AWS re:Invent 2022 - [NEW LAUNCH!] Privacy-enhanced collaboration with AWS Clean Rooms (ADM305)

Aug 16, 2023

AWS re:Invent 2022 - [NEW LAUNCH!] Privacy-enhanced collaboration with AWS Clean Rooms (ADM305)

In this session for developers and analysts, get a first look at how AWS Clean Rooms can help you more easily collaborate with your partners without sharing raw data with each other. Hear from AWS experts and customers on how you can use AWS Clean Rooms to create your own clean rooms in minutes, add participants, and start analyzing your collective datasets. You’ll learn how AWS Clean Rooms helps you protect consumer data and add restrictions on queries run by each AWS Clean Rooms participant with built-in, customizable analysis rules and privacy-enhancing controls.

Learn more about AWS re:Invent at https://go.aws/3ikK4dD.

Subscribe:
More AWS videos http://bit.ly/2O3zS75
More AWS events videos http://bit.ly/316g9t4

ABOUT AWS
Amazon Web Services (AWS) hosts events, both online and in-person, bringing the cloud computing community together to connect, collaborate, and learn from AWS experts.

AWS is the world’s most comprehensive and broadly adopted cloud platform, offering over 200 fully featured services from data centers globally. Millions of customers—including the fastest-growing startups, largest enterprises, and leading government agencies—are using AWS to lower costs, become more agile, and innovate faster.

#reInvent2022 #AWSreInvent2022 #AWSEvents

Content

1.56 -> - Hi everyone and welcome to the breakout session

4.05 -> on Privacy Enhanced Collaboration with AWS Clean Rooms.

8.76 -> We will be discussing AWS's new Clean Room service

12 -> that Adam announced yesterday during his keynote.

16.652 -> I'm Shaila Mathias, Senior Business Develop Manager

19.98 -> for AWS Clean Rooms focused on advertising and marketing.

24.21 -> I'll start this session with an overview of the new service,

27.48 -> what it is, why it was created, how it works,

31.08 -> the use cases will support for our AWS customers.

35.04 -> Next, Ankur Agarwal,

37.05 -> principle product manager for AWS Clean Rooms

40.17 -> will walk you through a demo

41.46 -> so you can see the service in practice.

44.64 -> After we will hear from an AWS customer, Comscore,

48.12 -> a leader in media measurement and analytics.

51.48 -> Brian Pugh, Chief Information Officer

54.51 -> will share how he sees clean rooms fitting in

56.88 -> and solving challenges for Comscore and their customers.

60.78 -> Last, we will wrap up

62.1 -> and share more information on AWS Clean Rooms.

67.054 -> Let's start off with a quick audience poll,

70.26 -> get everyone moving a bit more after three days at Reinvent.

74.52 -> My first question is,

75.96 -> whose company has faced challenges in securely collaborating

80.01 -> on data with entities outside of your enterprise?

83.7 -> And if you can just raise your hand.

86.22 -> Okay, I think I see 75% of hands up.

89.73 -> My second question is, who knows what a data clean room is?

96.63 -> Maybe less hands than that, 50%.

99.24 -> My final question is, whose business has tested

102.24 -> or is currently using a data clean room?

106.86 -> Maybe about 15 hands.

108.72 -> So I think we have the right audience for this session.

112.68 -> I'll begin with the AWS landscape,

115.295 -> the landscape AWS customers are currently navigating

119.04 -> as they wanna collaborate with on shared data

122.07 -> with a vast amount of partners,

123.81 -> but face significant challenges.

127.32 -> Data is siloed across business units within an enterprise

130.98 -> with different standards and approaches being developed.

134.4 -> This creates interoperability and scale challenges

137.61 -> in being able to incorporate past insights

140.64 -> into future business goals.

144.96 -> And companies have an unprecedented amount of data.

148.92 -> According to market intelligence firm IDC,

151.803 -> it is estimated that the amount of data created

155.01 -> over the next five years

156.793 -> will be greater than two times the amount of data created

160.5 -> since the advent of digital storage.

165.06 -> And while consumers value relevant experiences,

168.66 -> companies are faced with challenges

170.7 -> on how to better manage how data is collected, stored,

173.58 -> and used to protect consumer privacy.

178.95 -> As an example,

180.06 -> an advertising and marketing customer came to us

182.91 -> and told us they wanna better understand

185.34 -> their advertising effectiveness

187.29 -> and customer behavior by running analytics on data they own,

192.24 -> combined with their partners data

194.46 -> without either party revealing

196.86 -> or sharing their raw data with the other.

201.72 -> In light of these challenges, we created AWS Clean Rooms.

206.927 -> AWS Clean Rooms helps advertising financial services

210.6 -> and healthcare companies,

212.37 -> easily and securely match, analyze,

215.52 -> and collaborate on combined data sets

218.415 -> without sharing or revealing underlying data.

222.48 -> With AWS Clean Rooms,

224.04 -> customers can create a secure data collaboration in minutes

227.94 -> and collaborate with any other company on the AWS cloud

232.11 -> to generate unique insights around advertising campaigns,

236.1 -> investment decisions, and research and development.

241.59 -> Any AWS customer can achieve the services benefits

245.279 -> through its unique features,

247.44 -> supporting multi-party collaboration

250.53 -> for up to five members in a single collaboration.

254.277 -> Minimal data movement

256.56 -> through direct permissioning of data tables

258.991 -> from a customer's Amazon S3 Data Lake

262.74 -> to the Clean Room collaboration.

265.5 -> Easily configurable privacy controls,

268.08 -> restricting the type of analysis allowed on data,

271.35 -> for example,

272.183 -> only allowing aggregate statistics with a minimum output.

276.93 -> The option to pre-encrypt.

278.76 -> So only encrypted data is used within the clean room,

282.33 -> even when the analysis is being run.

285.15 -> And the opportunity to automate

287.64 -> and integrate AWS Clean Room technology

290.638 -> into existing workflows

292.83 -> to create your own white-labeled clean room.

298.71 -> For the purposes of today's session,

300.54 -> we're gonna focus on how AWS Clean Rooms

302.91 -> can help advertising and marketing customers

305.52 -> given the specific challenges that brands, media publishers,

309.93 -> ad technology and measurement companies are facing.

313.5 -> Let's take an airline as an example.

315.96 -> A brand marketer at an airline

317.91 -> wants to know which of their ad creatives

319.95 -> that ran on the media publisher's platform

322.5 -> led to the most ticket sales.

