Conceptualising Data Clean Room use cases

00:09:46
https://www.youtube.com/watch?v=QJzPgkgrlPg

Resumen

TLDRThe presentation discusses how data clean rooms (DCRs) enable one party to leverage another party's data to gain new insights about their own users. This key concept unlocks a range of use cases including addressable identity and audience activation, data enrichment, and predictive modeling. Addressable identity allows brands to discover new audience attributes for media targeting by using data from media owners or providers. Data enrichment involves leveraging partner data to enhance first-party customer information, aiding in product development and media planning. Predictive modeling in DCRs involves using ID graphs to improve data set match rates, allowing for robust data analysis. Although DCRs do not create entirely new use cases, they facilitate existing ones, providing secure, privacy-safe access to valuable first-party data, which was previously difficult to obtain.

Para llevar

  • 🔑 Key Concept: Data clean rooms enable insights using another party's data without exposing privacy.
  • 🎯 Use Case: Brands utilize data from partners to enhance audience targeting.
  • 🔍 Data Enrichment: Partner data enriches brand's customer understanding.
  • 🧩 ID Graphs: Improve data match rates in clean rooms for better collaboration.
  • 📈 Predictive Modeling: Create or import models to analyze matched datasets in DCR.
  • 🔐 Privacy Safety: DCRs ensure insights are gained securely without exposing personal information.
  • 🤝 Collaboration: Facilitates first-party data collaborations previously hard to achieve.
  • 🗝️ Unlocked Use Cases: Enables addressable identity, audience expansion, and more.
  • 🔗 ID Crosswalk: Links different user IDs for identity resolution in DCR.
  • 📊 Impact Measurement: Brands measure ad effectiveness using media owner data in DCR.

Cronología

  • 00:00:00 - 00:09:46

    The presentation discusses the concept of data clean rooms, emphasizing their importance in allowing one party to use another's data to gain insights about their own users, unlocking new use cases. It explains that data clean rooms are not for activating another party's data but for using it to reveal insights about one's own data. The speaker uses an analogy of stone tablets from Indiana Jones movies to illustrate this concept. Different use cases are discussed, including addressable identity and audience activation, where brands can find new audiences using media owner or data provider data. There's also mention of data enrichment and insight generation, where brands enhance their first-party data with demographic, psychographic, or behavioral information to gain a better customer understanding. Key points include the importance of identity resolution using ID graphs and the utility of data clean rooms for advanced analytics and predictive modeling. The speaker concludes by reflecting on the past challenges of accessing first-party data from publishers, which data clean rooms now help overcome, marking a significant shift in utilizing data collaboratively.

Mapa mental

Vídeo de preguntas y respuestas

  • What is a data clean room (DCR)?

    A DCR is a secure environment where one party can use another's data to derive insights about their own users without exposing private information.

  • How do data clean rooms facilitate addressable identity and audience activation?

    Brands use data from media owners or data providers to discover new audience attributes, aiding in media targeting, audience expansion, and more.

  • What is the role of an ID graph in a data clean room?

    ID graphs improve match rates between different datasets by linking user identities, allowing for effective data collaboration.

  • Can data models be deployed in data clean rooms?

    Yes, data scientists can build or import models into DCRs to analyze matched datasets securely.

  • Do data clean rooms create new use cases?

    While they don't necessarily create new use cases, DCRs allow existing use cases to be executed in a secure, privacy-safe manner with first-party data collaborations.

