lia so let me start with a question how many people in the audience have seen a Bundesliga game before okay pretty good majority um you know football is not only about 90 minute game right and it's about emotion ambition passion and in Germany the Lev of football is overwhelming Bundesliga has 26 million people watching the game live in the stadium every season they have more than 1 billion Fen worldwide that are following action and our partnership with Bundesliga started in uh 2020 and we've been innovating since then [Applause] [Music] [Applause] [Music] [Applause] oh [Music] [Applause] [Music] [Applause] [Music] hope you felt some of the passion and energy that we are feeling in the stadium here um but before we dive into specific use cases I like to talk about a little bit the data collection part or what kind of data bundes Lia actually utilized because this is a foundation with a lot of the products that we do uh build together so when we look at the Bunda data we are looking at like kind of four level um the first layer that you see here is um actually the match footage so Bunda has the wor largest football archive soccer archive um and there are about 100 more than 150,000 uh or hours of historical match footage and it's not just only about the video but there also metadata and the images as well so the second layer is the sports data and we are going to get into this a little bit uh more detail as we are talking about few products that are directly really using the sports data and the level three is more insights like the bundes Liga matchback Advanced an statistics that we build together with Bundesliga and the final the top layer is the business data those are the data the kpis or key performance indicators that we collect to help us inform uh this making a decision are we really effective are we delivering the right things so it's kind of the feedback loop that we are do uh we are collecting okay let's talk about the sports data so uh in sports data every Bunda stadium has 16 to 20 different cameras that that are located around the stadium those are not the broadcast cameras they are actually tracking data they're capturing data so we collect about uh 25 time position of the player and the ball position 25 times per second so in total we collect 3.6 million data points per game in addition to that we also collect what we call it event data those are the data important events that are happening on the football field so it's basically go penalty uh corner kick and so forth so we collect this data and we utilize this data to actually build products which is something that I like to talk about next Bunda match facts those are um using this data these are the realtime statistics called Bunda matcha to be the purpose of Bunda matcha is really uh the reason why we did this is to enhance fun experience and provide more insights into the game uh for fans we collect the data and with the AWS servus architecture and using um AWS services like Amazon Sage maker we process this data and we so far produce 15 uh bundes Lea match facts in total so let's see how they look like on the broadcast [Music] [Applause] get [Music] [Music] [Applause] [Applause] right so next time when you watch the Bundesliga game maybe you'll see some of those next I'd like to move from broadcast to actually like from data to personalization so one of the most important digital distribution channel for us to connect with fans are is the official Bunda app we have millions of users and since the beginning of our partnership we've done a lot of work to create more personalized Fan Experience that I like to walk you through so bundes Liga has um bundes Liga 1 and bundes Liga 2 in total they have 36 different clubs so one of the idea was how can we really personalize the content and engage more fans uh we did a survey for fans and we find out that each fan is on average are following four different clubs even if they have their favorite but they still like to know um more about the others as well so each fan is unique with their own preferences when you add like geography demographics and the additional data points that they are each unique preferences you actually get a lot of data so we get 1.4 million Club combination and in addition to that they are own preferences so we put all that information into what we call Amazon personalized to create a more personalized Fan Experience so let's see what that looks like so I personally follow four clubs and what you see here is actually my screen so those are the four clubs that I am supportive um at least for this season and you see that I'm getting the personalized content based on uh my preferences that I pick um we as soon as we launched the personalized we also like to get the feedback and input to continuously improve and innovate uh we saw 17% increase on app usage and um 32% increase uh in sessions with the article read and 67 uh% increase on total articles that was read here I also like to share this this is also official Bundesliga app uh what the bundes lia match fracks or those uh Advanced statistics look like you're able to select at the team level or the player level where you can really understand how your favorite player was performed what was the shot of that uh what was the speed of that shot that your favorite team uh player maybe uh performed in that game so with that uh I would like to invite Lucas to the stage to talk us about a little bit um what how Bunda is innovating and what strategy they are following so thank you all right thank you El if um also welcome from my side I'm Lucas and I'm the head of strategic corporations at dfl and um I was about to explain to you actually what Bundesliga and dfl is but this audience seems to be quite familiar with with Bundesliga so just so so you know dfl is what's behind Bundesliga so we organize Bundesliga and Bundesliga 2 uh think of us a little bit like the NFL but again remember no touchdowns we score goals all right El if has basically um explained to you how this partnership has been success successful for the last couple of years I today want to focus more on what's next what do we do with data and generative AI but before I want to do that we have to look at the foundation for that and that's the working backwards method that El if pointed out to as well so why do we work backwards well because we're customer obsessed out of conviction but also because we have to be why do we have to be customer Centric because we have a very diverse fan base um we have