325.8 -> The publisher wants to provide that insight

328.17 -> back to the airline

329.55 -> so the airline can determine which creative

331.89 -> was most effective in driving those ticket sales.

335.07 -> And ultimately the airline can invest more in that creative

339.3 -> to deliver more effective

341.19 -> and personalized messaging to their customers.

344.49 -> But neither the airline or the media publisher

346.92 -> wants this to come at the cost of consumer privacy,

350.16 -> nor do they wanna move what can be terabytes of data

353.67 -> given the time it will take and the risk in data exposure.

360.527 -> AWS is uniquely positioned to help with this customer ask

364.92 -> with 15 years experience

367.02 -> working with all the critical entities

369.36 -> across advertising and marketing.

372.51 -> Brands, agencies, advertising technology companies,

378.6 -> data and measurement, media publishers,

381.96 -> and divisions of Amazon have data workloads on AWS.

386.79 -> They are using Amazon services

388.86 -> to help with business challenges

390.81 -> around first-party data management,

393.12 -> advertising intelligence, and digital customer experience.

397.8 -> These entities

398.633 -> are already working with each other's shared data,

400.71 -> but that process isn't ideal,

402.87 -> there are compromises as it relates to data privacy,

406.05 -> security and data usability.

409.53 -> One party is sending data to the other

411.6 -> with a legal doc governing usage,

414.33 -> or companies are creating their own custom solutions

417.48 -> which take development time, resources,

419.76 -> and constant upkeep.

422.13 -> Or companies are turning to third parties,

424.95 -> which requires data movement,

426.72 -> increasing the risk of data exposure.

431.927 -> AWS Clean Rooms offers an easier, quicker

435.48 -> and more secure solution.

437.85 -> Let's talk through a high level architecture

440.34 -> of how AWS Clean Rooms works using that airline

443.55 -> and that media publisher as an example.

447.99 -> The airline wants to perform measurement analysis

450.51 -> using their own data and the media publishers

453.33 -> with all parties keeping their data

455.49 -> in their own respective Amazon S3 Data Lake.

459.51 -> In a few clicks, the airline can initiate

462.09 -> a collaboration inviting that media publisher

464.73 -> sending the invitation to their AWS account.

470.19 -> The media publisher receiving the invite can accept.

473.1 -> Once accepted,

474.12 -> they are prompted to associate data to the collaboration.

478.23 -> They can configure and associate any data tables

481.8 -> from their Amazon S3 data lake

484.29 -> specifying their privacy controls,

486.417 -> which are called "analysis rules" in an AWS Clean Room.

491.67 -> Once the analysis rules

493.41 -> including output constraints are set,

495.668 -> the media publisher can complete association of their data

498.81 -> to that collaboration.

501.54 -> The airline can configure and associate their own data

504.761 -> from their own S3 following the same process,

508.74 -> but determining their own specific analysis rules.

513.21 -> Please note that associating data to a collaboration

516.54 -> does not mean you are moving that data

518.79 -> from your Amazon S3 data lake.

521.34 -> Rather, you are giving AWS Clean Rooms permissions

524.82 -> to allow the query to run

526.329 -> as long as it meets the privacy control set

529.199 -> by each data owner.

532.65 -> Once all members in the collaboration

534.81 -> have completed association of their data,

537.128 -> the airline in this example

539.01 -> can start running their analysis,

541.2 -> the output going to their specified S3 bucket.

545.61 -> That analysis can be visualized

547.98 -> or used for further analysis.

555.899 -> AWS Clean Rooms drives business value

558.75 -> for all members in a collaboration.

562.35 -> In my example, it offers the airline interoperability

566.28 -> in analyzing their ticket sales

568.837 -> that's stored in their Amazon S3 data lake

572.07 -> against the media publishers exposure data

574.71 -> stored in their Amazon S3 data lake

577.71 -> without either party revealing the raw data to the other

581.49 -> or moving their data

582.72 -> from their own respective Amazon S3 data lake.

587.7 -> For the airline, this translates to specific insights

591.54 -> into what media placements

593.007 -> or ad creatives are driving ticket sales

596.79 -> to help the airline make more efficient

598.86 -> media investment decisions in the future,

601.44 -> benefiting the airlines and customers

603.33 -> with more personalized advertising.

607.272 -> For the media publisher, AWS Clean Rooms creates a new,

611.351 -> more secure and monetizable offering for the airline

615.84 -> and any other brand

617.34 -> or agency customer of that media publisher

620.19 -> to extract their own insights

622.26 -> without directly accessing the media publisher's raw data.

627.93 -> Although we are focused on brands

630.09 -> and media publishers in this example, AWS Clean Rooms

633.581 -> can provide business value for all AWS customers,

637.5 -> including other advertising and marketing personas.

641.22 -> You'll hear more about the benefits for other entities

644.04 -> such as measurement companies very shortly in this session.

647.61 -> I'm now gonna turn it over to Ankur for a service demo

650.58 -> so you can see some of the reviewed use cases in practice.

657.36 -> - Thank you. Shaila,

659.939 -> Hi everyone, I'm Ankur Agarwal.

662.048 -> I'm a product manager supporting AWS Clean Rooms.

665.889 -> And I'm really excited to show you

668.907 -> AWS Clean Rooms in Action.

671.04 -> For the demo today,

672.06 -> we have an airline who is launching an ad campaign

674.82 -> for its frequent business travelers

676.56 -> on a publisher's platform.

678.75 -> They're trying to glean two types of insights

681.06 -> that requires data from both the parties.

683.79 -> Pre-campaign,

684.623 -> they would like to understand how many of their users

687.831 -> that are business travelers are also active

690.63 -> on the publisher's platform recently, and post-campaign,

694.41 -> they would like to be able to understand

696.305 -> the creative performance

698.25 -> and see which creatives are leading to highest sales.

702.51 -> In order to do this,

703.65 -> they will need their data in Amazon S3 Data Lake

706.681 -> and the publisher will associate impressions data

710.04 -> and users data,

712.05 -> and the airline will associate its ticket sales data

715.68 -> and their customer data from their CRM systems.

721.763 -> To recap what Shaila demonstrated earlier,

724.59 -> we'll have four distinct steps

726.69 -> in how we will go through the demo today.