Ver más resúmenes de vídeos

Obtén acceso instantáneo a resúmenes gratuitos de vídeos de YouTube gracias a la IA.
Subtítulos
en
Desplazamiento automático:
  • 00:00:00
    all right so next let's think about how
  • 00:00:02
    to conceptualize different data clean
  • 00:00:04
    room use cases so what's on the screen
  • 00:00:07
    right now what's on the slide this might
  • 00:00:09
    be the most important concept from this
  • 00:00:13
    entire presentation so if you end up
  • 00:00:15
    forgetting everything that I've said
  • 00:00:17
    everything that I'm going to cover later
  • 00:00:18
    please remember this part okay so data
  • 00:00:22
    clean rooms allow one party to use
  • 00:00:25
    another party's data to reveal new data
  • 00:00:28
    points about their own users
  • 00:00:30
    all use cases are ultimately unlocked by
  • 00:00:33
    this concept so if you're using a data
  • 00:00:35
    cleaning room you're not actually adding
  • 00:00:39
    or activating the other party's data
  • 00:00:41
    what you're doing is you're using the
  • 00:00:44
    other party data to reveal insights
  • 00:00:46
    about your own data and with those
  • 00:00:48
    insights you're then able to unlock new
  • 00:00:51
    use cases so it's like one of those
  • 00:00:53
    stone tablets from like an Indiana Jones
  • 00:00:55
    movie that reveal the truth when the Sun
  • 00:00:58
    Shines on it at the right angle at the
  • 00:01:00
    right time of day the tablet is your
  • 00:01:02
    data and the Sun is the other party's
  • 00:01:05
    data um probably not the best example
  • 00:01:08
    but I think I think you understand what
  • 00:01:10
    I'm getting at uh so let's go through
  • 00:01:12
    some use case categories to illustrate
  • 00:01:14
    and for our purposes we'll assume the
  • 00:01:16
    perspective of a brand being the party
  • 00:01:18
    that's setting up the DCR that said most
  • 00:01:21
    of the these use cases can apply to
  • 00:01:24
    Media owners and Publishers as well okay
  • 00:01:27
    addressable identity and audience
  • 00:01:29
    Activation so so here the brand wants to
  • 00:01:32
    discover existing or new audiences for
  • 00:01:34
    media targeting and or measurement at
  • 00:01:36
    scale party a here would be the brand
  • 00:01:38
    and party B could be a media owner or it
  • 00:01:41
    could be a data provider and in these
  • 00:01:43
    cases brands use the data provided by
  • 00:01:46
    the media owner or data provider to
  • 00:01:47
    reveal new attributes about their own
  • 00:01:49
    users or customers which can then inform
  • 00:01:52
    Downstream auding Spas use cases like
  • 00:01:54
    Discovery targeting expansion look like
  • 00:01:58
    modeling or segmentation
  • 00:02:00
    and it's worth noting that the benefits
  • 00:02:02
    here can work both ways and that a media
  • 00:02:04
    owner can also take these insights to
  • 00:02:06
    better understand what audiences on
  • 00:02:08
    their properties can drive more impact
  • 00:02:10
    for the brand
  • 00:02:12
    partner next we'll look at data
  • 00:02:14
    enrichment and inside Generation Um so
  • 00:02:17
    here brands are looking to enrich their
  • 00:02:20
    existing first-party data to learn more
  • 00:02:22
    about their customers to improve
  • 00:02:24
    everything from product development
  • 00:02:26
    creative messaging uh and media planning
  • 00:02:29
    and here party a would be again be the
  • 00:02:31
    brand party B could be anyone with value
  • 00:02:34
    adding first-party data and in this
  • 00:02:36
    category brands use the collaborating
  • 00:02:39
    partners data to enrich their existing
  • 00:02:41
    customer data to reveal new data points
  • 00:02:44
    whether that's demographic psychographic
  • 00:02:46
    behavioral uh Behavioral or any other uh
  • 00:02:49
    consumer attribute and that uh this
  • 00:02:51
    allows them to better understand their
  • 00:02:54
    customers and optimize uh things like
  • 00:02:56
    identity resolution so moving more
  • 00:02:59
    toward it's a holistic single customer
  • 00:03:01
    view uh and worth noting that identity