millions of fans worldwide in the ape region alone we have 400 million Bundesliga fans and uh when it comes to Bundesliga content those uh fans in the ape region have fundamentally different desires compared to the fans that are the local fans that go to the stadium each and every week we as a Bundesliga want to provide them with a personalized experience and in order to do that we have to understand their challenges to consume Bundesliga content in a way that they want to want to do so and so we have to work backwards from these challenges understand how to best apply technology to solve these Solutions and provide them with the content they need that's number one number two is actually we innovate for our unique clubs our clubs are among the ones with the most tradition in the world most passionate fan groups and we have the second highest um spectatorship uh behind the NFL stadium Stadium average uh attendance actually I should say after the NFL in professional sports as a Bundesliga we have we're the home for all of these 36 clubs and we want to help them grow into internationally the third reason International Partners right our International broadcast partners are the ones making it all possible and just like us they have audiences themselves and we want to provide them with a Content offering that supports their business model so three Target groups where we listen to where we innovate upon and where we Implement solutions to help them that is essentially the working back backwards method of Amazon and um I think it's a testament to our partnership that we have implemented that in our corporate culture as well and actually we use gen to help us in that process so as part of our Innovation process we compile PR FAQ documents these are documents that um describe the solution from a point where we actually um send out a press release or would potentially send out a press release and attached to that that there are frequently asked asked questions about the product and about the solution this helps us internally to have Clarity and transparency of what we want to achieve and now we use Amazon B Amazon party rock as an as an application to help us draft these documents so we answer in an um in in Party Rock the most important customer questions and and party rock creates prfq drafts that we can use then and iterate we're just quicker to start start to innovate by that all right that's all you know the rather theoretical part but theory is really the gray matter it really happens what um it really matters what happens on the pitch that's a finite saying of a Bundesliga Legend and so let's go closer to the pitch for that let's take a look at our strategy our strategy is what we call glass to glass from the glass of the camera in the in the stadium until the glass of the um device of the fan we control the media value chain so in terms of production data platforms and content as well as pushing it out by our sales units we control that media value chain why is that important because that allows us to use AWS services expertise know how across this uh media value chain so we don't have to go to any other third parties to ask them hey you do do you want to do that we can decide that ourselves that's also why the partnership that we have with uh AWS is focusing on the most important pillars of our media value chain which is data production and the fan and we want to create the next level of data services media production and Fan Experience powered by AWS these are not just um you know marketing slogans that that we came up with we actually came up with these together with our teams and the teams from AWS because this is our focus and our customers off fense focus so let me start with data services um in 2016 we started our JV sporttech Solutions or STS as we call it which is enabling us to control our official match data IL pointed out to that before SCS as we call it is not just providing data services for ourselves but also for the FIFA or the MLS here in the US and it all starts with the collection in the stadium after that um the data is pushed to our Bundesliga data Hub which runs on AWS where life life analytics are performed where data is stored of course where our data feeds are processed and delivered and we build products up on that products like the Bundesliga match packs our our nextg stats products like life support for commentaries and um products life match sensors or as we call them life tickers on the letter two let's focus on them our first use case here that we want to go a little deeper in is our gen live ticker today our fans already demand up to-date real-time information about about games right that process however it's still quite manual and it doesn't really allow us to scale and uh the Gen life ticker is a fully automated solution that is based on our event data and it's providing a play-by-play text based commentary and that fills exactly that Gap let's look at the solution so you know that that first part about the Bundesliga Stadium the members of sportex solutions actually collect that event data still quite manually they push it to the Bundesliga data Hub and then it's let's look at our backend solution that runs on AWS as soon as an event is coming in so an event like a card or um a goal uh an AWS Lambda function is triggered that creates a prompt that asks uh Amazon Bedrock via an AP call to create commentary on that event and um this then is we want to provide that then for our Bundesliga ticker editors who watch the game at the same time and right now um compile these sticker entries manually and we want to support them by giving them them these automated um entries and reducing their workloads and have so they have more time for creativity and we push it then out to our bundes league applications how does it look like an action so we have a game here brusa Dortmund is hosting their local Rivals bom it's a draw just 15 minutes to play great pass clinical finish and of course like in every Bundesliga Stadium the crowd is going absolutely bananas so um let's look at the solution with the life ticker in place there's already an entry from the from minute minute 73 and now the pass is coming and then the the goal is scored as soon as the goal is scored there's an automatically created goal card with a player picture included and the context of the goal created so this is all done within roughly 7 seconds it's still way ahead of the broadcasting signal which is important because we need to be fast so that's