Skip to main content

Build Precision Marketing at Scale with Databricks and Acxiom

Summary

  • General Motors built an in-house marketing data platform on the Databricks Data and AI platform with Acxiom as its data and intelligence partner, gaining ownership and control over its own data, models, and audience segments after moving away from vendor-dependent systems.
  • Identity resolution—combining first-party and third-party data to understand who customers are and what they enjoy—is described as the performance multiplier that enables GM to reach the right customers at the right time across every touchpoint.
  • The platform supports a three-stage roadmap from governed audience segmentation and near-real-time activation to a data flywheel measured through clean rooms, building toward a future operating model that anticipates and responds to individual consumer intent.

Build Precision Marketing at Scale with Databricks and Acxiom

Watch: Build Precision Marketing at Scale with Databricks and Acxiom
General Motors and Acxiom transformed precision marketing by moving from vendor-dependent systems to an in-house marketing data platform built on Databricks. this video explores how they established a customer data hub, implemented identity resolution across channels, and applied propensity modeling to reach the right customers at the right time.
Learn how GM and Acxiom combined first-party and third-party data within Databricks to create governed audience segments, activate marketing in near real-time, and measure impact through clean rooms. The session covers strategic priorities, identity as a performance multiplier, and a three-stage roadmap for operationalizing customer intelligence across the marketing lifecycle.
🤝

Chapters

FAQs

Why did General Motors move marketing data capabilities in-house?

GM moved in-house to gain ownership and control over its data, infrastructure, models, and audience segments, which this video describes as essential for visibility, accountability, and flexibility as the marketing landscape changes. Building on the Databricks Data and AI platform also means that model compute and AI costs are internally governed and auditable rather than buried in vendor contracts.

What role does Acxiom play in GM's marketing data platform?

Acxiom serves as GM's data and intelligence partner, providing identity resolution capabilities that combine first-party and third-party data to build a complete picture of who GM's customers are and what they enjoy. This video describes identity as the performance multiplier that underpins GM's ability to run precision targeting at scale.

How does identity resolution improve marketing performance?

Identity resolution links customer records across channels and touchpoints so marketing can be directed to known individuals rather than anonymous device IDs. This video explains that Acxiom's identity capabilities, combined with GM's first-party data in Databricks, enable an always-on retargeting use case where the right message reaches the right customer in the right moment.

What is a data clean room and how does GM use it?

A data clean room is a privacy-preserving environment where two parties can analyze overlapping datasets without either party exposing raw records to the other. This video describes clean rooms as the measurement mechanism at the third stage of GM's marketing roadmap, used to validate the impact of spend and close the loop on the data flywheel.

