AWS re:Invent 2022 - Innovating Ad & Marketing Tech

Please welcome General Manager and Director of Solutions for Advertising and Marketing Technology. AWS. Tim Barnes.

Well, good afternoon, re:Invent! What a week? Well, we've made it to the last leadership session of re:Invent 2022. And I hope you were able to absorb some of the amazing content from our various sessions, workshops. Hopefully, you got to connect with colleagues old and new and hopefully you'll leave with some new ideas to help your company accelerate their business.

My name is Tim Barnes. I'm the General Manager of the Ad Tech and MarTech vertical here at AWS and I wanted to begin with a quick story.

The year was 2006. Coincidentally the year that AWS was founded, but it has nothing to do with this story. And I was working with the agency Razorfish and our CEO Clark Kokich held a company meeting and we were all in the common area and people were milling about and chatting and talking and Clark jumps up on the table with a cup of coffee in his hand and he yells "Coffee!" at the top of his lungs and everybody sort of stops and looks around a little bit like what's he doing? And he get says it again, he holds his cup up and says, "Coffee." And at this point, everyone sort of staring at him and he said, "Do I have your attention?"

He said, "Uh 20 years from now, you're not gonna remember what you did today or what you did yesterday, but you'll likely remember this presentation or this conversation." And that's the point of advertising to break through the clutter and to create memorable brand experiences. And while that's still true today, innovations and marketing and technology have dramatically changed how you plan execute and measure campaigns.

For example, we've seen explosive growth in media content and consumption, creating many more opportunities to connect and reach your customers across new channels like connected television and social media. We saw the rise of programmatic auctions which allowed marketers to efficiently and effectively target their customers but also created amazing technical challenges to run millions of queries per second at sub millisecond latency.

With the rise of data and consumer data marketers had a broader challenge trying to bring all that data together and create a customer 360 degree view to provide more relevancy and op uh optimization. And with all that data, there was a greater need to collaborate across the industry but also a need to focus on privacy and security of that data.

And to address these challenges, we are pleased to announce a new initiative AWS for Advertising and Marketing, which is a comprehensive suite of purpose built services solutions and over 100 and 50 partner offerings. This new initiative, thank you. This new initiative simplifies the process to select the right tools and partners for acceleration and helps customers innovate faster, operate more efficiently and collaborate at scale.

Now, this initiative is a culmination of 16 years of innovation on top of AWS and I wanted to walk back through some of that innovation and give you a few customer examples of how our customers have been able to take advantage of AWS s offerings.

So let's start back in 2006 where we launched the first of our core compute technologies, Amazon EC2 and um Amazon S3. And just a few years later, we follow that up with Amazon uh EC2 Spot instances. And one of the big problems facing the industry was the inconsistent spikes in advertising demand. And the old way of handling that demand in the programmatic world was to build out hundreds of servers scaled to millions of queries per second. But often that load was sat idle along came Amazon EC2 Spot instances which changed the game and allowed the technologies to pay only for the compute that they use.

Now. A good example of this is our customer uh Quantcast who processes over 40 terabytes of data daily for their real time advertising audience insights and measurement technology for over 100 million websites and mobile destinations, leveraging EC2 Spot instances. Quantcast was able to save over 25% on their comparable reserved instances. And our customer GumGum, which is a company that uses computer vision and natural language processing to contextually present relevant ad opportunities processes over 100 terabytes of data per day and 300 billion transactions. And they've seen the savings greater than 62% by leveraging Amazon EC2 Spot instances.

The next major innovation with data and analytics and this is really foundational to advertising and marketing as data is the core of what we do. And AWS is innovated in many, many ways in this space. um with services like Amazon Elastic MapReduce, Amazon Redshift and Amazon Dynamo DB.

Now prior to these innovations, processing, big data was virtually impossible and these services really unlocked the ability to process efficiently and effectively petabytes of data. And these are used for things in the advertising industry like ad measurement, marketing analytics, customer modeling and insights.

One of our customers Axiom uses Amazon Elastic MapReduce to compress the time to run model inferences for over 4500 machine learning models. And they're able to do that in hours where it used to take days and they do that at petabyte scale with 50% less cost than they would do in an on prem data center and Salesforce in 2019, launched their customer data platform, Salesforce CDP providing a single view of customers with real time event processing at exabyte scale. This is a major challenge and Salesforce turned to Dynamo DB to solve this problem which allowed them to not only scale but scale in new regions in weeks versus months.