automation but we also have personalization here and um we have the opportunity to provide this live ticker in basically each and every language that we desire so that's a lot of opportunities and potential for our d2c Ambitions um other than just localization we can also adjust the prompt that it more fits to younger audiences for example so on the left hand side you see the um that we call the Sports journalist style um however on the right hand side you see what we call the bro code or the Bro style and it's more targeted for younger audiences let me be honest we don't know yet if that really resonates with that audience we have to check that but this is basically to to show you what's possible and how we can scale such a such a solution so bringing it all together uh while previously the whole process was a mixture of human intelligence artificial intelligence combined we are confident that we can actually automate that to a very very high extent we still want content moderation and the human flavor in between because we are a passionate product we are we want to be close to our fans but in terms of growth in terms of personalization and in terms of productivity we need the technology to support our teams let's look at example number two it's our data story finder so the data story finder is a solution that provides realtime insights for our B2B customers via our commentary live system so data stories that they can feed to our fans via VIA broadcast this is all done in what we call the commentary live system uh we'll take a look at that in in a bit why do we do that like what's the reason behind it well because both fans and uh broadcasters alike they are looking for data stories they're looking for the deeper story of the game and data now allows us to tell these stories for example via Bundesliga match facts but they don't have the time be it in pre-match postmatch also during the during the match to look for you know these statistics look for the data stories and have the time to actually query them so that's where the data story finder comes in and supports our B2B partners this is our commentary life system so we provide our commentators with uh personal like player data um team data Bundesliga match facts data comparisons so they have everything that they need at their fingertip so as soon as an event is coming in for example in this case it's a goal an funwood scores the first goal of the game we automatically support them with context to that goal so in that case the the commentator can already say hey after leading one nil it's very unlikely for Frankfurt to lose a game because we know that from the data um as as well as exceptional performances or or records we can easily detect in our database because we all access the same structured uh database which runs on AWS you can see here an example of a speed alert from a bun Lega matcha system which is as I said easily to detect for the data how does the solution look like so it's you know I don't have to go through the data collection part again we we all know that by now I'm sure and then we have basically a rule set system that says as as soon as an event is triggered and as soon as we detect a data story like you know there's a goal for example provide context to that goal as soon as you detect a uh a record out of the system someone going up in a in a top scorer list list that is an automatic text provided for the uh for the commentator generative AI comes into play when we think about match summaries so as soon as a um halftime whistle is is happening or full-time whistle is happening this triggers again an aw Lambda uh function which then creates VI an AP API call to bedrock uh a summary based on these data stories plus event data so our common ators can use that post game but also for analysis at half time similar to the AI live ticker um I said we use the same data but also let's be honest um human beings are also capable of uh of um detecting these kind of Records or these kind of data stories is not nothing that is just possible with technology but scaling it and doing it in real time when you actually needed how can you do that when we add 3 terabyte of data each season it's close to impossible for human beings to keep up with that so this is why we think also here we can automate to very high extent and provide these Services automatically in the future what's next for data three things and we want to start with what a chatbot no right we don't want to do a chatbot well we do however want an interface that solves an issue that we don't solve so far because we provide data stories yes but they are always kind of one directional and what if the commentator needs a story from a player that is sitting on the bench you know they need data from him and uh the event data that is a automatically triggering the the stories doesn't provide them with these kind of stories so we want to uh we want kind of a conversational opportunity for our commentators to engage with that data and why just offering to commentators when you can offer the same also for our fans so that's also something that we're after and then there's skeletal data so I'm sure by now everybody here knows 3.6 million tracking points per match which what you think this is going in the future any ideas we multiply that roughly by 40 starting basically next season that's a lot of additional data coming in for performance analytics for health and also for gamification and then there's automation you know right now that's process is right now quite manual we are in the process of um making this more automated using algorithms and II because um this will help us to create more efficiency and we already see that our solutions that we have have in an MVP phase right now provide 95% of accuracy when it comes to the foundational base event data so we are very confident that we can go to a semi-automated solution in the future all right that's it from my side I'll pass over to Bastian for his part thank you very much okay this is football it's just a game 22 players some chasing a ball others stopping in nothing serious this is just a regular weekend and these are just numbers numbers that change when the ball hits the back of the net over and over and over again this is just history being made and this is just footb but it means everything so this is one of more than 10,000 content assets we produce every year to bring the fascination of the Bundesliga to the fans out there in the world [Music] but we want to do more right