  • 00:03:05
    resolution requires the use of one or
  • 00:03:07
    more ID graphs which we'll cover more
  • 00:03:09
    later in this section and this can help
  • 00:03:12
    with inside generation enrichment and
  • 00:03:14
    modeling as well and then attribution
  • 00:03:17
    measurement and optimization so here
  • 00:03:19
    Brands want to better understand the
  • 00:03:21
    impact and returns generated by their ad
  • 00:03:23
    spend so party a again would be brand
  • 00:03:25
    and party B uh would be the media owner
  • 00:03:28
    where the brand is running their media
  • 00:03:29
    media activity so here Brands would
  • 00:03:32
    leverage the media owner data to reveal
  • 00:03:34
    the data points that are required to
  • 00:03:36
    better quantify the effectiveness of
  • 00:03:38
    their marketing advertising campaigns so
  • 00:03:40
    things like ad exposure engagement
  • 00:03:42
    conversion sales all of these um could
  • 00:03:45
    be the data points that are required to
  • 00:03:47
    carry out these use cases and then uh
  • 00:03:49
    the outcomes would inform Downstream use
  • 00:03:52
    cases like campaign reach and frequency
  • 00:03:54
    audience validation so the brand making
  • 00:03:57
    sure that they're actually reaching the
  • 00:03:58
    audience's the they want and
  • 00:04:00
    incrementality which is a um a technique
  • 00:04:04
    to measure how much of the campaign
  • 00:04:06
    outcomes are actually being driven by
  • 00:04:08
    advertising versus the advertising
  • 00:04:10
    simply taking credit for outcomes that
  • 00:04:12
    would have happened with or without the
  • 00:04:14
    ads being placed then lastly uh
  • 00:04:16
    predictive modeling of different kinds
  • 00:04:18
    and different
  • 00:04:20
    types so I touched on identity graphs
  • 00:04:23
    earlier and the main purpose of identity
  • 00:04:26
    graphs or ID graphs in clean rooms um um
  • 00:04:30
    are to improve match rates between data
  • 00:04:33
    sets from different parties um why is
  • 00:04:35
    this important well cleaning rooms work
  • 00:04:37
    by joining different data sets right and
  • 00:04:39
    we've established that um earlier on
  • 00:04:40
    this presentation and this is done
  • 00:04:43
    typically using some sort of user ID as
  • 00:04:46
    the common field and Beyond scenarios
  • 00:04:49
    where data sets um already share the
  • 00:04:52
    same identifier the ability to join data
  • 00:04:55
    will depend on an ID graph or multiple
  • 00:04:58
    ID graphs working together and this is
  • 00:05:00
    crucial for data collaborations as match
  • 00:05:03
    rates really or they often dictate the
  • 00:05:06
    scale and effectiveness of use cases so
  • 00:05:09
    as an example if you look on a slide
  • 00:05:11
    starting from uh starting from the left
  • 00:05:13
    side and moving moving right if data set
  • 00:05:16
    a contains a user record with user ID
  • 00:05:20
    equals 1 2 3 4 5 um so that's the the
  • 00:05:22
    purple circle on the left and data set B
  • 00:05:26
    contains a user record with person ID
  • 00:05:29
    equals 3 4 5 6 7 that's the um you know
  • 00:05:32
    the light orange circle on the left um
  • 00:05:35
    these these are two different ID forms
  • 00:05:37
    right user ID and person uh person ID so
  • 00:05:40
    they have to be linked somehow and this
  • 00:05:42
    is done through what's called an ID
  • 00:05:44
    crosswalk um or walking from or crossing
  • 00:05:48
    over from one ID set to another and then
  • 00:05:51
    it's determined through this crosswalk
  • 00:05:53
    that user ID 1 2 3 45 is the same as
  • 00:05:56
    person ID
  • 00:05:58
    34567 this indicates that the identity
  • 00:06:01
    um between those records have been
  • 00:06:03
    resolved or matched and that process um
  • 00:06:07
    leads to a higher match rate which leads
  • 00:06:09
    to better results and better utility
  • 00:06:11
    from the clean
  • 00:06:14
    room so we we lightly touched on this
  • 00:06:17
    when going through the use case
  • 00:06:19
    categories but for more advanced use