In 2014, AWS introduced AWS Lambda. Now, this serverless technology empowered ad tech to efficiently handle spiky traffic and focus on applications, not infrastructure. And our customer Nielsen needed to create an event processing engine and data warehouse for their new marketing cloud initiative which processes over 250 billion events per day or 55 terabytes. And they turned to AWS Lambda and Amazon EKS.

Now this allows them to scale up or down from one terabyte to six terabytes of processing per hour and to do that at less than $1000 per day and at 20% efficiency on their compute utilization.

Another major innovation is machine learning which really transform what's possible in advertising and marketing while solving for the inefficiencies inherent in data science. In 2017, we launched Amazon Rekognition which is a allows customers to extract insights from images and video for things like contextual advertising and brand safety. And in 2018, we launched Amazon SageMaker, unlocking machine learning for a array of customers who lacked data science resources in the past and allowed them to easily deploy models for things like audience segmentation forecasting and so on. So less time on data engineering more time on data science.

Now I really like this example. This is a company called TripleLift and they're an ad tech company that focuses on new and innovative ad formats. And they used Amazon SageMaker and Amazon Rekognition to analyze video content frame by frame to find suitable ad placements, flat spots in the video that they could serve. an add on to. The result was a seamless nonin interrupted solution for video product placement and a completely new revenue stream brought to market in under six months.

And FreeWheel is a programmatic advertising customer uh company for publishers. And they had a challenge of forecasting inventory far in advance across billions of ad placements and across multiple dimensions including audience segments, geographies, media types. They leveraged Amazon SageMaker to create a machine learning data pipeline for long range inventory prediction and the result improved the accuracy of those predictions but also reduce their cost by over 50% and their dev cost by more than 60%.

Now, we talked about compute instances and compute innovation earlier in the in the program. but there are several new innovations in this space over the last few years. AWS Global Accelerator allows us to intelligently route traffic across AWS global network background at ultra low latency. And then this is really important for programmatic ad technologies.

And in 2019, of course, we launched our first Graviton based compute instances, unlocking significant price performance as well as reduction in carbon footprint and helping our customers reach their sustainability goals.

Now, our customer Trade Desk uh wanted to open three new bidding sites in global regions and, and they were able to do this with no major architectural changes. They move their entire re uh real time bidding workloads into the cloud in weeks as opposed to months. And with AWS Global Accelerator are able to load balance millions of transactions per second across those regions.

In addition, they are now running their user profile store in the cloud processing over 30 million events per second. And they do that with technology from an AWS partner called Aerospike, which is the no sql data store optimized for real time processing customers like Trade Desk use Aerospike for ultra performing readwrite capabilities.

And last year at re:Invent, we launched a new solution with them making it easier for adtech to deploy Aerospike in their own environment. A few weeks ago, Aerospike announced Aerospike on AWS Graviton instances with a 60% price performance improvement scaled to process over 25 million transactions per second and deliver on energy efficiency helping them reach their sustainability targets.

So the transformation we just mapped out demonstrates how core AWS services have helped our customers innovate in many areas. And our AWS for Advertising and Marketing initiative allows us to take this a step further by offering purpose-built services and solutions for the industry across five critical solution areas:

• Data Management, helping customers create a single view of their customers
• Add intelligence and measurement using machine learning to power insights and analytics
• Add Platforms, helping programmatic ad tech customers buy and sell trillions of ads per day
• Digital Customer Experience, creating meaningful connections with consumers.
• And one of the biggest challenges customer face is around data privacy and collaboration.

Now two days ago, we preannounced the launch of AWS Clean Rooms, a new service to help customers share data and insights and here to share more detail about this exciting new service is Akram Shati bgm of AWS Clean Rooms.

Thank you, Tim. Hello everyone. My name is Akram and I am the GM of AWS Clean Rooms. And I'm very excited to tell you more about this service that we announced a couple of days ago.

As many of, you know, there are hundreds of thousands of data lakes running on AWS yesterday and advertising and marketing customers like yourselves have always been pioneers in using AWS analytics and machine learning tools to store process and analyze your data in the cloud. This includes media publishers and broadcasters, social media companies, data vendors and measurement companies at tech and martech platforms, brands and their agencies and Amazon itself.