and we just um said it before we already have this one billion people that we read through all the contents that we produce but we want to reach the next billion of fans and Lucas was just stating it um before to do so we need to be the most fan Centric league in the world because there's one pity of those billions of fans out there there are just a few who really enjoyed the game in the stadium and it's totally different if you have once enjoyed a game in the stadium and felt that passion that is happening there compared to when you sit in front of a glass and consume video content for example or articles or whatever and this is what we need to do we need to bring this Joy these emotions as best as we can to the fans out there so being fan Centric has a lot of dimensions and I'm not going to tackle all of those Dimensions today but there are a few if you want to reach more people you need to think about accessibility right so how can you make content accessible to people it's about channels how many channels do you have but it's also about l language right we know from research that people prefer to consume content in their mother tongue right and that is a threshold for us as of today because most of the content that we produce is in English is in Spanish is in German and then there's just a few contents that we produce in other languages which we then push out through different channels so language is really key to address the next billions of people out there but it's not only about language it's also about localization and let me give you one example a commentator in Brazil is celebrating a goal totally different than a commentator in Germany for example does right I'm not doing it how they do it but you can imagine they shout the world the word goal for minutes right and the German commentator says that was a nice goal great if we now would only Translate speech to speech for example the German commentary in an highlight video to a Brazilian Portuguese commentary it would not be authentic right those people consuming the content in Brazil would think H this sounds a little bit awkward because that's not the way they consume that content right and as I have stated before it's a lot about emotions right football is a lot about emotions and therefore we need to translate this this emotions again and there are other characters to localization which uh I will show you later on so this is really really really key for us when we think about how can we address more people in a way that we can scale it and I think this is a question a lot of you have asked or have heard throughout the last month right and of course there is an AI for that and it's creative ey which can help us to scale content production localized in different languages in the upcoming months so let's start with the broadcast portfolio that we produce of course we produce the live games but in addition to the live games we produce a vast amount of shoulder programming we produce that in English we produce some parts of it in German we produce some parts of it in Spanish and that's basically it if we think again about accessibility localization helping our broadcast Partners in the world to have content that they can put on their Ott or on their broadcast channels it totally makes sense to think about how can we a translate and be localize this content and let me give you some examples so we do have a Bunda highlight show right which is more or less 40 minutes of um highlight clips of those games right they are in English it's easy to translate that stuff in the future in all other languages right which then helps the broadcast Partners to put it on their channels again because they don't have any additional cost to produce it right and of course if we think about Indonesia Bahasa at the moment we don't produce the Highlight show in Bahasa Maybe the broadcast partner doesn't do it as well and just puts it on the channel in English if they put it on the channel if they don't put it on the channel the likelihood is much higher if we just deliver the Highlight show for example in Bahasa right same is the bua weekly show which is a format which somehow um looks back at the last match day and somehow looks into the future what is happening on the next match day so same here right we can just translate that localize that and provide it to our broadcast partners and the third one is the bun midweek interview you can imagine it's an interview where we send an um we send a team with a camera and anything that you need to a player and we do an interview 30 40 minutes maybe in their home right in the future I could think about sending a Word document with questions to that player he and the pr um Department of the club is answering them sending back the word document and I could create a synthetic video out of it where I never ever need to send out any kind of camera team or any content guys to produce that kind of a video format right um and then I can just produce that video in any language that we want to right so that really really helps us to broaden our let's say broadcast portfolio and with that we manage to get better in accessibility for the fans and we can create better stories for them then of course we have digital channels right and of course we do have an app we do have a website we do have a fan fantasy manager and the app as of now for example is in German is in English and it's static in Spanish that's it and set it in Japanese and that's it if I would think about now creating apps in any kind of language of course ni will help us there as well and then I have another accessibility and I have another localization and with the personalization features we were already talking about and I'm thinking more about individualization which is more to me than just personalization we do have apps in any language tailored to every every one of you everyone of you gets his or her own app in your language with the content that you like with the stats that you like and so on so forth and the same is for operat channels right we use social media channels like every Sports organization and more or less every company does use it for the top funnel marketing right so to convince people that the Bundesliga is a great league and it's worth to follow it and we do have social media channels in English and that's it we don't even do it in German and what we do is we do geot targeted postings and contents for several languages but that's it right and here again I don't need to tell the story