729.27 -> The first one would be that one of the collaboration members

732.42 -> would initiate a collaboration

733.92 -> and invite the other member to join the collaboration.

737.97 -> Once all the members have joined the collaboration,

741.21 -> they would need to create what is called a configured table.

744.78 -> A configured table

745.77 -> is a first class resource in AWS Clean Rooms,

749.526 -> which holds the reference to your Amazon S3 data source,

754.32 -> as well as contains analysis rules that determine exactly

758.19 -> how your data would be used inside a collaboration.

760.86 -> I'll talk more about it during the demo.

763.98 -> Once each member has created their configured table,

766.627 -> like in this case, we'll have four configured tables,

769.431 -> they would then associate it

771.279 -> with an AWS Clean Rooms collaboration.

776.031 -> Once that association is done,

778.115 -> then the member that has the permissions to query the data

782.19 -> will be able to query the data

783.63 -> using either AWS Clean Rooms APIs

787.08 -> or through the AWS management console.

789.84 -> All right, so then let's jump right into it.

794.25 -> Right.

795.083 -> So the first step is for the airline

797.52 -> to create a collaboration and invite the publisher.

800.34 -> I'm currently in the airlines account in the dark mode,

804.21 -> I'll go ahead and create a collaboration.

806.183 -> I'll give it a name and I will give it a description,

811.545 -> the same description.

814.002 -> Next, I will add the collaboration members.

816.813 -> I'm AirlineCo, and I will add SocialCo.

820.747 -> And what I would need

823.14 -> is the AWS account ID of the publisher.

827.76 -> So I have it here and I'm gonna go ahead and add that here.

832.535 -> For the purposes of this demo,

834.63 -> we have two collaboration members,

836.19 -> but AWS Clean Room

837.36 -> supports up to five collaboration members.

839.55 -> For example,

840.383 -> you can have third party data measurement companies,

842.82 -> identity providers, or even multiple publishers.

845.85 -> Each collaboration member can associate data

848.94 -> and one collaboration member can run the analysis.

851.82 -> So in the next step,

852.72 -> I will select who will be the collaboration member

855.24 -> that will be running the analysis.

857.13 -> In this case, it's the airline that's looking at

859.59 -> extracting the insights from this analysis.

861.87 -> So I'll go ahead and select the airline.

864.9 -> You can then select whether you want to enable

867.3 -> query logging for the collaboration.

869.31 -> If enabled, detailed logs of the queries,

871.59 -> including query text is sent to AWS CloudWatch,

875.79 -> which is the centralized

877.05 -> logging and monitoring service for AWS.

879.57 -> Each collaboration member gets to pick this setting

882.57 -> for their own while joining or creating a collaboration

885.75 -> so that they get their copy of logs

888 -> for the queries that reference their data.

892.47 -> I'll go ahead and enable this and say yes.

895.26 -> Finally,

896.31 -> what I have is cryptographic computing for AWS Clean Rooms.

899.3 -> So this allows you to pre-encrypt some or all of your data

903.03 -> before even associating it

904.41 -> within AWS Clean Rooms collaboration.

907.35 -> If enabled, AWS Clean Rooms will run the queries

910.35 -> on encrypted data based on the parameters

912.96 -> that are selected by the collaboration creator.

916.08 -> For the purposes of this demo,

917.67 -> I will disable this and go ahead

919.35 -> and create this collaboration.

924.06 -> So once done,

925.2 -> I can see that collaboration has been created

927.48 -> and an invitation should have been sent.

929.4 -> I'm gonna go and move ahead to the publisher account

932.76 -> in a lighter mode.

934.186 -> So let's reload this

935.88 -> to see if an invitation has indeed come through.

939.15 -> All right, I see campaign planning

941.149 -> and I open the collaboration invitation.

944.91 -> And I can see that, all the details

947.28 -> around who the invitation is coming from,

950.73 -> whether or not cryptographic computing is enabled,

953.1 -> is query log supported, and who exactly are all the members.

956.67 -> If everything looks good, I'm gonna create a membership.

960.12 -> As a collaboration member contributing data,

962.28 -> I also have an option to enable query logging.

965.1 -> If I enable it,

965.94 -> logs for queries that reference my data tables

968.16 -> will be sent to my AWS CloudWatch service,

970.65 -> which is separate from the one

971.7 -> that the airline had configured.

974.1 -> So I'll go ahead and say yes and create a membership.

978.695 -> So this concludes my first step of creating a collaboration.

982.77 -> We have a collaboration now with two members in it.

986.91 -> I can go and see it in the members tab,

989.43 -> and both the members are there.

992.37 -> You can see that there's no data in it yet,

994.2 -> there's no data that has been associated with it yet.

996.63 -> The collaboration,

998.685 -> the next step for me would be to create a configured table

1003.2 -> using the impressions data that is stored in Amazon S3.

1006.5 -> I'll go ahead and select that

1008.197 -> and I can select the glue catalog

1011.9 -> that is used to generate the schema.

1014.54 -> I'll go ahead and select this.

1016.25 -> I'll look at,

1017.209 -> I can also view the schema right from within the console,

1020.66 -> so I'll go ahead and look at that,

1022.46 -> it contains the identifier, impression state ID,

1025.58 -> all of this looks good,

1028.07 -> and I'm gonna go ahead and move to the next step.

1032.877 -> I also have an option to allow,

1034.76 -> list all the columns in my underlying AWS glue table

1037.363 -> or only allow list a subset of these.

1040.31 -> This allows me the flexibility

1042.23 -> that I can use the same underlying AWS glue catalog

1045.05 -> without having to create a new table

1046.79 -> for every single combination of the columns.

1049.19 -> For the purposes of this demo,

1050.399 -> I'm going to select

1051.899 -> a lot of these including creative ID campaign.

1055.79 -> They look good.

1057.283 -> Impressions ID is something

1058.7 -> that's actually internal to my system,

1060.5 -> so I'm not gonna allow listed for this configured table.

1065.39 -> I'll leave the other details as is,

1066.915 -> and I'm gonna go ahead and create a configured table.

1070.73 -> You can see that a table has been created,

1072.8 -> but it is not yet enabled for querying

1074.87 -> and that is because we haven't yet specified

1076.88 -> how exactly it will be used within a clean room,

1079.7 -> including

1080.533 -> and what type of queries can be performed on this table.

1082.97 -> So we will go ahead and do that

1084.95 -> using the create configure table,

1087.157 -> create analysis rule workflow and I'll.