All of these customers have their data in the cloud and they've told us that they've been increasingly expecting their data to be interoperable with their partners data in an easy and secure manner.

Take for example, a publisher and a brand, the brand may have conversion data such as store sales and the publisher may have advertising data such as a impressions or clicks. And they both want to generate new insights based on their collective data. But at the same time, they wanna do so while reducing friction related to moving and copying data back and forth, which today takes months of work only to get one collaboration up and running.

More importantly, they also want to do so while improving data protection so that they can only share the outputs that are needed for their business use cases without necessarily sharing the underlying inputs and exposing the data with one another.

So I'm excited to share that we announced AWS Clean Rooms, a new analytics service that allows multiple customers to easily and securely analyze and collaborate on their data without sharing the underlying raw data sets with one another.

AWS Clean Rooms was built to make these collaborations easy at scale customers who have been using homegrown solutions or third party platforms have told us that they wanted something that can easily scale programmatically so that they can include it into their existing solutions and applications.

They also told us that they want something that operates on their data natively in the cloud so that they don't have to maintain and copy their data and you have yet another copy that, that they store in another system. They also told us that they want um a system that allows them to have multiple collaborators in the same clean room so that they can collaborate with 123 and even more collaborators in the same clean room.

AWS Clean Rooms is built such that we invest in the advanced data collaboration and and data protection access control so that you can focus on the business analytics and the business use cases without having to re invent the wheel and do the research and write the code and innovate in these privacy enhancing data access controls.

The service includes a broad range of data access controls including query restrictions that allow data owners to have fine grained controls on how their columns are used. Such that the data analyst can write queries just like they do today. And the service automatically detects whether that query is allowed or not based on the analysis rules defined by each data owner, no need to write any code or to define any complex systems to do. so.

The service also has output constraints so that each data owner can have output constraints such as minimum aggregation levels so that the output rows that do not meet these aggregation levels are automatically redacted and only the outputs that are compliant with those minimum aggregation levels are returned to the analyst.

The service also has query logs so that data owners can audit the queries that are run on their data. And one feature that I'm excited about is uh is cryptographic computing that allows customers to pre encrypt their data before even uh setting it up into the aws clean room so that the data is encrypted even when queries are right on top of it.

So these are some of the examples of the privacy enhancing data access controls that come with AWS Clean Rooms. So customers in advertising and marketing have been asking for solutions like this for use cases that span the entire campaign life cycle. This includes campaign insights, audience insights and planning, as well as measurement and attribution and of course audience building and audience generation for activation.

Customers also use clean rooms for uh data enrichment so that you can add attributes to your data without necessarily sharing the entire data sets with one another. There are a number of customers and partners that are excited to use AWS Clean Rooms. You'll hear from some of them today and I'm very excited uh to see what you all do with this new service.

Um and thank you very much for helping me.

Thank you, sir.

Thank you, Akram. Joining us to talk about how migrating advert advertising workloads to the cloud over the years has helped them better innovate and serve their customers. is Anna Nikonov CTO, an elect omnicom.

Hello everyone. I'm Anna and I lead technology at Analect Omnicom created a act to reimagine the future of marketing at the crossroads of data technology and analytics. Today. I'm extremely excited to share what we're building and how we're building it. I will highlight some of the newer technical architecture that we're using and some of the newer tools that we're adopting like the clean rooms. My talk is going to be cooking show style. I'm first going to present the final product and then I'm going to talk about the ingredients and the methods of preparation.

One of the main products of Analect is Omni an open operating system that helps marketers orchestrate better outcomes in a nutshell. Omni is a suit of proprietary application that can be easily integrated with clients and partners first party data and technology to deliver marketing orchestration across all key workflow stages. The tool fall in four main categories of work, insights, planning, activation and optimization for insights to allow explore lifestyles and behaviors of people to create marketing strategy planning tools. Let users translate the insight into tactical plans from decisions on media vehicles to creative and messaging activation tools deliver hyper personalized creative variants across channels

And finally, once the campaigns are in flight downstream performance reporting and algorithmic multitouch attribution allow users to achieve best ROI on their plans. So the platform does a lot to give you a sense of the scale of the platform. I want to share some of the key statistics.