Full transcript

[00:09] All right. Um, hello everyone. Uh, my name is Jenny McCloskey. I am the senior strategy manager over technical enablement and marketing applied sciences within General Motors. That's a very big mouthful. Um, and Zachary? Yes, thank you. Good afternoon, Zachary Van Dorne. I'm SVP product strategy and
[00:25] partnerships at Acxiom and I lead the customer data lake go-to-market and initiative on behalf of Acxiom. And this afternoon, uh, we're going to walk you through how GM built out our internal capabilities to drive precision targeting at scale, what
[00:41] that unlocked for GM, and where we're trying to go in the future. So, let's get into it. Yep. If I can press the button. There we go. All right, so to understand why we built what we built, I want to
[00:58] first look at what's driving our decisions at GM. And these priorities shape every data decision that we make within the marketing applied sciences org, or as we like to call call ourselves the MAS team. A couple team members over there. Hello. So, market share growth. It's really,
[01:14] you know, reaching not just all of the customers, but the right customers. And having ownership and control over our own data, our own infrastructure, models, and audience segments provides that visibility and accountability to our marketing stakeholders. Having that ownership and
[01:32] control means we can easily flex as the marketing landscape changes, and model compute and AI token costs are internally governed and auditable. And we want to be good stewards of our marketing marketing stakeholders' dollars. And so, we want to ensure that
[01:48] every dollar is spent in the right place with the right person. And with Acxiom's help, we are now identity-led. So, we know who our customer is and what they enjoy, which is absolutely critical to finding the right audience.
[02:04] And finally, it's important to us to build a clean infrastructure and I am super nervous. Let me start that ONE RIGHT OVER.
[02:22] I TOLD YOU I'D BE JAZZ HANDS. And finally, it's really important to us to build a clean architecture that scales with enterprise needs. You know, these six priorities have driven the work that we've done since we started this team a little over 2 years ago and it's what drives us today. Nice setup.
[02:38] So, if I were to project myself into the future, what would I see for GM? I would see an operating model where we anticipate and respond to individual consumer needs, to individual consumer intent and activate the right experience in the
[02:55] right moment across every touch point. And well, we're well on our way with what the marketing teams have done over the past few years. Now, today we think about this across three dimensions. The person, the identity, and the message. Now, the
[03:10] person is the who. So, leveraging GM's own first-party data, we have signals such as site behavior, app activity, and historical purchase records and enriching this with Acxiom's individual additional signals and intelligence on that profile
[03:27] creates an addressable dynamic profile for every consumer. Now, if the person is the who, then the identity is the how. And so, it's that connective tissue that creates a consistency layer across platforms from individual to household
[03:43] to device and it's what makes all of this work at scale. And then the message is the what and the when. So, the vehicles, the features, and ad experiences personalized based on where someone is in their purchase journey. Contextualized by intent. And
[03:59] increasingly driven by AI. We're not sending an ad to everyone and hoping and praying that they go buy a vehicle. We are sending an ad to the right person and we're being intentional at that individual level. So, let's talk a little bit about how we got here.
[04:16] A few years ago, GM's marketing data and audience capabilities were heavily dependent on third-party vendors. We didn't have full visibility into the models being built and how this was driving audience inclusions and exclusions. And we knew that that had to change. So, we made a fundamental shift
[04:33] from vendor dependency to a GM-owned in-house marketing system. We established the customer data hub, a scalable governed data lakehouse that serves as the single source of truth for our GM customers. It is that 360°
[04:48] view of our customer that powers our in-house propensity models, helping us find the right customer at the right time. Yeah. Yeah, and this is Yeah, and this is where we we come in, right? So, Action, all right, our role in this is to You know, think of us as a force multiplier
[05:05] on the on the data foundation side, right? So, if you look at all the the raw first-party signal from from GM and you look at how that is ingested and normalized and and and, you know, computed within this data foundation, we bring a an optimization insofar as
[05:22] understanding that intention, right? Are they in market? At what phase of market are they? The qualification relative to what would be the the most qualified and best product to to offer and the persona, like who are they, right? What is their lifestyle? What are their interests? What are their life stage?
[05:39] All this is context and understanding that adds to that foundation that brings context to then the the personalization and orchestration layers, right? So, it's typical the combination of the first-party and the authoritative consumer graph that we bring, we're meshing that together in
[05:56] Databricks to to really optimize the foundation. Right. Um yeah, and so what that brings, right, is then actionability, right? So, that's where I'll just go to the next slide. Absolutely. Yes.
[06:11] That brings in Databricks, right? And they bring a force multiplier with regard to the platform, to the technology, right? So, if you look at Acxiom with GM, you look at our the the amalgamation of our first-party and third-party data sets
[06:27] together, um the Databricks brings a lot, right? So, one, like the speed and agility, right? So, this idea and being able to capture all this intent signal in near real time to be able then to automate