1092.501 -> So Analysis rules is configured in three steps.

1096.2 -> First you select a query template

1098.33 -> to specify what are the types of queries

1101 -> and analysis that you want to run within a collaboration.

1104.12 -> Second, you configure fine green column level controls

1107.33 -> on the data using query controls,

1109.1 -> and finally you select output constraints.

1113.875 -> AWS Clean Rooms provides two flexible query templates.

1117.26 -> The first one is of type aggregate.

1119.57 -> Aggregate template allows queries

1121.55 -> that output aggregate statistics

1123.14 -> such as counts, sum, averages on collective data.

1128.57 -> This can be used to support use cases such as reach,

1131.21 -> measurement, attribution,

1133.01 -> or even finding the overlap of the user segment,

1136.52 -> which we'll be doing in our use case today.

1140.7 -> The second is of type list.

1143.077 -> List allows queries that output roll level data

1146.487 -> for the overlap between the data tables

1148.43 -> from different collaboration members.

1150.5 -> The list can be used to enrich data

1152.24 -> with additional attributes

1153.47 -> by combining it with a common match key

1155.69 -> or output a list of IDs that can be used for activation.

1159.17 -> For the purposes of our demo,

1160.49 -> we are looking for aggregated insights.

1162.14 -> So I'm gonna go ahead and select the aggregate template.

1167.12 -> So the next step is for me to configure query controls.

1170.06 -> We start with specifying exactly which functions can be used

1173.695 -> on which columns within this configured table.

1176.87 -> For my demo, I know that the airline wants to measure

1179.24 -> the size of the overlapping segment,

1181.19 -> so I'm gonna allow count distinct of that of identifier.

1184.88 -> I can also select other aggregate functions

1186.83 -> such as sum, averages, or some of the other functions,

1191.63 -> but for the purposes of my demo, I only need count distinct.

1196.4 -> Next, I can specify joint controls

1198.83 -> to control whether this table can be joined directly

1202.22 -> by the airline without them

1205.07 -> having to join with their own table,

1206.9 -> or do I only want to allow them to,

1209.416 -> or do I want to require them to have an inner join?

1212.727 -> I want to enable the airline

1214.07 -> to only perform analysis on the intersection of the data.

1216.77 -> So I'm gonna go ahead

1217.603 -> and require a join for this analysis to be performed

1221.51 -> and I will select the identifier as a jointee.

1225.347 -> So go ahead and select that.

1229.28 -> Next I can select which columns do I wanna make available

1232.64 -> to be used as dimensions in the analysis.

1235.46 -> I know that the airline

1236.45 -> wants to understand the creative performance,

1238.19 -> so they'd want to group by the creative,

1240.26 -> so I'm gonna go ahead and enable that.

1242.27 -> I also know that

1243.53 -> they'd want to be able to use impressions data as filters

1246.23 -> in their queries to better attribute

1248.36 -> and have attribution logic.

1249.62 -> So I'll enable that as well.

1253.16 -> Finally, I can also select a custom list of scaler functions

1256.407 -> or allow all supported scaler functions.

1259.64 -> I'm gonna go ahead and allow all of them for this demo.

1264.35 -> The last step in the analysis rules configuration workflow

1267.41 -> is aggregation, is specifying aggregation constraints.

1270.53 -> This allows you to automatically filter out rows

1273.08 -> that do not meet a certain minimum threshold

1276.02 -> for aggregated numbers.

1277.88 -> This can be used to further mitigate the risk

1280.1 -> that information about a small group of individuals

1282.83 -> would be released through the analysis.

1285.08 -> For instance,

1285.913 -> if you group the data by US states for a specific query,

1290.06 -> some of the largest states like California

1291.92 -> may have a large number of users,

1293.66 -> but some of the smallest states like North Dakota

1295.73 -> may have a small number of users

1297.17 -> and you may want to protect that information

1299.127 -> in the analysis.

1301.85 -> So for the purpose of this one,

1303.44 -> I'm gonna use it on identifier and select a value of 25.

1309.08 -> So that concludes the analysis rules workflow.

1313.34 -> I can review all the details

1315.47 -> and go ahead and create configured table.

1322.04 -> All right, I can see that it can now be used for querying,

1324.71 -> which means that it can now be associated

1326.6 -> to a collaboration as well.

1328.46 -> So I'm already on my third step,

1330.26 -> I'm starting to associate tables,

1331.778 -> but before I do that,

1334.52 -> I'm gonna go ahead to the collaboration.

1337.58 -> I am going to say "Associate Table".

1340.91 -> I can select my table that I just created from impressions

1344.575 -> and I can review the schema

1346.7 -> to make sure everything looks good.

1348.26 -> I can see that the joint column is indeed the identifier

1351.895 -> and I'll leave all the details as default.

1356.87 -> I'm gonna go ahead and say associate table.

1358.94 -> In the background, what AWS Clean Rooms is doing

1361.34 -> is creating a scope down IM rule

1364.975 -> to get read only access to this table

1367.61 -> only when the queries will be run.

1369.26 -> At this time there is no data that is moving.

1371.455 -> So we've associated the first table.

1377.66 -> I also created a few configured table already

1380.24 -> using the same analysis rules workflow

1382.7 -> so that I don't have to do it four times in this demo.

1385.31 -> So I'm gonna go ahead and select that.

1387.62 -> I'm gonna show you the users table,

1389.57 -> I can see that it has some attributes, including country,

1392.831 -> the analysis rules specify, identify as a joint column.

1398.57 -> Everything looks good

1399.95 -> and I'm gonna go ahead and associate this one as well.

1404.379 -> And at this time, we are just giving information

1407.99 -> to the AWS Clean Room service

1409.4 -> that this is where you go look for data

1411.08 -> when a query is actually initiated.

1414.71 -> So this association will take a couple of seconds.

1417.08 -> All right, so we are done from the publisher side,

1419.72 -> we created a configure table,

1421.34 -> created analysis rules and associated both the tables.

1426.08 -> So we are ready to go back to the airlines

1428.12 -> and I'm gonna refresh

1430.28 -> to see whether this data shows up in the collaboration.

1434.57 -> And indeed there are two tables in the collaboration,

1438.535 -> and you can see that those two have been associated.

1443.24 -> I'm gonna go ahead and associate the first party data now

1446.24 -> that I have from the airline site,

1447.65 -> including the ticket sales and the CRM data.