We have more than 40,000 users who collectively build more than 80,000 audiences per year across more than 400 layouts. But the numbers that I think this audience here would appreciate more as a technical statistics that summarize the technology that is required to run this platform.

We just about 11 petabytes of data, just our audience data consists more of 16 quadrillion attributes that we need to process at a rate of 5 500 million per second to create a seamless user experience for constructing audiences. And we run about 400 lambda functions to ship constructed audiences and plans across the workflow as our users plan and optimize to run these operations.

We heavily rely on AWS products across the stack from caching and networking for our web applications to computing and data processing for our data assets. The platform is pretty big. So to provide a proper technical overview, I want to zoom in into one specific tool audience explorer.

The main purpose of the tool is to construct custom audiences like people who drink smoothies, the tool gives you the answers to questions like do they make smoothies at home or do they prefer to go to trendy green bars? Which fruits do they shop for? Do they play VR games? The answers to those questions then allow the users to decide on custom strategy that allow one to be very precise and targeted with communication without wasting large media and advertising budgets to serve this kind of insight audience explore as a tool needs to be very data heavy and it needs to work pretty fast.

We currently evaluate audience sizes at about 500 million attributes per second, which requires a lot of compute power. After the audiences are created, the data doesn't stay static within the tool, it needs to be shipped to other activities within omni for planning where to find those audiences and reporting on how this audience is performing. Every audience object consists of granular data where every row stands for the anonymized individual and their behavior. And the meta data was the overall information who created this object uh audience object when and what's it called to achieve this high performance for users? The tool doesn't call the database directly but it does. So through a layer of data publishing API s that access compressed and pre index data for fast performance.

Before being published, the data undergoes various stages from from landing in omni environment to being preprocess in the data lakes and getting finally into the data warehouse. As you can see from this diagram, we definitely duplicate data and it's not the best architecture. But what if we could use this data without the need to copy it from and move it from our partners?

We all know the issues with data divergence and data refreshes and data storage. This is not wishful thinking because the technology already exists and it's known as clean rooms. The purpose as clean rooms is to allow mutually agreed partners to do collaborative data analysis without moving or copying data or even disclosing sensitively personally identifiable information.

Each partner brings their data set without disclosing full data to the partner agrees on specific queries that are allowed to be run once the data is joined and then get out of the clean room. The aggregate insights that they need wid analects have been using clean rooms for a while and i'm extremely excited to be on stage with aws team to join on the them on the launch of a was clean room wid a elect have been huge fans of the simplicity and the ease of use of this capability.

The setup to configure the contract and doing the handshake is very easy and straightforward to say nothing about the configuration convenience for those of us who store a lot of data assets across s3 buckets and red shift tables, what it means for us and for the architecture economy, it means that now we can leverage aws clean rooms to exchange data with our partners without moving it around or trying to set up sensitive pi i data environments.

The outputs of clean rooms can be used for data enrichment, insights, activation and measurement. I hope i was able to give you a flavor of how clean rooms provide a more technically efficient data architecture to do marketing orchestration in a privacy compliant way. Thank you very much for your attention.

Thank you, Anna. Amazon Ads runs advertising workloads at scale on AWS and collaborates with agencies like Analect to create audiences, provide reporting and attribution. And here to talk more about that is Tony Krier, Senior Principal Engineer, Amazon Advertising.

Hi, everyone. I am Tony Krier, a senior principal engineer uh coming from our Boulder Colorado development center. I joined Amazon seven years ago and I've worked in Amazon Ads for each of those seven years. Today, I am really excited to walk you through the Amazon Marketing Cloud. What most everyone just calls AMC and how AMC will utilize AWS Clean Rooms to help advertisers answer important questions about their advertising on Amazon Ads.

Now, Amazon Ads helps brands of all sizes design and exper add experiences that delight customers around the world and deliver meaningful business results. With hundreds of millions of active customer accounts worldwide, as well as first party insights based on shopping streaming browsing signals. Brands can confidently craft relevant campaigns that enhance the customer experience.

Our bread of solutions at Amazon.com combined with crowd favorite services like Twitch Freebie and Amazon Music, make Amazon Ads and amplifier for brands to reach the right audiences in the right places, both on and off. Amazon properties.