  • 00:06:21
    cases or at least more analytical use
  • 00:06:23
    cases data scientists can deploy data
  • 00:06:26
    models within clean rooms in uh a couple
  • 00:06:29
    different ways uh the first is um and
  • 00:06:32
    this is the the like the Box on the left
  • 00:06:35
    side the first is essentially building a
  • 00:06:37
    new model using the match data set as
  • 00:06:40
    training data and it's important to
  • 00:06:42
    remember that the match data set that
  • 00:06:44
    comes from a collaboration use case um
  • 00:06:46
    it can be used like any other data set
  • 00:06:49
    as long as the data doesn't leave the
  • 00:06:51
    DCR environment and the model can then
  • 00:06:53
    be deployed on existing customer data
  • 00:06:56
    again within the DCR where analyst
  • 00:06:58
    analyst can direct D insights and
  • 00:07:00
    because these insights come from
  • 00:07:03
    aggregated data this Ure ensures that no
  • 00:07:06
    uh Pi is exposed and then the second way
  • 00:07:09
    which is the the right side box um the
  • 00:07:11
    second way is import meaning bringing an
  • 00:07:14
    existing pre-trained data model into the
  • 00:07:17
    DCR for deployment against the data
  • 00:07:19
    that's already inside for similar use
  • 00:07:22
    cases and in terms of the types of
  • 00:07:24
    models that can either um be built or
  • 00:07:28
    imported into to a uh DCR there's really
  • 00:07:32
    no limitation right so any kind of data
  • 00:07:34
    model that's used within advertising and
  • 00:07:36
    marketing things like audience
  • 00:07:37
    segmentation models attribution LTV
  • 00:07:40
    propensity CH prediction and so on and
  • 00:07:42
    so forth everything's fair
  • 00:07:44
    game so at this point you might be
  • 00:07:47
    wondering whether um DCR simply exist or
  • 00:07:51
    S simply allows for existing data use
  • 00:07:53
    cases to be carry out albe it in a
  • 00:07:55
    secure privacy Safe Way or whether they
  • 00:07:58
    actually facilitate say net new use
  • 00:08:00
    cases the answer as best as I can tell
  • 00:08:03
    is that most of these use cases most of
  • 00:08:06
    the um the use cases we outline they
  • 00:08:08
    aren't um they aren't new per se but
  • 00:08:11
    dcrs provide users with access to
  • 00:08:14
    first-party data sources with whom to
  • 00:08:16
    collaborate that were previously
  • 00:08:18
    practically impossible so to give you a
  • 00:08:21
    personal anecdote uh along time ago back
  • 00:08:24
    in the early days of my career um in New
  • 00:08:26
    York City I was more of a standard
  • 00:08:29
    digital planner buyer and at that time a
  • 00:08:32
    lot of the job um was basically working
  • 00:08:35
    directly with Publishers so tons and
  • 00:08:37
    tons of Publishers um negotiating rates
  • 00:08:40
    asking about inventory and then data as
  • 00:08:42
    programmatics started taking off and
  • 00:08:44
    there was a time when we were looking to
  • 00:08:47
    and in hindsight hindsight this is crazy
  • 00:08:50
    but we were we were trying to get a
  • 00:08:51
    publisher or several Publishers to allow
  • 00:08:54
    us to either pixel their site or do a
  • 00:08:58
    DMP sync to basically gain access to the
  • 00:09:01
    first- party data um and every time I
  • 00:09:03
    had that conversation they look at me
  • 00:09:06
    like at three heads right uh or they'd
  • 00:09:08
    ask for what would be a practically
  • 00:09:10
    impossible um amount of investment or in
  • 00:09:13
    order to unlock it so the point is that
  • 00:09:16
    kind of conversation was theoretically
  • 00:09:19
    possible back then but practically off
  • 00:09:21
    limits um and these days with data clean
  • 00:09:24
    rooms that kind of thing is unlocked so
  • 00:09:27
    it's a big deal in terms of unlocking a
  • 00:09:30
    whole new way of working um at a time
  • 00:09:33
    when we really need it when it comes to
  • 00:09:35
    first party data
Etiquetas
  • data clean room
  • audience activation
  • data enrichment
  • predictive modeling
  • ID graph
  • privacy-safe data use