1450.8 -> So I have those two tables that I configured earlier

1453.474 -> and we can see the schema here

1456.62 -> and it has a lot of rich details about the price

1459.44 -> and the transaction ID and a lot of other columns

1463.28 -> that would be used into analysis.

1465.14 -> I can see the analysis rules

1466.7 -> and I can see that the joint column is again,

1469.04 -> is of type identifier

1470.323 -> and I also am allowing sum in this case on the price

1475.55 -> so that I can understand the revenue impact.

1479.03 -> So I'll go ahead and associate this table,

1482.12 -> it'll take a few more seconds

1484.7 -> and then we'll be ready to associate our loss table.

1488.48 -> All right, so we have three tables in this collaboration

1491 -> and we'll associate our fourth one now,

1493.34 -> which is going to be the airline CRM data.

1496.22 -> So go ahead and select that,

1498.38 -> and I can see that it has a lot of rich information about

1501.763 -> whether or not the user is of type business

1504.523 -> and a lot of other demographic information about the user.

1509.095 -> Again, this is my first party data.

1511.663 -> I'm still configuring analysis rules,

1513.86 -> but I'm configuring it in a way that is more permissive.

1517.451 -> So I'm gonna go ahead and say associate.

1526.54 -> Should be another second or two.

1531.92 -> All right, so we have our collaboration,

1534.26 -> all the data has been associated

1536.03 -> and since airline is the one that is running the analysis,

1539.6 -> we have a queries tab as well.

1541.37 -> So I'll go there and I can see on the left hand side

1544.37 -> right alongside my code editor,

1546.26 -> I can see that the analysis rules are configured,

1549.92 -> the airline tables do not require an overlap,

1554.72 -> because again, it's my first party data.

1556.7 -> I can see all the information about

1559.61 -> which are the joint columns, what are the dimension columns.

1563.18 -> I can see here that the publisher requires a join.

1567.05 -> And again, I can see all the details of how analysis rules

1570.38 -> have been configured for this table as well.

1573.02 -> The last step is for me to specify the S3 bucket

1575.39 -> for where I will send the output of the queries.

1580.55 -> So I have something called ACR demo that I created earlier,

1584.54 -> I'm gonna select that and we can start running queries now.

1589.225 -> So I have something that I had written earlier.

1594.62 -> So let's try this particular query.

1598.16 -> Why don't we try to

1599.617 -> output a list of identifiers from this analysis?

1603.44 -> So I'm gonna try to do that and I'm gonna try to stay

1606.393 -> by joining the two tables of customers and user.

1611.3 -> As you can see,

1612.133 -> it has been immediately rejected by AWS Clean Rooms

1614.75 -> and that is because the analysis type is a type aggregation

1618.44 -> and we are trying to output role level information

1620.9 -> in this query.

1622.837 -> Let's try to do something else.

1625.79 -> Let's try to query just a publisher's table.

1628.61 -> So I'm gonna comment out the other part

1630.457 -> and I'm gonna try to find the count distinct,

1633.23 -> which is an liable operation on the identifier.

1636.95 -> And I'm gonna try to run this without a join.

1639.989 -> Let's try to do that.

1642.23 -> And when we do that, again, it has been immediately rejected

1647.36 -> and that is because a join was required by the publisher

1650.665 -> and this can only be,

1652.329 -> only the queries that have a join will be allowed

1656.197 -> in this collaboration.

1658.1 -> So let's try something that might work.

1659.96 -> Let's try count distinct.

1662.12 -> Okay, we need to fix that syntax error.

1664.94 -> But as soon as we do that, we run it again

1666.89 -> and we can see that the query has been accepted

1668.69 -> and it's running

1669.86 -> and it's gonna take some time for it to run completely.

1672.14 -> So what I'm gonna do is

1673.07 -> I'm gonna switch to another collaboration

1675.08 -> that has results from the same queries

1677.224 -> and a couple other queries run from earlier.

1681.21 -> As you can see here, this is a number of users

1684.65 -> that are common between

1686.538 -> frequent business class travelers on the airline

1690.8 -> and those who were recently active

1692.18 -> on the publisher's platform.

1693.23 -> So this gives me an idea

1694.67 -> what my addressable user segment size is

1697.88 -> for this particular campaign.

1699.38 -> As you can see your use,

1701.75 -> you can again see all the information

1704.09 -> from the analysis rules window.

1707.111 -> I can also use other types of queries,

1710.09 -> like I can group by the creative

1711.74 -> and I can run the analysis by joining the impressions table.

1716.452 -> And here I can see that some of the value oriented ones

1720.83 -> are not working that well in terms of driving sales,

1723.77 -> but the aspirational one is working particularly well.

1726.558 -> So this kind of helps me inform

1729.329 -> the types of creatives that I want to build.

1732.177 -> I can also, for instance, do the sum of price

1736.19 -> to understand what was the total impact of the revenue

1738.59 -> that was generated as a result of this ad campaign

1741.29 -> by better attributing it through a where clause

1744.44 -> where I'm making sure

1745.28 -> that the impression occurred before the sale was made,

1748.4 -> and I can get more specific there as well.

1751.19 -> So analysis rules really allow you to

1753.26 -> write a lot of different types of queries

1755.24 -> within the constraints of what's defined.

1756.68 -> And because the data is an S3,

1758.36 -> you can really very easily plug it

1760.43 -> into a AWS analytics tools such as Amazon QuickSight

1765.08 -> and SageMaker.

1766.924 -> So here I've plotted the graph of the impressions

1771.62 -> and the sales that they're driving on a daily basis.

1775.55 -> And this really helps me

1777.05 -> better visualize the relationship between the two.

1780.11 -> I can also easily visualize

1781.624 -> how the creatives have been performing.

1784.04 -> I can slice it by daily, weekly numbers

1787.28 -> to understand how the creatives have been trending.

1790.13 -> Assuming the same exposure, I can see that certain creatives

1793.82 -> are doing better than the others.

1795.56 -> So that really concludes our demo today.

1800.863 -> These are just some of the examples

1803.48 -> of how you can use AWS clean rooms

1805.61 -> to extract insights from collective data

1808.148 -> without sharing raw data

1811.328 -> or moving it outside your AWS account.

1815.465 -> We are really excited to put it in your hands

1818.18 -> and hear about all your interesting use cases.