Over four years ago, I helped to design and implement AMC so that advertisers could answer new questions they had about their advertising efforts. Now, advertisers already had and still have excellent reporting capabilities on Amazon Ads and these capabilities help them understand the effectiveness of the advertising and better use. Amazon Ads, product offerings but as the complexity of advertising has continued to grow, we learned that our advertisers wanted more customizations on how they viewed advertising performance.

For instance, using AMC Spark X and Ad Tech agency created a geographic analysis for Energy. And that analysis enabled Energy to better advertise the renewable energy products. It is to enable advertisers and their partners to answer these kinds of questions that led us to build AMC. And of course, all the reporting that is run in AMC only returns aggregate and anonymous outputs, maintaining our long standing commitment to customer trust.

Since launch AMC has grown very quickly. In the past year, we've seen a four fold increase in active users which translates to thousands of marketers getting the insights, they need to make complex business decisions. We currently manage many thousands of AMC instances, which means managing many thousands of AWS accounts to give you an idea of the complexity involved from a network traffic perspective.

We transfer dozens of terabytes of data each day and for each distinct AMC instance, we manage over 30 cloud formation stacks, deploying over 900 resources across two AWS accounts. We do this over many AWS regions where deployments are automated using a very complex state machine.

AMC provisions, a dedicated AMC instance to an advertiser that is isolated from all other instances for each advertiser, their advertising events such as impressions and clicks are placed into their AMC instance. And advertisers then use AMC Sequel, which is a dialect of Postgre Sequel to create customized reports that are anonymous and aggregated.

One of the nice things about AMC is that advertisers can optionally add their own signals into their AMC instance, which allows them to answer ever more complex questions about their advertising. When an advertiser adds their signals into their AMC instance, that information stays in the instance and can only be used by that advertiser using AMC to combine Amazon ad signals with their own.

Advertisers can create reports that they couldn't before. As an example, marketing agency Cognition, Digital recently helped multiple automotive industry advertisers to add their CRM sales and service signals into AMC. They observed a 60% overlap between Amazon Ads impressions and their dealer CRM information using the overlap. They were able to establish that their media was performing at around $300 cost per vehicle sold.

This is a big improvement, a 71% reduction in cost per lead month, over month and an overall 80% reduction in the cost per vehicle sold.

AMC. In its simplest form is a signals router to send Amazon Ads pseudonym my signals to each AMC instance along with an API surface. Advertisers use to interact with their individual instances. An advertiser uses the API surface to send pseudonym my signals to and to receive aggregated and anonymous outputs from their AMC instance and to configure recurring reports or to create an ad hoc report.

Advertisers use the API surface to schedule queries for processing the reporting. API is also used by the Amazon Ads console for advertisers managing their queries from there under the hood. Though AMC is a single tenant application, each advertising customer gets their own unique instance. Information pertaining to and originating from the advertiser is isolated and separated from any other AMC customer where advertisers use a dedicated single tenant API to add signals into their AMC instances.

After query is processed, aggregated and anonymous query results flow out of AMC and into a preconfigured S3 destination that is chosen by the advertiser. The AMC orchestrator manages the health of every instance deploying and updating the many cloud formation stacks that comprise an AMC instance and the MC orchestrator is also responsible for ensuring that AWS services don't become overloaded by acting as a gatekeeper that spreads compute load over time.

However, a single tenant application model introduces challenges that require continuing heavy lifting to provide AMC's features to our customers. On the one hand, the ability to query event level signals across media sites, inventory formats and conversion channels implies moving advertiser scope signals to each AMC instance. This process is complex and brings us close to AWS scale limits.

On the other hand, AMC also allows advertisers to add their own signals to their AMC instances. This feature requires our advertisers use developer resources and operational resources to add those signals. Advertisers must also develop manage and maintain applications that ensure their signals in AMC are fresh and current.

Finally, each AMC instance is a self contained application. And while the deployment of one instance is straightforward performing deployments across thousands of instances is complex. Cloud formation is a great tool but managing tens of thousands of cloud formation stacks is kind of hard.