[06:43] and/or I'm a like automate the with the marketer, with an agent working side by side with a marketer, the ability to calibrate and create an audience and operationalize and action that audience relative to this in-market intent signal, right? Um you look at the the
[07:00] partner leverage, you look at the the foundations of that, right? So, you look at clean room calibration like clean room collaboration constructs and ability to further bring second-party data, publisher level data, all of that the data that's going to be exposed within that marketplace, that
[07:15] ecosystem, that matters relative to creating that personalization, right? And so, that brings tremendous value, right? So, there's the value that you would normally associate with a CDP, right? So, creating an audience, seeing the audience makeup, seeing the overlaps,
[07:30] um but then there's the compound of that kind of baseline value for Databricks and there's gigantic layers that come to that. We'll get into that in a in a few slides. Okay.
[07:46] Um so, I will talk about the phases of CDP and so Tatsu earlier at the keynote he talked a lot about this so I won't revamp that but I'll add some color relative to how Action sees this these phases of CDP from the very beginning to to where we
[08:01] are right now, right? So we look at this in terms of just problems that they were intended to solve and then we look at okay, what problems did they create? And then what these additional phases came into to solve these problems, right? So package CDP so this was the problem
[08:18] marketers not having accessibility data. Yeah, marketers not being able to act on the data in a any sort of relevant time, right? This is the days of data being locked in on-prem environments. This is the days of marketer having to submit a Jira ticket
[08:33] to get an audience data set into a ERP or CRM system or a email marketing system. So the CDP package CDP was really designed for that accessibility empowerment of the marketer to have their hands on that data to them being able to act on that data, right? What
[08:50] that meant is that data that was traditionally locked in IT and technology environments to get that into a a structured environment where they can act on that, right? And so that be get the sort of the traditional package CDP where they're ingesting, they're storing, they're computing
[09:06] and they're resolving data all kind of within the self-contained well kind of well kind of multi-grain. Um so then the problem became well as all that first-party data moved into the cloud foundations from from away from those on-prem locked environments to
[09:23] a cloud foundation like Databricks then you you you you find yourselves with a silo with a duplicate duplication of of data that's in the CDP and that is in the the warehouse and then increasingly marketer data science teams and engineers are building
[09:39] intelligence on that data foundation in the cloud environment. And there is a just immediate silo relative to the data between between the two, right? So, that begets the the rise of the composable CDP where CDPs got out of the business of
[09:54] data management and they brought all that application layer on top of the data warehouse and they're pouring down at that that sort of centralized horizontal phonetic data foundation. And they still perform those functions, right? Um So, that rise
[10:09] relates to the next problem, right? The next problem is uh Well, it's also you know, talked a bit about it, right? The the fact that within package CDP and composable CDPs, you still have the the traditional framework of a waterfall um from
[10:24] planning to ideation to audience creation to activation construct or foundation. So, if you think of uh something like in the context of GM where intention matters and time to market matters, this uh lag can take anywhere from four to
[10:41] six to eight to 10 weeks relative to okay, this person is in market, we've got to get a message to this person. Okay, where do they fit within the audience uh segmentation construct? What campaign we put them in? Um how do we get them syndicated, right? It just
[10:57] weeks and weeks and weeks and and opportunities lost, right? And so, with the Gentik, um this provides that great opportunity to to automate and bring from an agent working side by side with a marketer to really so, supercharge that ability to to
[11:12] calibrate and design and recommend and operationalize audiencing and decisioning and and planning within this all within this framework, right? Um I'm going I'm speeding along, aren't I? Yeah. Gives us plenty of time.
[11:28] Okay. We're good. Is it Is it a good thing that it comes? And this really drives, you know, what does this a lot to the marketer, right? Um And we talked about um speed to market the relative to intention and to qualification and persona. So, they're
[11:43] being able to create an audience, right? The smarter audiences, I know you know, at glance it's a little trite, it's a little cute, but it's to me this is the true power of an Agentyx CDP. Um I found that uh when you give
[11:58] marketers an application that it has a UI that has access to a data foundation that is could be hundreds of millions of rows, hundreds of thousands of columns, like it is a massive quantitative
[12:13] load, right? To be able to accept, you know, conceptualize and operationalize a massive amount of data when you're like strictly kind of {{}quote} {{}unquote} a human, right? So, what audience do you create? What journey do you create? What channel, you know, do you do you decide to to communicate them to? What
[12:31] offer sequence do you send them to, right? And so, the ability to have an agent side by side with a marketer to really solve that problem of what audience to create, right? That I'll say creating the audiences is easy, right? And audience segmentation creation has
[12:46] been around for for decades, right? Um that is not the problem. The problem is the decisioning and the intelligence that goes into how do I how do I automate that? How do I optimize audience creation when you're dealing with hundreds of millions of of rows of
[13:01] record at a profile level basis and columns that are attributes and context and uh and models and signal, how do you how do you truly bring that to market, right? So, that that for me is the great promise, right? Um
