Increase the Lifetime Value of Media Content with AWS Media Intelligence Solutions

welcome uh to this aws talk about media intelligence solutions my name is chris kudun i'm the head of worldwide business development direct to consumer here at ews and with me today is michael shields a general manager of advanced advertising at triplift a great partner of aws in this talk we're going to discuss how to increase the lifetime value of media content while reducing costs with aws machine learning services so let's dive right in audio and video data is growing exponentially all around us in the u.s alone consumers have over 300 video services to choose from last year or like the last 18 months or so we saw disney plus hbo max peacock and more launch and surprise surpass their subscriber goals in record time but it's not only these new and exciting and expanding platforms that drive the exponential demand for content it's also like all those businesses who don't have media or video as their business but use audio and media and and video to drive their business to advertise their business uh to drive interest for their business and so forth all these dynamics uh definitely accelerated over the last 18 months or so but already the time before so and it's really driving this exponential growth of of audio and video overall and as already mentioned more and more industries are relying on media for better engagement so examples here are education 92 percent of educators think uh that video increases the student satisfaction uh and also the learning experience 84 think uh it's increasing the achievements of students so it's really making an impact there obviously you know again most recently it has been mandatory to offer education through video or audio uh then there's ad tech where 86 percent of businesses are using video as a marketing tool and it's working you know 84 of people say that they have been convinced to buy a product or service by watching a brand's video and i already mentioned media but you see here also a statistic that shows that really it's exploding the amount of up video that had that's being created that's uploaded and being consumed a couple of overall trends they are as well notable like you know consumers don't necessarily own or want to lean back and passively uh consume video but they're expecting enhanced new experiences they're expecting personalized experience they're expecting additional information coming with the video they're expecting interactivity opportunities you know chat uh potential e-commerce opportunities in the video that they can use uh when they consume the video there's also uh really an expanded reach necessary for these videos these videos uh or this this content uh is being consumed globally so there are a lot of regulations to be followed but also uh the accessibility of the video is absolutely crucial for um you know for example for special special needs groups but also you know for for you know language and and so forth the next one would be there are it's very important it's very expensive to to to produce this content uh it's it's often the most expensive item of these companies to produce these kind of uh to to produce quality content so customers and and and companies are looking into ways of monetizing into new ways innovative ways of monetizing content more effectively so to sum it up you know consumers expect more and organizations really need to accelerate the innovation to differentiate themselves and maintain profitability yeah so let's go a little bit deeper there because the the way how this can be achieved is for example through machine learning and you know organizations need machine learning solutions because you know creating and traditional media management media creation and media delivery workflows are having a couple of issues they are like very manual and time consuming uh they are subjective and error prone when you're reviewing videos again and again for many many hours you know it could create errors they're expensive and they're really not scalable that well and like machine learning can help with these issues for example it can help lowering the cost by automating the very repetitive parts of this process of review processes for example by for example pointing out where there might be issues in the video and then have them reviewed on like have have these areas reviewed specifically uh machine learning can help boost the user experience you know consumers it it's just really really difficult to you know go into every piece of content that's being created go deep into there and see what's in every like frame for example there like what objects are in there what's elaborate design there right machine learning can help to surface those and make the whole archives and the the whole content more discoverable that way uh machine learning can also help with compliance and brand safety issues there are certain uh um you know regulations across the globe but also for brands individually like they don't want to be associated with specific content there's obviously regulations to certain content not to be shown at certain times of the day or in general so that's uh machine learning can help to identify issues there and then it can drive it can drive revenues by like you know using machine learning to find you know better areas for example to place ads and and so forth we're going to talk about that much more in this conversation quickly like working through how ews actually enables developers and and users potential users of machine learning to use machine learning services through our aws ml stack very high level it's three layers and each layer can be used by you know specific target groups here you for example on the bottom layer it's really for the deep experts petitioners of machine learning that are comfortable with building their own models training their own models then employing them and maintaining them the middle layer is an integrated development environment amazon sage maker that really puts everything that you need to manage your machine learning pipelines and workflows behind one pane of glass and then on the top layer this is the api layer where you use these services by calling an api and getting results like for example with vision uh you you're bringing a piece of content a piece of video you use you call an api the recognition api and you get you know which objects are in that content as a result back you know same for audio use text tracked you know extract that kind of create a transcript of the content so in this talk we're really focusing around the ai layer uh and i have also marked there you know mainly around vision recognition uh speech poly uh and transcribe and then on the text side like comprehend and translate and we are already outlined in a little bit but like we're looking into these