So AMC brings unique value to advertisers by offering unified signals and enabling custom analytics across a wide range of topics. But the single tenancy model does provide technical challenge from an operational as well as software engineering point of view, not only for Amazon and software engineers, but also for those of our advertisers as well as i mentioned before.

Customer trust is foundational to how Amazon Ads builds things using AWS Clean Rooms. AMC continues to have the same privacy guarantees it as it does today. Query results are always aggregated and anonymous and as an added benefit using an AWS Clean Rooms collaboration now allows the collaborator signals to remain stationary, adding signals to an AMC instance will no longer require Amazon Ads or an advertiser to move data around instead signals from both Amazon Ads and an advertiser will be offered to an AWS Clean Rooms collaboration and now just like Amazon or sorry. AMC instance works today. Only AWS Clean Rooms is permitted to query over the signals in a given collaboration. And also like in AMC instance, works today, queries running within an AWS Clean Rooms collaboration will be performed securely within the collaboration boundary.

Advertising customers will get the benefit of joining their own signals with those of Amazon Ads. But with the simplicity of using anybody's clean rooms, data access controls, they're using AWS Clean Rooms.

AMC will get to be able to rely on AWS to handle much of the single tenancy concerns and enabling Amazon amazon Ads to focus on feature improvements rather than instance management.

So in the upcoming AMC architecture with AWS Clean Rooms built in, we no longer require the advertiser to add signals into an AMC instance. Instead, both Amazon Ads and the advertiser will offer signals to the ABOS Clean Rooms collaboration and then AWS Clean Rooms will manage processing queries within the collaboration.

AMC plans to leverage a clean rooms to enable advertisers to participate in clean room collaborations with Amazon Ads in 2023 using AWS Clean Rooms will remove the need to build custom solutions to add signals to AMC. This reduces an advertiser's operational overhead and the reliance on developer resources.

AMC with AWS Clean Rooms will accelerate the advertiser's time to collaboration while still providing the same protections that AMC does today. As AMC transitions to using AWS Clean Rooms, the complex single tenancy environments we manage now will be able to be replaced with a multi tenancy model that is more efficient.

Each collaboration is single tenant nature, but compute resources managed by ADBS will be better utilized. Finally, the AMC orchestrator will move to managing AWS Clean Rooms collaborations and it will become AWS Clean Rooms responsibility to effectively and efficiently process query requests.

Thank you and we are looking forward to building solutions on top of AWS Clean Rooms in the very near future.

Thank you, Tony. And finally here to talk about an added intelligence and collaboration story from a publisher's perspective is Fox. Please welcome to the stage, Lindsay Silver SVP Data Platforms, Fox.

Alright. Thank you all. I'm really excited to be here. It's a privilege to watch this, this unfold. My name is Lindsay Silver. I'm the head of engineering and commercial products at Fox. Watching that timeline at the start made me remember back to the first time i deployed an advertising pixel which was around 2008. And the my first attempt. This was truly an attempt instantly crashed the blog network of sites that i was running the pixel on. I had to frantically try to learn the AWS cli tools to actually bring things back up and, and figure it out.

It's really neat to watch how far we've come since then. Fox is the company i work for. It's a conglomerate of several media companies.

Um, we are as, as kind of a cooper. We work as a corporate function in between all of these groups and we're constantly building collaboration between them. And so when we think about an platform for powering our advertising, for powering commercial data, we're thinking about how to build a platform that allows these businesses to work together.

Uh the scale is substantial. Uh we're at about 20 to 30 million data points per second across our platforms. Uh we have a substantial number of about 500 million uh users every month who are viewing our content uh across our own sites, our media players, our third party partners uh and we're creating thousands of pieces of content uh per month. And so the platform that we have has to analyze that content and also allow us to collaborate and work together on it.

Uh, so what do we do here? I thought I'd start with a couple of examples of things that our data teams have done before talking a little bit more about the platform that powers this.

Uh, so hopefully you're all watching the World Cup right now. Uh in the US, it's provided by Fox. Uh, most of us are watching it as a second screen on our, our monitors while we're working. Uh our teams are fortunate enough to get to think about this from the data standpoint.