1820.91 -> Although I went over the console today,

1823.22 -> the entire process can be automated

1825.77 -> using every operation that I performed on the console today

1830.18 -> as it will be available through AWS SDK, through APIs.

1834.05 -> And this entire process can be automated using AWS ETL tools

1838.34 -> for automated data ingestion.

1840.68 -> You can use AWS clean Rooms APIs

1843.11 -> to create and manage collaborations and run the analysis

1847.22 -> and you can easily feed it into analytics services

1850.01 -> such as Amazon SageMaker for machine learning

1853.16 -> or Amazon QuickSight for better visualization

1855.728 -> or Amazon Redshift for advanced analytics.

1859.91 -> We are really excited to hear about

1861.26 -> all the things that you would do with AWS Clean Rooms.

1863.708 -> Thank you for being here.

1865.222 -> I would now like to invite Brian Pugh from Comscore

1869.048 -> to talk about industry trends

1870.95 -> and share about some of the ways

1872.36 -> that they're excited to use Clean Rooms.

1878.54 -> - Thank you.

1882.53 -> Am I on?

1885.29 -> Can you hear me okay?

1886.43 -> All right.

1888.14 -> Hi everybody, nice to see everyone here.

1890.48 -> Thanks Ankur, and it's a pleasure to be here.

1893.69 -> Data clean rooms

1895.046 -> are an area that I have a lot of interest in.

1898.91 -> I'm CIO at Comscore,

1900.95 -> I'll talk a little bit about what Comscore is

1903.728 -> and some of the changes that we're seeing.

1906.47 -> I mean, there is massive change happening

1909.26 -> very quickly in the media space,

1912.86 -> and that's where Comscore operates.

1914.87 -> We'll talk about a clean room use case that we designed

1918.156 -> that is very illustrative

1920.51 -> of how we would use this type of technology.

1923.39 -> And then the types of offerings

1926.45 -> that data rooms can help us implement.

1930.44 -> So who is Comscore?

1932.909 -> We are a media ratings company,

1935.84 -> so we provide different ratings

1940.58 -> around how many people are visiting websites,

1943.37 -> what is the total reach of a advertising campaign,

1947.206 -> how many people viewed television?

1949.25 -> We're a digital ratings and TV ratings company

1951.86 -> and also movie ratings.

1953.57 -> And if you read about like what

1955.49 -> the top movies were this past weekend,

1957.848 -> it's probably Comscore numbers that we're using.

1960.52 -> We are in the business of measurement

1964.13 -> and our data's used for things like

1966.77 -> planning an advertising campaign

1969.11 -> or how can I possibly reach the best audience,

1972.23 -> we're a neutral third party that can provide those insights

1975.23 -> between advertiser buyers and publishers

1978.98 -> who are selling inventory

1980.87 -> and then evaluating that was a campaign successful

1984.62 -> that I reached my audience successfully.

1988.7 -> The thing with measurement though, with all of the,

1992.99 -> I mean, it's huge, the most media used to be

1996.41 -> before the internet was TV and radio

1999.5 -> and you didn't need a whole lot of data

2002.25 -> or as much data to be able to measure,

2004.69 -> but there's so much cardinality, so many niche audiences,

2007.84 -> so many different ways to consume media

2010.036 -> that collaboration really is how we work with the industry

2015.49 -> and in order to be able to provide measurement.

2018.916 -> So we know media companies

2021.7 -> capture a lot of data about their audiences

2024.092 -> and they have analytics data

2026.02 -> about people interacting with their websites.

2028.876 -> We integrate data with MVPDs for television,

2032.83 -> and we have billions of measurable events

2036.79 -> and impressions every single day that we ingest to measure.

2041.476 -> And the collaboration piece is absolutely critical,

2045.58 -> but one of the things that's happening

2047.88 -> is more awareness around how that's being done

2052.93 -> and how measurement is happening

2054.28 -> and people don't want to be tracked and things like that.

2056.59 -> And that's really where data clean rooms can help out.

2061 -> We're doing an audience analysis,

2062.716 -> what type of audience brings context to the measurement?

2066.1 -> It's not just counting impressions.

2068.468 -> How many, are you reaching your target audience?

2071.26 -> Are they auto intenders?

2075.092 -> Are they intending to buy a car?

2076.15 -> Are they mothers?

2078.4 -> Those types of things for demographics

2080.56 -> are really important to advertisers.

2082.81 -> And then how do you reach those people?

2084.79 -> Where do they spend their time?

2085.99 -> Where are they engaged?

2087.67 -> Where are they most likely

2088.81 -> to interact with an advertisement?

2092.344 -> The data we ingest is pretty much,

2097.068 -> if you wanna interact with different websites,

2099.22 -> and I'll show an example in a second,

2100.66 -> there's a measurement that's sent to Comscore,

2103.16 -> that is the collaboration.

2105.13 -> We work with different media companies

2106.908 -> and they integrate Comscore measurement

2110.65 -> and that measurement is ingested,

2113.05 -> and we use that to create insights.

2118.87 -> And we combine this with first party data that Comscore has,

2121.9 -> panel data, we recruit people around the US and the world

2127.78 -> to interact with us as part of the panel.

2131.17 -> We offer incentives, we get to monitor the behavior

2134.11 -> and then we integrate that with all this other data,

2136.954 -> which I'll call census data.

2139.78 -> That census data is data we're getting from media,

2142.354 -> from different media companies.

2144.88 -> That is how we provide the context.

2147.1 -> That is the data that we're joining together.

2149.89 -> Today we do that all on our infrastructure.

2154.184 -> But as we put these pieces together,

2157.57 -> there's some sensitivity

2159.04 -> around how that's been done in the past

2160.9 -> and how we need to move forward.

2162.94 -> We believe at Comscore,

2164.23 -> that privacy is a right that should be supported.

2167.341 -> So if there are things that,

2170.29 -> artifacts of advertising technology

2173.08 -> that consumers aren't uncomfortable with,

2175.06 -> how do we adapt measurement

2176.77 -> so that we can provide to our customers what they need

2180.31 -> to be able to do media planning, evaluate audiences

2184.204 -> and respect consumers as far as

2187.082 -> what they would like to have happen

2189.352 -> when they engage with these different media properties.

2194.049 -> So this is the way,

2196.48 -> what we call unified digital measurement works today.

2199.21 -> We have a panel, we can observe everything that's happening

2202.356 -> for integrated devices on that panel.