four key use cases you know search and discover discovery uh subtitling and localization compliance moderation and content monetization i want to also point out how you can use aws media intelligence solutions to solve these sad use cases there's really two flexible approaches on the one side there's the media intelligence partner network so this is for customers who want to leverage the deep expertise of our apm partners for media intelligence these partners have a lot of experience building these kind of use cases that i just outlined that are gonna you know name those partners in in the following slides and then on the other side or in conjunction with the first approach is our media solutions these are for customers and partners who want to build integrated solution uh using frameworks uh starter uh architectures and and concepts designed by ews and then kind of built from there so you can go either or you can obviously go the combination there so let's type dive into these specific use cases and as already said search and discovery is like definitely a very big one if not even the biggest use case there uh and it's really like show using machine learning to help users easily find the content they want um so it's like used for content classification but also it's not just like finding the video specifically but also you know using audio transcription for example to generate subtitles but then also to to find you know where a certain statements made within a content right so enhance the search uh it's used to to to to faster generate highlights and previews of content if i find the specific highlight faster uh if i'm looking if there was a news event happening and i want to go into my archive and find you know all the highlights that you know are related to this specific use event uh you know if i don't use machine learning that could be a very lengthy task and that cannot could potentially not be reacting to the use news event fast enough but also the metadata being generated here can be used for personalization right so he brings in a lot of a much richer understanding of what the content is and if i relate that back to the viewer and how they're reacting to that specific content i obviously can inform a personalization engine and and kind of you know create a bit of a flywheel of more and more tailored uh uh content personalization for the users you see partners here uh that are really experts uh and in using these services uh and we're working very closely with to enable these workflows and also you see the the solutions that are relevant in in this area uh in order to make that bring that alive i'm now uh switching over to a demo where i'm gonna show you how this looks in in real life okay so here we see the demo we uploaded a commercial here that is running and on the left side you see after running these services you see the objects that have been identified you see person uh and there's a lot of person in there so you see a lot of red dots so maybe take another one so couch for example you see red dots uh identifying the the areas in the video that have like in this case a couch on it you also see a percentage there um that indicates the uh confidence core that this is actually a a couch other things here as well sunglasses headphones and so forth so these are all the uh objects that have been identified with a confidence core higher than ninety percent you see celebrities there's another api from the computer vision service you see multiple celebrities here identified same concept of moderation this is a commercial there's no results obviously on the moderation side faces words is interesting you see basically all the words that are being identified on the on the screen so let's say uh we go to amazon uh identified here on on this alexa device uh beside the vision results you also can see the transcript here from the audio uh translation look for key phrases and so forth so i'm not going through all the details here obviously but uh you can imagine this these are all inputs into like a potential search uh that actually you know support then this search and discovery use case okay after this we go back to the presentation okay so now that we saw that in action uh let's go how uh and talk through a couple of examples how customers have been using these technologies uh to optimize their workflows so here for example you see nfl who significantly improved the speed of their content search uh you also see uh like a partner of ours synchronized uh that uh are you working with a key european customer and they're saving 50 of the editorial time spent creating thumbnails so that's like a specific use case right if you want to build if you want to identify the right frames within the video that could be the most interesting to then use as a thumbnail you can use these tools as well right identify potential areas and then review it and and use those as a summary you can even personalize the thumbnails that way and then a partner of ours promo uh who are reducing their promotion creation process from two days to two hours significant improvements in like you know again analyzing the assets they're dealing with in order to generate promo material the second use case is subtitling and localization improve the reach and accessibility of media content this is really all around creating transcription and translation so you know using machine learning to transcribe the content uh to make it accessible through like subtitling uh and so forth uh and then also translating the content here in over 70 plus languages and again making that also accelerating that process of transcribing and translating the content this is really important there's a lot of content out there that's aging very fast uh the if if you are uh having a news event if you're having a an event of like interest that's like very time sensitive you just don't have the luxury to take very long time to create subtitles and then like having them translated the quicker you can get these things out but it's also not only news events if you have like an educational piece you know you want that thing out there as fast as possible so that you know you get also the return on investment on this asset you obviously can reach a wider audience you know by making it accessible but also making it uh you know translating it uh for new regions and so forth and there's obviously all the regulatory requirements uh for transcription and and and captions and so forth um again here you see two key partners that are working with these solutions and also the solution that would help you to get the poc started or to get like experimentation going there here a couple of customer examples and you see that like one of our partners worked with a customer there with allianz and reduced the subtitling uh the costs of their subtitling production