from A to Z it's the same right I can create any kind of social media channel in any in any language and again I have accessibility top funnel marketing where I can bring the joy of the bundes to those PE people again in their language right so there's a lot of opportunities for us when we think about accessibility language and also localization of it right but once again it's not about translation if you just stop at translating stuff it could work for some pieces but it will not bring the joy to the people out there and then there's digital campaigns we do a lot of brand campaigns right it's about season start it's about the Bundesliga still has a winter break right and we need to tell people that the winter break is over the next match day is ahead right and we are having um brand campaigns around our most important games like the classic it's Bayern Munich against brua Dortmund for this and this is for a year we create those 707 match graphic assets we create those almost 400 non-match graphic assets and we create 846 videos we create this for our own channels but we also have to create this for our broadcast Partners in the specific territories and we also have to create this for our clubs at the moment it's all done manually so every time we do those graphics for B lus for B Dortmund for Bayern Munich there is a graphic designer who is designing those graphics for the clubs if we produce it for ESPN Plus or for Sky perfect in Japan or for any other broadcaster it's also amended by a human being right in the future this will be different right in the future gen will help us to do all this stuff localize it and even localize with other players right so if we now would do a graphic for ESPN in the US of course it makes sense to put in American players that are playing in the Bundesliga right in the future the AI will do that right we just tell them it's America use an American player use the ESPN logo and go for it right and then we can scale in languages we can scale for broadcast Partners we can scale for um we can scale for influencers we can do whatever we want to do and we can do it at scale right at the moment I would have to build up resources and resources for which I don't have the money money at the moment right of course there is a cost to AI as well we know that but it's not comparable build to bringing up people and setting up agencies and so on and so forth so that's a glimpse of where gen will help us to scale content to create more accessibility to localize content to tell different stories as we are doing right now but of course we also thinking about new content formats so how can geni help us to create new stuff less for the topic of accessibility less for the topic of language and localization of course we can put that in there as well but we thought about how can we deal with for example retention and engagement right and what we did we created the Bundesliga stories so what is the bundes Lea stories when I talk about stories it's pretty much the same format that you know from social media right so circles where content is behind those circles in Instagram Facebook and you name it so at the moment we are creating more or less 4,000 articles right they are mostly created for the website because they are used for CU purposes those articles have around 800 words in average right so we have a Content asset of which we thought okay how can we leverage this for example to the app so what we were thinking about to say okay let's take those articles and let's think about how can we use technology to create a new format out of this article that is already there and create the Bundesliga stories for the app and in the end this is what it looks like right we just take the article and with the article as a basis we create those Bundesliga stories and I just tell you how this happens so first we take the backend article right which has the words which has the jsons which has the HDML in there and we normalize this article because there's a lot of information we don't need to create a story then the team does the prompt engineering and you can imagine the prompt is not a prompt that we put into chpt when we want to know get an answer to something it's a very sophisticated more than 100 200 300 lines of a prompt where we we tell especially Bedrock what to do right so we tell them take the article and create 10 slides or five slides we tell them use photos we tell them use citations of the article and so on so forth so we give the instructions with that prompt to the system we create M met metadata almost automatically right there were photos for example in article skills so we use AWS recognition to create metadata of the players that are covered in the photo so that the system knows okay if I'm now creating a slide with Harry Kane there's already a picture from Harry Kane and of course this picture then needs to be cropped right from 16 by 9 to 9 by 16 um and all the other metadata that you need is created there then what I said there's an image selection right because the story is 10 slides and every slide needs to have a photo so the image the the image creation is done and in the end the story is created it just takes a minute and that's it right what we do is we still have the chance to have a human in the loop So within our own CMS which we build on our own um the team created an editor where a human being can still have a look at those 10 slides that were created and if the crop of a photo is not correct the editor can do right and left and take the right sizing for for the cropping maybe there isn't a photo in there because there's a citation of a doctor of a team of which we don't have a photo so the editor can say okay we take that slide out it's just a nine slider or we do something else so at the end there's still the possibility of a human in the loop but to be honest my hope is that pretty soon no more human in the loop is needed because it's the system is so good that we can just fire it out right so what does that mean um you know that the diagram from all um the other things we are doing in the end we take out a lot of human intelligence because it's not needed anymore um and we use artificial intelligence important for us is we never ever do that to cut away people right it's not about cutting away people it's about creating something new with the use of Technology because otherwise we would have not been in the position to create those stories because budget wise we wouldn't have done it right um and even