[13:16] Less waste, more precision, right? This also is is highly relevant, especially for GM, right? Is um you with that um precision comes an then increase the opportunity to go really into
[13:32] qualification intention that is really right and really correct and really suppresses those who are truly not in that market, right? And I know suppression has been around for forever, right? But this really brings it to a whole new level, right? Yeah, and and before we go to the next
[13:47] slide, I I put this in green because I wanted really zone in on this. In the automotive industry, we only have 5% of the population in market for a vehicle at any given time. It is incredibly important that you are
[14:02] finding the right people to target. Of those 5%, who knows how many are going to have affinity for your brand, right? So it's so important that we have the right models and the right data and the right infrastructure to support this kind of precision. Yeah.
[14:20] Can I ask a quick question? No questions right now, sir. Thank you. We're We're presenting right now. Yeah. I appreciate the enthusiasm. We're flying through this. Yeah, we'll we'll touch on yeah, we might We'll we'll see. Okay. Please Please continue.
[14:36] take my time on this one. So this one is exciting. Um This really exemplifies the or the example of an always-on uh one-to-one retargeting machine. Um always-on, right? So if you think of
[14:51] intention, right? Of signal, right? And so you imagine users coming into chevrolet.com, it's going through the truck section, you all this is raw, right? And so you're getting um of course page views, you're getting
[15:06] button clicks, you're getting um um the custom configurations, you're getting video clip plays, right? It's like very very raw signal. And you can of course capture that, you can collect that, you can of course put that into some baseline
[15:21] levels of high mid to to low intention, you know, groupings. Um Action, where we come in, is we provide an ability to recognize and persist a profile object in real time, right? And so this is inclusive of
[15:38] coordination with a digital first-party ID. We incorporate that and mesh that with our Action Real ID or a graph ID. We can syndicate that, coordinate that with other uh conversion conversion APIs that expand addressability enrichment.
[15:54] That all that raw signal is collected. That data then is streamed into Databricks. Databricks customer lake will ingest that within their um I think it's called the the lake flow ingest protocol. That then goes through the resolution process, right? So the resolution process when Databricks would take
[16:11] the ID signals that we are collecting, right? So that could be the first-party digital ID, that could be the Real ID, that could be some conversion API IDs. They can be other IDs that we are collecting. That can be brought into Databricks and through customer lake
[16:27] I am a profile agent, they that provides the means to really normalize that into a persistent profile object, right? That can then have a continuum relative to a a journey, right? That begins It can begin at this this point of a first-time visit, right? Um
[16:43] that goes in. That could then correspond and and can be categorized relative to what type of profile it is. Is it a prospect in the kind of either a third-party or pseudonymized first-party context? It is a customer,
[16:59] right? Do they authenticate? Uh do they permission themselves? Do we know who they are relative to the first-party data? If we do, great, like we can like Databricks would be uh customer lake would be able to flow that into a a first-party known segment as opposed to
[17:15] say a pseudonymized unknown prospect segment, right? So, that ability for It's a great example of an action as an authoritative graph ID provider working with a identified CDP that's within Databricks that has like a very
[17:30] enterprise level ID resolution process coming together to normalize that and create that foundation, that data foundation, right? So, that all goes all in to the foundation and then you go into customer lake. Sure. And that that's where you have the
[17:46] markers being able to access that data, right? And this is where we can operationalize with additional context, right? So, the GM's data science team they can come in and they can add additional context and and relative to additional models that
[18:02] we put on that, right? So, that could be propensities that could be intent-based clustering, that could be indices, right? Um and that could also then be further enriched by our graph, right? So, first-party raw signal with action,
[18:17] psychographic, demographic, and to see data, you know, attributes amalgamated together, you have GM's data science team building additional context on that. That then is accessible to an agent and then you can imagine the power of an agent being able to have all this rich
[18:33] feed of data and context to be able to then automate the audience creation, right? In real time or near real time, right? And and near real time, yes. Near real time. It was right. Yeah. Um Yeah, I have a question. And that then creates that flywheel, right? And so, the syndication can
[18:50] occur, right? As near real time, right? We can then, using Databricks clean room, we can begin to extract the kind of exhaust coming out of that data. We can join that back at various levels of granularity depending on the channel, right? So, it
[19:06] could be the ID-based, it could be the household-based, it could be the audience-based, the channel-based brought back in, and you can imagine that's sort of that infinite loop flywheel that Tosha was talking about. To be able then to calibrate and tune the agents, and then
[19:21] calibrate and tune the audiencing and the sort of the decisioning power and planning power that you would have within within the customer journey. So. And this is just one of the many um opportunities that we are exploring right now. Uh we were lucky enough to be
[19:38] in a private preview with Data Bricks on this. Uh so this is really, you know, a big dream for us, but it could be a reality, which is very exciting. Yeah. So what's next for us? Um what are we doing to enable this big