by eight times uh nascar used these kind of tools and reduced the costs by 97 when it comes to to to their subtitling uh uh workflows and then you have all the other partners uh that like making learning assets much more valuable and accessible overall that has been as already mentioned in other verticals especially in education a lot of interest for for these kind of tools and then compliance and brand safety so this is really to review media content against predefined or or custom criteria to ensure compliance and also other requirements like brand requirements and so forth so again you're using rich metadata to automatically identify inappropriate content in general objective content uh but also to for example uh check if like certain brands are placed properly if the logos are placed at the right place or also if if the placement of the logos or the brands is objecting to any uh you know communicated rules by the brands right also within the advertising but also within is in generated content so really using machine learning to automate that process also using machine learning to automate the qc process overall uh and again save time and cost to do that as the the the volume of content is growing exponentially again here's a couple of great partners we're working in this area with and then the solution you could use to get started okay and now to counter monetization so this is the fourth use case i already outlined the first three and for this specific use case i'm to hand it over to our partner triple lift and michael is going to walk you through the amazing experiences they're building for their viewers for their customers using the tools that i just outlined in the area of content monetization okay michael take it away thanks chris and thanks to um everyone in the aws audience we're super excited to be able to showcase to you a variety of new ad products for television that we've developed in in partnership with aws so today i'd like to introduce you to triplelift and a variety of ad products that we're creating for the future of brand supported television so first of all i'm michael shields i'm the gm of advanced advertising overseeing our efforts in in developing new products for television and before i tell you a little bit about those products just an overview of triplift as a whole so we are a late stage venture-backed company we've had tremendous growth uh in recent years and triple lift first entered the market to develop uh native advertising and really pioneered the marketplace for programmatic native advertising we utilize a computer vision-based technology for peering into ads breaking them down into components and then building them back up into native environments across the open web we have evolved that technology over the years to look at other lines of business and specifically create ad products for other forms of media so we launched a business and branded video and then branded content and then in 2019 started assessing the marketplace for connected television and tripolis focus is not on just enabling how advertisers can transact how they could buy advertising but really defining the what for any given media and we're going to show you a variety of ad products that we're creating for the future of television but first a little bit of a why we're focused on this area so you know streaming has evolved tremendously over the last couple years 2020 was really a banner year for the evolution of of the streaming wars if you will pre to 2020 you know streaming was concentrated in a handful of large services major media companies major television companies struggled to reach consumers uh uh through streaming applications notably launching a variety of tv with everywhere applications that required authentication um but in 2020 all of the majors regrouped and we had as chris referred to we saw the launch of services from disney plus to peacock to hbo max all of which are in the tens of millions of subscribers now and as as of this recording i believe disney has just reached 100 million uh barrier um and if you think about like what this means for consumers right i think that the average number now is consumers are subscribers to four different services but they're also having a lean back experience around television content in new areas of streaming including on free ad-supported television services like pluto tv and 2b and others that have spurred a lot of m a activity in television in the last couple of years so the industry's going through a major change and the eight minutes per half hour of television advertising that per was persistent across television networks of the past and traditional linear environments that eight minutes per half hour is not going to be tolerated by consumers right who are increasingly gravitating towards s-vod platforms where there's very little advertising opportunity for brands or avod platforms with significantly lower commercial loads so tripolis ad experiences are aimed at responding to this trend in consumer behaviors and really providing a solution for the future of brand supported television and indeed our solution is focused on not what's in break not what's in commercial breaks but what actually occurs in show and attempting to automate integrated ad experiences in show indeed this is already happening across a lot of major services but in a pre-production and unscalable way so you may or may not know that services like netflix and hulu and amazon prime have a lot of brands in the programming that they bring to market indeed the the uh market for product placement in television advertising in television shows is already an 11.4 billion dollar market and that's within the 100 billion dollar global market uh for for television advertising as a whole but there's no automation there's no technology for that existing product placement marketplace so we're going to show you a variety of tools that we've created to develop integrated ad experiences at scale based on a series of of different amazon technologies several of which chris just referred to the different products that we're bringing to market include yes standard ctv spots or standard commercial breaks but those commercial breaks paired with integrated experiences that appear in the show so we utilize the traditional lower third space used for promotion to offer contextually relevant dynamic overlays we trigger experiences what we call the inaction six or split screen ad units um at lulls in the action or natural breaks in pro and programming that appear in show and then we render a variety of 2d and 3d assets in post-production to superimpose elements of product placement without any requirements of the show producer so without any kind of pre-production concerns and we deliver each of these experiences dynamically so just as you would expect in streaming