2205.42 -> It's all opt-in, people join it.

2207.94 -> And then we have our census network,

2210.31 -> which are measurements that are integrated

2212.56 -> all over these different media properties.

2215.23 -> And we actually merge these two,

2216.79 -> we have an intersection of what our panel sees

2220.6 -> and what it comes through the census data.

2222.07 -> It's a tiny percentage of all that data,

2224.59 -> but we get all of the census data as well.

2227.2 -> We add that up, we have methodologies,

2229.66 -> algorithms that turns that into a rating.

2232.6 -> We use our panel

2234.25 -> and other data sources to tell us the context,

2237.07 -> what type of people are interacting with that media.

2241.27 -> And then this is the type of report that we provide.

2244.15 -> The scale is huge, right,

2246.07 -> we're basically trying to measure ratings,

2247.81 -> audiences for the entire internet.

2250.214 -> So that's why we're ingesting so much data.

2253.879 -> We're reporting on 10s of 1000s of websites,

2257.68 -> we're reporting on all national local television.

2260.99 -> And so that's where the integrations,

2263.59 -> a panel cannot do this on its own.

2265.324 -> So we collaborate, we work with our media partners.

2269.26 -> They want us to work with them

2271.51 -> so that we can provide

2272.5 -> the most accurate measurement possible.

2275.29 -> And we're also a neutral third party,

2277.6 -> we're treating everybody the same.

2279.19 -> That's really important as far as media goes.

2283.63 -> Buyers want to trust that the measurement

2286.18 -> is representing everything without bias.

2291.19 -> So use cases for buyers or agencies,

2295.96 -> and advertisers reaching their target audiences,

2300.19 -> evaluating different sites, competitive sets,

2302.62 -> those types of things.

2304.33 -> And expanding their reach,

2305.71 -> like how do they reach the most possible people

2308.5 -> for an advertising campaign.

2310.72 -> And then for publishers that's really around benchmarking.

2314.464 -> What is the size of the audience?

2316.54 -> How do they compare to the rest of their category,

2320.05 -> or who they're selling against?

2321.16 -> How do they show that their audience has a lot of value

2325.99 -> so that they can increase the value of their inventory?

2331.18 -> So what's changing?

2334.459 -> Measurement needs to be privacy forward

2338.53 -> in that things like cookies and mobile ad IDs

2344.02 -> are advertising technology artifacts

2347.74 -> that are commonly used in measurement.

2349.3 -> They're used for analytics, they're used for integrations,

2353.53 -> they're used for counting.

2355.552 -> And Comscore has used those in the past,

2358.822 -> that has been a trust

2360.79 -> or a way to be able to measure these things with accuracy.

2365.74 -> But there is sensitivity, right?

2367.96 -> The consumers, governments have started to realize

2373.33 -> that there is a lot of power in that data

2375.49 -> and it needs to be protected.

2377.86 -> It cannot be misused.

2379.99 -> So not all measurements opt in, right?

2383.56 -> So how do you make it so that we can work with our partners

2387.37 -> and they feel trusted or they feel that the data

2390.516 -> that we're integrating cannot be misused?

2396.07 -> Well, we do have to support collaboration.

2398.59 -> There's so much fragmentation.

2400.6 -> It is impossible for a measurement company

2404.14 -> to measure everything by itself and have it be great,

2407.83 -> collaborations where it all comes together

2410.71 -> because we can see exactly what the media companies see

2414.43 -> and we can integrate that into our measurement,

2416.44 -> and that gives the best possible result.

2420.04 -> All these integrations though, are very,

2422.56 -> there's a lot of friction,

2423.49 -> there's a lot of different integrations

2425.47 -> and we want to be interoperable,

2427.93 -> we want to go where our customers are going,

2429.996 -> but there's a lot of ETL happening, there's a lot of,

2434.23 -> if we're sharing data server to server, let's say,

2438.012 -> we're integrating at a lot of different points.

2442.24 -> And something like a data clean room,

2443.77 -> like you want it to be where our customer's data is

2447.82 -> just to reduce all that friction

2449.11 -> so we're not moving data around

2450.7 -> and having to absorb all this costs.

2453.64 -> And then that's where the interoperability comes in.

2457.328 -> How do we operate where our customers are operating,

2461.02 -> reduce all of that friction.

2465.28 -> So clean rooms can enable these things

2467.32 -> so they can enable Comscore

2468.58 -> to provide the best possible measurement,

2470.452 -> can enable our data partners

2473.05 -> to trust that the data that they're providing

2474.97 -> is safe and protected.

2478.12 -> So the way data sharing with works with Comscore today,

2481.15 -> this is just an example of a pixel tag

2483.49 -> that's sitting on our website.

2484.75 -> We have them all over the web.

2486.731 -> What it's doing is sending

2488.14 -> just a little piece of information back to Comscore

2491.05 -> that says somebody visited Comscore.com

2493.42 -> and there's an http header that has cookies

2497.47 -> and all of those IP addresses and all that stuff in it

2500.74 -> because that's how the internet works.

2503.47 -> And that's also some of the elements

2505.15 -> that we use in the measurement.

2506.98 -> We also work with a lot of companies

2508.57 -> on server to server integrations

2510.34 -> where they're pushing that type of data to us.

2512.89 -> And that's happening in a lot of different ways,

2515.23 -> FTP, S3 integrations, I mean,

2518.855 -> whatever our customers are comfortable with

2521.2 -> because we're interoperable.

2523.878 -> I will go back for a second.

2526.21 -> I mean, if you look at our whole AWS pipeline,

2529.886 -> we're using CloudFront as our CDN

2534.01 -> and we're ingesting everything through AWS

2536.47 -> and just landing in S3.

2538.143 -> And then we're running analytics

2540.07 -> on top of that to create our results.

2542.62 -> So our data is already there, right?

2544.66 -> So and in this case, it's the same thing,

2548.8 -> we're integrated data, maybe S3 to S3

2551.315 -> and then we're then putting that into our data lake

2555.7 -> and then we're creating reports

2557.391 -> or our ratings off of that on the AWS infrastructure.

2564.22 -> So in a clean room,

2566.53 -> if you could think about that server to server integration

2569.62 -> where a company might be providing data to us

2573.01 -> through an S3 bucket, and it has all those things

2576.7 -> that they might not be comfortable sharing.