Uh, and one of the really interesting features I think that we launched this year powered by Amazon and, and some of the ML models and processes that Fox had built on Amazon is our Catch Up feature. Uh, and so for the first time when you turn on a game, uh, now if you turn it on in the middle of the game, you probably do because it's running at, at 3 to 4am in the morning. Uh you'll see a Catch Up that is built based on advanced machine learning models that are looking at the game and extracting moments in the game that are of high importance and combining those using Amazon technologies to deliver a replay instantly for that point in the game. It's the first time we've done it, it's a really cool feature. Uh it's running across all of our platforms as of today.

Uh, the second thing we're doing with data that I think really lends uh lends itself to collaboration and, and how we think about collaboration as a data platform. We're using insights that we have that our partners have to deliver Heads Up uh insights to our sportscasters and production teams. So uh the teams on the ground in Doha and around the world are watching uh heads up displays that are powered by ML that are extracting moments in the game in real time and taking and finding historical insights based on those. So they're giving insights about the player who has the ball, matchups that happen as they happen, any of the, the stats that are happening around the game happening in real time, uh it's the first time we've done that and for soccer and it's a really exciting product that we have. It's also delivering promos to the game. So any of the highlights you're seeing printed up on screen are delivered via AI and, and the systems, we've built really neat set of products.

Second of all, we're jumping uh again to uh one of the other sports that obviously Fox is well known for our NFL market, uh teams or market products. We're delivering every game. There are 14 to 16 games on a Sunday. Uh every Sunday, our teams historically have had to make those decisions about those products uh manually. So we have a team of analysts and schedulers who are looking at all of the match ups, looking at the competitive data every Sunday and trying to decide what to play this Sunday afterward.

Uh this year for the first year, we're using AI and we're using models that we've created in partnership with Amazon to deliver insights and forecasts for these games in real time leading up to the, to the day of the game. So we're actually making a decision on the 420 six or some markets that we have in real time changing the game and you'll actually see us improving the, uh the projections throughout the weeks coming up to the game. And so there are potentials where we switch depending on how a team is performing or depending on how much engagement a matchup has, wouldn't be possible without our Amazon partners and the processes and systems they've given us.

Uh, so we're a content company. Um, obviously on the news side and, and in general, news is a hard topic for advertisers. Uh, brand safety becomes the most important thing. Um when I came to Amazon a few or to Fox a few years ago, one of the things we were missing was uh a deep understanding of our content, both our, our video content as well as our print content that we had.

Uh we invested heavily over the last couple of years in building out what we think is, is the next generation of contextual platform within the industry. So we're taking content, we're building an index in real time of our contents that streams through our systems. All of our video is tagged um at a sub minute uh level in, in all of um and in entities in all kinds of factors over about 100 and 50 attributes um per point in time. And then we're building models throughout time for brand safety and for other uh contextual attributes for our clients.

This is something that takes a huge amount of processing. Um it, it's an expensive process, but we found that it's actually massively worthwhile for a company like Fox to have. Again, it wouldn't be possible without some of the Amazon products that it's built on.

Uh next, we're a large company that relies on our audiences. And so we've invested again in, in building audience systems that allow us to pull in that data um across Fox. And so we're building out our own unified audience pixel uh with a unified idea across the Fox platform. The trick to that as you, as you all know is that when you're dealing with an ecosystem like ours, where you have several different businesses, several different PnLs and several, several different privacy policies, all of these systems in order to collaborate, have to work within those constraints. And until today, we haven't had good tools tied to our infrastructure to do that. So we're really excited about Amazon Clean Rooms. For this reason, we believe that it can help us with identity. I believe we can help us with a lot of the tools that, that we need to build these types of collaborative systems for audience analysis.

So really quickly on our approach to data, um Fox takes a pretty simplistic approach to data, but I think actually it's, it's very powerful. Um we focus first on our domains. Uh this is a simplified view of our domains. But as a media company, content audience and customers are domains, we create content audiences react and engage with that content, customers buy those audiences or look for those audiences.

What we found over time from a data standpoint is the most valuable assets that we have are those second level assets that relationship between each of these things. How does our content relate to our audiences? How do our audiences relate to our a or to our commercial products? And how do our commercial products fit in with our content to build experiences?

What we've discovered more recently and what we we're aiming toward as we go into the next generation or the next chapter of products that are built on data is to build products that optimize each of these domains based on the relationship between the other two domains. And that means that uh our experiences, our audience experiences are built on the relationship between content and our commercial products. Uh that means that our content distribution decisions are built on how audiences respond uh to commercial products within those experiences. And that relationship we think is a recipe for building out a platform that really powers the next generation of Fox products.