environments with standard commercials to receive a different commercial a different message from an advertiser based on who you are as a user we're now delivering these different integrated experiences based on the profile from the audience leveraging traditional programmatic advertising technology and i'll give you a quick view of what these experiences look like so first if you're a sports fan you've seen the inaction 6 or split screen unit proliferated across sports but the process for launching these units and sports is still very manual oftentimes there's an individual in a mobile production truck that's actually triggering the experience and sending a note to the commentator with the help of amazon machine learning technologies we're doing the analysis of the video looking at a variety of elements including doing speech to text doing natural language processing on the closed captions all with a view towards finding the right moments to trigger these experiences so we can actually find that player substitution or pitching change we can find a natural break in the action that's relevant for uh triggering these types of ad experiences we utilize the same machine learning tools to find the right moment in the context of a particular show to utilize this lower third space traditionally used for promotion but now to launch contextually relevant advertising right so we can trigger these dynamic overlays at the right moment not interfering with the narrative of the show and providing a contextually relevant moment uh for bringing an advertiser into the show i mentioned before that we're also utilizing computer vision to find images and to locate particular surfaces where we can render 2d and 3d assets from an advertiser in post-production so this is a much more scalable way of doing product placement if you think about some of the concerns that advertisers have had to meet in doing product placement in the past it's very difficult to get physical product on set it's very difficult for an advertiser's marketing budget to align against the production schedule for the show right but by rendering these 2d and 3d assets where we've located uh particular surfaces right we don't require any additional resources of the show producers right and we can deliver these dynamically with impression level tracking so different audience members see different experiences a much more scalable much more efficient way to do product placement to do this correctly we actually use a variety of amazon technologies and there are really four key elements to our solution one is the video analysis so i mentioned before some of the tools that we're using we're utilizing computer vision based video analysis to recognize images and to recognize surfaces both on the background of a shot to do things like replacing a back a billboard in the background of the shop we're also doing um speech to text so that we can translate all of the spoken words in a show into text that we do then do further analysis on we do natural language processing on both that transcribes text using utilizing tools like amazon transcribe and then amazon comprehend for doing natural language processing on that and on the closed captions we combine all of these different mode methods of machine learning analysis to do what we call total scene analysis so this is an attempt to replace with artificial intelligence the very high touch integrated marketing function of pre-production product placement finding the right moment to trigger these ad experiences in show we've also invested in a variety of tools to have the right assets from advertisers advertisers actually don't have a lot of the right 2d and 3d assets they need to do this sort of thing at scale and triple if now has assets in our system from 100 of the ad age top 100 advertisers we leverage other aws technologies notably aws elemental where we integrate with play out and server side ad insertion system so that we can leverage the existing ad infrastructure for delivering these experiences for those of you that are familiar with online advertising we deliver all of these experiences as simple vast responses so the 6 or 8 or 10 or 20 seconds of edited video that we've made to insert a product into a show we return that as a vast response and stitch it together to the content stream utilizing tools like amazon's media tailor and then the fourth piece has nothing to do with technology but we've led our business development effort by engaging hollywood studios production companies directly to make sure that even down to the show runner that we have their buy-in for the consumer experience that we're offering and now i'll give you a piece by piece slightly deeper dive into the technology so i mentioned part of our machine learning tools is utilizing computer vision it analyzes so that looks for surfaces and analyzes for images all with a view towards where we can do the rendering of 2d and 3d assets product placement in post-production all of these experiences i mentioned before are rendered dynamically so we utilize the same space for rendering multiple brands and again this doesn't require any additional resources of the show producers i'll show you some work we're doing with nbc universal that really showcases the programmatic technology so once we've identified a given place for a brand to come in we can then rotate different brands based on the audience profile of who's watching the show and again we're doing this machine learning to find the right moment to insert things like that in an action six that you you just saw right so again this was a very manual process uh in sports but our machine learning based technology can find things like a closing credit sequence so that we can standardize the rendering of some advertising experiences like the inaction six or dynamic overlays we know what's going on in a particular scene we can analyze mood and tone uh so that we can match the right advertiser at the right moment and again based on what's actually being said on the screen and of course we can utilize some of the tools that chris introduced before for finding brand safe moments right so that we know the right contextually relevant moment for rendering an advertiser but we also make sure that they're coming in at the right place so we utilize these machine learning tools not just for finding the right moment to include brands into a particular shot but also to exclude them from these moments and this is a key concern uh for the advertising industry so you know we recognize moments that uh of potential violence or if we can actually take requirements from a show producer so for instance if there's like a alcoholic character and a show right we wouldn't want to make the render any alcohol brands right so we can take those exclusions we can also take exclusions for where existing talent