if we do something else like we do AI match reports for example right which is for an editor a quite boring task right it's just writing down what has happened on on the pitch during the 90 minutes right they don't really love to do that we are now doing that with AI as well which is pretty amazing because you don't feel any difference between the AI match report which is done by AI compared to what has been done by humans before and now they have the time to do the stuff they like right so that is what we how we think about it at the moment so and of course we can use tools that are out there to create new contents as well right and here's one example have you ever heard about generative AI yes I have they use uh artist's voice to generate music um that they create on AI today we're here with AWS and they have a new tool it's called Party Rock would you like to create your dream pitch today sure yeah let's do it what about the location says in the mountains at night time dark you see the stars it's a little bit chilly a little bit cold but I'm from Canada so I'm used to it so ooh with the mountains and the skyline obviously with the Stars I I like it I think we get distracted from the mountains you know just long ball you be like wow beautiful and then miss handle the ball hi I'm aonso Davies I'm with buesa and AWS Party Rock everyone can build AI apps so that's one example which we created on our media days so what is the media days the media days is a format where at the beginning of the Season we go to Every club we have a Time window of up to 3 hours where we create content with all the players and all the stuff right it's from photos as you can see to social media content anything that you need content for our Media Partners content for our sponsoring partners and Commercial Partnerships like tops all the tops cards are photographed there and then produce afterwards um we are now able um to bring in remotely editors from Skype perfect from Japan to directly talk to ja Japanese players right they don't have to fly over anymore we can just do it via the system right and then they can tell the Japanese players what they would like to hear from them right as promotional support for Sky perfect for example so that's really really cool that we have those bonus leer media days and of course party rock is one example we all know those tools out there which we can use as long as it's legally fine I would say uh which we can use to create contents and contents and contents is not about only creating content it's also about creating more efficiency in all those operational tasks that there are out out there today so what's next um and what are we doing together with AWS based on the road map that Al was referring to at the very beginning we are now thinking about how we can deal with intelligent metadata creation so where can gen help us to create metadata um and we all know right data is the basis to use AI right and we produce a lot of content a lot of video content which is still somehow locked and needs to be locked and taged right but there is um possibilities and tools out there which can help us to do that in a more efficient way and gather much more quantitative and qualitative data we were talking about content localization we are right now and I will show you an example afterwards um discussing with AWS and doing the first MVPs and Pock which tools to use to do this translation and then the content localization and we start with speech to text and of course um we would love to have speech to speech or even geni commentary live for the live games right that is the holy graay to me if we manag to do that in different languages with all the challenges and hurdles there are out there um that that would be fascinating but we are aiming for stuff like that um then we are thinking about bu ites which is a little bit to help um internal users of our archive and the mum system that we have to find content in a better way right um and we are thinking about how we could deal with push notifications because push notifications is an for us a very strong tool to bring people back into the app right but here again it's about how many people do you have to create editorial push notifications not just system based push notifications and there we thinking about how we how we could leverage J as well right and let me just make one final sentence on that the stories that we had I was referring to engagement and retention and those people in the app who use the stories show much better retention rate and show much higher time spent in app right so the hypothesis that we had with creating this new format for more engagement and more retention and more time spending app uh um was I would say verified uh at that stage this is what we are focusing on it just looks like four points on the PowerPoint slide it's a lot of work that we have to do right to get where we want to get um but the teams are making great progress and as I'm saying I'm happy that we can partner here with AWS because they really help us to get to the next level so here's just one example delivers he's taken the buddist leag of Mike stor his season has been a goalscoring procession across Germany the captain of England ripping up the Bundesliga he scored 36 goals in his first buus Le campaign a feat no player had previously achieved only G Müller and Robert levandowski have ever ended a buus leing campaign with more goals so this is the first test that we did um when we were referring to content localization with content OS how how we call it right um I'm pretty sure every translation that was in there and every localization was 100% correct I didn't check check it at all but the team said it's all fine so because I said is that all cool I will be with that in Las Vegas at AWS reinvent there will be people from everywhere over the world if there is in there people will tell me right so please make it good and I think it is good um so this is a first test as I said it is speech to text right and the Holy Grail is to make it in every language even more language than there speech to speech people not ever feeling that this is a machine doing it it's just a human being translating and local and localization and localizing the stuff that we do so that's it from me um that's it from me a if and Lucas thank you very much for attending um that session and um if you have any questions we will be around uh for a few minutes so thank you very much and all the best to you

2024-12-14 14:14

Other news