2579.04 -> You know, there's a lot of data like cookies,

2585.28 -> first party IDs,

2588.07 -> IP addresses that are useful to connect to our panel

2592.01 -> so we can provide that context.

2594.076 -> But if in a data clean room environment,

2597.79 -> we can actually do that merger

2599.571 -> inside of the data clean room.

2602.44 -> And then those sensitive elements

2604.24 -> never need to leave the data partners infrastructure, right?

2611.103 -> We're gonna run an analysis,

2613 -> we're gonna get the pieces we need

2614.35 -> to perform our ratings measurement,

2616.6 -> and then that's gonna be the result.

2620.41 -> So this is how it'll work.

2622.24 -> And this was the test that we designed

2624.67 -> because we have all these elements,

2626.11 -> we have those integrations on websites

2629.17 -> and we have our panels

2630.243 -> so we can find a key that would be the same key

2635.782 -> that we might use with a media partner, right?

2638.47 -> Which would be, let's say a cookie.

2641.92 -> And then instead of us ingesting all that information

2646.51 -> and doing all the analysis all behind our firewall,

2650.38 -> we can join those things in the data clean room

2652.87 -> and then get back what we need.

2654.1 -> Let's say it's an aggregate,

2656.943 -> a good example of the type of aggregation that we would do

2661.09 -> is a media company may have demographics on their side.

2666.55 -> Comscore has demographics because of our panels.

2669.55 -> And we can join based on the intersection of our panel

2673.72 -> with the media company

2675.34 -> and the media company will have their demographics,

2678.34 -> we'll have ours, we can create a matrix,

2680.23 -> which is basically like an error correction matrix

2683.14 -> that we can use to integrate in our measurement downstream.

2686.703 -> That is one really good example.

2689.65 -> And then to get the rest of the measurement,

2692.02 -> it could be sent to us without those sensitive keys

2697.051 -> or it could be provided in aggregate potentially.

2700.33 -> So, and that would reduce the amount of data

2703.12 -> that's being ingested.

2706 -> So this is an example similar to what Ankur went through,

2711.347 -> when I looked at this, if it looked and felt like Athena,

2715.15 -> you know, you're writing SQL,

2716.32 -> you've set up a configuration based on

2719.47 -> which party can have access to what,

2721.667 -> and you can run SQL on it.

2723.97 -> And if I just added a demographic

2726.28 -> to that statement with a group eye,

2728.86 -> the demographic from the partner

2730.291 -> and from our Comscore panel,

2733.36 -> then I could create that matrix I was talking about

2735.67 -> and I never see the keys,

2736.9 -> I never see what the intersection is,

2738.43 -> I don't know which data,

2740.996 -> or I won't need to see all the data from the media partner.

2747.58 -> Or we could create a list,

2749.05 -> we could join our panel to that information,

2751.66 -> we could pull back at the row level,

2753.88 -> at the panelist level information that might be useful,

2757.15 -> but we wouldn't need like the sensitive keys

2760.383 -> associated with it.

2766.06 -> So we look at data clean rooms like a,

2769.756 -> they're absolutely necessary for the future of measurement.

2775.154 -> We want to find solutions like Amazon's data Clean Room,

2779.77 -> which where our customers already are,

2782.11 -> and then we can scale that.

2784.48 -> One of the challenges Comscore has

2786.16 -> is scaling all of these integrations

2788.751 -> and as our customers needs evolve

2792.843 -> around which data they can share or will share.

2796.547 -> And we want them to trust

2798.764 -> that the information that they're providing to us

2801.67 -> is exactly what they're willing to provide

2804.22 -> and it's gonna be used in the right way,

2806.62 -> they can set up the configuration

2808.545 -> and then we can automate ingesting that data

2814.39 -> or the results of that data clean room queries.

2817.93 -> And that will eventually reduce friction,

2820.18 -> we can standardize those things

2821.77 -> and we can approach these integrations

2823.39 -> with literally 1000s of media partners.

2827.5 -> And the types of things we can do,

2830.5 -> I've talked about validating the data,

2832.09 -> the demographic data sets, that's a really good example,

2834.91 -> something the Comscore panel does very well.

2838.21 -> We can standardize and scale these types of integrations

2842.043 -> and then we Comscore can offer things back.

2845.86 -> You know,

2847 -> we can hope with predicting what type of audience

2850.083 -> could be on the publisher site

2851.896 -> by basically sending the data back to them

2854.8 -> that we're comfortable sharing.

2855.97 -> We're also not really willing to share who our panel is.

2859.9 -> That would be bad for Comscore, right?

2862.57 -> But we can share insights back to our customers

2867.07 -> and we'd like to do that, and do that in a way that

2871.06 -> is easy for us and for them, but also provides that privacy

2875.419 -> and security that everybody requires.

2879.01 -> So.

2882.04 -> I'm gonna invite, I just kind of summed all of this up.

2886.81 -> I'm gonna invite Shilah back up to close us out.

2891.43 -> Thank you very much.

2893.747 -> (crowd applauding)

2901.57 -> - Thank you Brian for sharing perspective

2903.85 -> on how you see clean rooms fitting in

2905.83 -> and bringing value to both Comscore and your customers.

2909.318 -> We focused a lot on advertising and marketing

2911.89 -> throughout this session,

2913.03 -> but the use cases for AWS Clean Rooms can extend broadly

2916.72 -> to other industries as well, such as financial services,

2919.69 -> as well as healthcare.

2922.03 -> Before we wrap, let's review

2924.285 -> AWS Clean Rooms benefits for AWS customers one last time.

2930.49 -> Create your own clean room,

2931.9 -> add participants and start collaborating with a few clicks.

2938.05 -> Collaborate with 100s and 1000s of customers on AWS

2942.43 -> without sharing or revealing underlying data.

2947.11 -> Protect data with a broad set of privacy enhancing controls

2951.13 -> for clean rooms.

2953.44 -> And use flexible, easy to configure analysis rules

2957.4 -> to tailor your queries

2958.87 -> according to your specific business needs.

2964.42 -> The service will be available in preview in a few weeks.

2968.08 -> You can get more information on our website,

2970.72 -> the URL is listed here.

2975.76 -> And on behalf of Brian, Ankur and myself,

2977.92 -> thank you very much for attending our session

2980.14 -> and your attention and enjoy the rest of your event.

2983.283 -> (crowd applauding)

Source: https://www.youtube.com/watch?v=YxWYEeEAvv4