The other part of this is a company that spans or several companies internally uh is that we need to decouple access and governance from that those layers of the infrastructure that allows us to act. And so the other approach, the other philosophy we've taken uh as a platform is we're really focused on how do we decouple these things. And it's interesting when we listen to the talks about Clean Rooms and these concepts that a Clean Room is really an access layer, right? It's a layer on top of all of your data on top of your data infrastructure that allows you another level of control.

Uh we used to have table level control and then row level control or column level control and then cell level control. Now you have probabilistic control, which means that your partners, even your internal teams know to some extent how accurate that piece of data is. And that's an immensely powerful dimension to add to these systems.

So right now when we look internally, we've built in a world and we're building toward a world where we're completely decoupling access from the underlying infrastructure and data sets. I think the next generation uh will be one where we're actually giving probabilistic control to that.

I decided to go against the grain and not do the big logo slide here. Um if anyone wants one, I can send you, I'm happy to send one to you offline. This is the rough architecture of our systems. So uh going right, right to left, we have several sources that flow into our systems both and this is in the hundreds or thousands of ETLs, both streaming and batch ETLs that flow into our systems. Uh we have real time ingestion layers and, and fire hoses that are ingesting that data and then bringing it to our leg and uh through our processing and into our data warehouses. Last step is our query interfaces. That's that access layer i talked about and that's where uh either human or programmatic consumers of that data have access.

Uh it's obviously more complicated than this, the point that I wanted to make here. I think that it is different than a lot of the way that people approach. This is that we're trying to give access to our users of this platform at the level they need access and that's extremely important. So um users who need access to metrics or highly governed data should have access to it. If they can't get that access, they should be able to drop down a level into the semantically accurate data, drop down a level into our data lake. Um if they really can't get the ax answers they want and only then should they start the process of building a new source or building a new process?

Um what I think again, that that Clean Room bring to this is they bring that next level to this. They allow second party partners to have that discussion as well. Um I think this is probably too simple to even bear repeating. But all of the point of designing a system like this is to make to break down uh the process of creating feedback loops.

Um if you think about the exponential nature of companies like Google or Facebook in its in its heyday, these companies built and designed an automated systems that made themselves better and more profitable and grew based on data and insights. And so we're aiming to do the same thing. We're aiming to automate these processes and it takes time and, and money and effort to do it. Uh we're sure thankful for partners like Amazon because they make it a little bit easier for us.

So what's next? Uh I want to give this picture. I've mentioned this a little bit, but I think it's really important to say that when we look at companies like Analects partners on our demand side, partners within our ecosystem or even within the Fox ecosystem internally, what we're asking ourselves is how best to take and use this access layer that we have and share data across those. And until today that men extracting data, pulling a data set that we knew was cleaned and handing it over to that partner.

Now, we're starting to think about building infrastructure and architectures that include that second party layer within them. And I think within the next year to two years, you'll see companies like Fox offering a surface to our partners that includes that probabilistic layer uh or that Clean Room layer.

Now, these are most of the, the things that you've heard all week, but I think it's worth calling them out again. Data co ops and marketplaces are becoming a very interesting conversation within the supplier demand world and in advertising right now, we're starting to think about what are the assets that we have that contribute to those?

Um obviously advanced audience activation is interesting and hooking these systems directly into activation platforms. Uh they ad servers or SSP partners uh that we're working with are things that we're working on right now.

Um measurement is the most obvious thing that people shout about Clean Rooms. I think it's a valuable part of, of the consideration, but i do think it has to go hand in hand with, with activation and we, and you should be asking of your partners. Hey, what are you doing for activation ting to tie into your measurement?

Uh and then internally, we obviously are working to build a spine around identity across Fox. I think that these Clean Rooms like Amazon's offering will help us immensely in that uh our new insights for our publishers we're looking at at how do we share second party data across other sports castors, other publishers, this type of interface allows us to do that. And then obviously our uh data collaboration in in general, being able to collaborate among our business units in a way we haven't before is is key.

So uh with that, I'll turn the stage back over, but I'm really excited to be here and thank you all for the time.