has uh endorsements to make sure that um you know advertisers from that particular category don't flow into the show as a result we are really uniquely in a position to engage publishers about what the future consumer experience should be for television so we know that consumers are requiring lower ad loads but when we go to publishers we're not just talking to them about how they can use data or targeting to enhance the value of their existing advertising inventory we're really talking to them about how they can provide a better consumer experience so they can go from three ads in a break to two or one but replace it with integrated advertising experiences in the show we're giving more opportunities for marketers to engage audiences we're providing a better consumer experience with lower avalos and we're helping to drive yield for publishers so it's really a total solution for the future of transported television and i'll show you what that future is going to look like so here's some work that we're doing with our our partners at warner media and specifically a show tacoma fd where we rendered some of our ad experiences so you know you might start off a show in the opening sequence by seeing a sponsored split screen ad then as you move into the show we again find all the opportunities to insert 2d and 3d assets in post-production so we start to have a variety of brand integrations into the show they're all dynamically served and impression level tracked then we then find the contextually relevant moment for offering something like a dynamic overlay and of course this applies to the key use case that allows us to then lower the ad loads across the entire stream right so we have fewer standard commercials but we have a variety of these integrated experiences again rendering 2d and 3d assets in in post-production without any additional resources of the show and it really opens up the windows for marketers to get involved right because now they don't have to have physical product on set they don't have to fit within the production schedule and you see over the course of the stream that this creates a variety of new advertising experiences so here we are having far fewer commercials and providing a better consumer experience and yet we're allowing partners like warner media and other streaming services to increase their monetization opportunities across the show as a whole right and through the closing sequence where again utilize machine learning tools to find things like uh the credits in the show and then trigger the last ad experience so you really get a sense of what the future of brand supported television is going to look like i mentioned before that we utilize tools like amazon media tailor as well so within the elemental suite um we leverage existing ad infrastructure and specifically server side and insertion framework so that we return each of these 6 or 8 or 10 or 20 seconds of edited video right as a vast response and then we stitch it together with the content stream so that different users see different experiences but advertisers are given all of the same level of targeting and reporting that they would expect with standard commercial bricks here's a view of our product as a whole and all of the different touch points that we have with amazon technology so first we actually receive shows up front from uh television companies so that we can do our video analysis and often times they're giving us that content securely through tools like amazon s3 buckets then we use a variety of these machine learning tools that chris referred to earlier including recognition sagemaker comprehend and transcribe to do that total scene analysis so that we can replace with artificial intelligence you know that high touch integrated marketing function we're utilizing uh uh oftentimes partners are using tools like amazon amazon elemental for their playout system and we actually set new markers in the manifest associated with a show with a video file for our specific triple lift markers so that it can call out to us and trigger the ad experience and then as i mentioned we're stitching together that ad with the existing content stream utilizing tools like aws media tailor so um amazon's really been a key partner to us in the development of this this product line and we're looking forward to increasingly aligning with them on go to market as we we build the product further early efforts have really helped to reinforce our product development effort right so some of the early campaigns are showing that consumers prefer this experience that they have a positive shift in brand perception associated with the show i might say that across the campaigns we've run we've seen a 46 positive shift in brand perception against only a two percent uh negative shift um which is rarely seen in advertising so really i think consumers are choosing this as a future model and of course it's working to drive marketing outcomes and upper funnel marketing metrics for advertisers including uh significant double-digit increases in things like brand recall brand awareness and preference right so we really think that we've cracked the code here for the future of brand supported television um we again thank our partners amazon for the suite of tools that have helped us to build this product and for the opportunity to engage the aws audience and thank you for attending great thanks so much mike it's really fascinating what you guys are building there uh and uh really interested how you help customers monetize the content better and using machine learning to do so so thanks a lot okay and i'm just going to spend another minute or so to uh give you some pointers on how to get started with all these things we have been talking about through the last half an hour or so here again you see the technology partners that we are working with uh that you know are really deep specialists on on those workloads you see consulting partners that can help you uh build workflows uh that are like very tailored for your need and you also can see the solutions and you know open files so like open source frameworks that you can use to get started you also have the location where you can find all that information on on our website so hope that makes it easy for you to get started and and also to to find the right partners for your workload yeah and that was it uh again thank you very much for your time and for your attention i hope uh this was interesting and yeah wanted to thank you and wish you a great rest of your day thank you very much bye

Channel: AWS Online Tech Talks

Views: 202

Rating: 5 out of 5

Keywords: amazon web services, aws, captioning and localization, cloud computing, content moderation, content monetization, media intelligence, webinar

Id: DP6SROsa6QQ

Length: 37min 47sec (2267 seconds)

Published: Mon Apr 26 2021