[19:54] beautiful dream? Is we're thinking about the road map in three stages. And so we have the identity and privacy foundation that is done. It's already unlocked. We have a unified identity across site, customer, and digital, a privacy safe workspace, privacy safe.
[20:12] Um for speed and agility, and we have the ability to set up governed partner data sharing via clean rooms. We are actively exploring what's possible through triggers and automation. Site interactions become marketing events in near real time.
[20:29] Near real time. Yeah. In near real time. There we go. We got it. Dynamic audiences uh refreshing as customer signal intent comes in. We can have AI monitoring signals, AI flagging opportunities, AI initiating actions. And now on the horizon, we're
[20:46] really looking at this continuous agentic loop and how that personalizes across channels. A system that runs, learns, and adapts with minimal human intervention, but with full human oversight. I mean, this is not just AI
[21:02] in marketing, but it's exploring how AI can help support the humans within marketing. All right, Ann. All right. So, we're going to do a little in lieu
[21:17] of Q&A, we're going to do a little fireside chat. I think we're going to get comfy though and get some chairs first, so give us a second. Okay. So, I I've been at Axiom for about 3 years. And 3 years ago,
[21:34] um I I didn't have a chance to come to my first Data Bricks conference, but um a stakeholder at GM did. And I've always had this vision of sharing the stage with GM. So, thank you for the leadership. Thank you for the partnership. I'm super excited to be
[21:51] here. Um I always like to start with the end. So, uh people are at the core of everything that we do. And it's ultimately the future of AI. So, taking us through a lot today. And I am curious, we'll start with you,
[22:07] Jenny. What is one lesson that you'd like our audience to walk away with today? Oh, okay. Um So, I was lucky enough to be part of the marketing applied sciences team uh from
[22:22] its inception. And it's it's been an incredible journey, and I've learned so much along the way, and it is really hard to pick just just one thing. Um It would have to be cross-organizational alignment.
[22:39] Get that now. Sit down together, build an achievable roadmap. Sit down with your leadership and tell them about your risks, your challenges, your timelines, so that they are not out there promising on work that you can't deliver.
[22:55] Sit down with your stakeholders and get feedback, understand their priorities, understand the business context. There are so many teams out there who know why a campaign worked or why it didn't work. They know why a customer
[23:10] segment behaves in the way that it does and that that data is so incredibly valuable, but you're not going to find it in any database. And without it, you you're not going to be successful, right? And so so we when you miss that people alignment, you miss the story
[23:26] behind it. And so I I ask all of you be Don't let perfect be the enemy of good and don't let personal conflict get in the way of your company's achievements. You know, I I can throw a rock in this room and I can hit 10 wonderful data
[23:42] scientists who can do our jobs. You are all amazing, wonderful people. And budget is relatively easy to get. I but it's You're laughing. I can see it, mate. Um but it's the people. It's the people who
[23:57] are going to make you successful. It's the people who are going to drive change. I would not be up here if it were not for my colleagues who have worked so near and close to me over the years. A lot of them are in that row right there. I love you all. You're amazing. And
[24:17] You know, I had a I had a leader once who used to say stake partner and I always thought that was like a little a little cheesy, but um I know like I going into any project with that kind of a collaborative mindset, that's what's going to make the hands-on keys successful. That's
[24:32] what's going to make your data scientists successful. I I think that's so excellent. And it's usually not a technical capability that prevents us from getting from point A to point B. And with that being said, Zachary, I'd love to get some perspective from you. If we could walk
[24:48] away with anything today, what would be your advice? Yeah, point A to point B. Um Well, I had a canned answer but I I'll I'll scrap it because I thinking about Tasso's keynote um and I'm thinking about uh the this this is a very big pivot for and
[25:04] pivot of moment for Databricks and now bringing a an actual business user marketer user application to to the market. Great. Um it's no longer for IT and data. It's no longer just for the CIO and CTO. Like the CMO and the teams within the CMO now
[25:19] have an actual get a value and and and uh usage of a direct especially usage of Databricks and so I I hope that uh we were able to bring what was a very top line uh maybe somewhat abstract, you know, uh framework and positioning of
[25:36] what it's customer like and bring it to a more real tactical and exemplified framework in so far as how a brand like Jean would be using that, right? And so going through some of the use cases, going through some of the the phasing of what this means from a to be an identity
[25:52] CDP, what that really means for a marketer. I think it tends to get abstract and ethereal and um it's hard to sometimes understand, okay, what does it mean for my day-to-day? So I hope we had we give some clarity to that. And it's really good to hear. No, that's that's excellent, Zachary.
[26:08] Thank you. Uh we're going to move on to the next question and really kind of the the theme of this one I think is identity as a performance multiplier. Uh Jenny, I'm going to go with with you first and I'll throw it over to you, Zachary. A strong data foundation with
[26:25] identity is invisible, but it's also core. So why is it important to everything you've shown today from modeling through to activation? Wow. I'm going to let you. Okay. I'm going to tell a little teeny tiny little story. Uh when I first got
[26:42] out of grad school, I wanted to be a data scientist so so bad. I just I wanted to sit there and and model and and tune my hyper parameters and and it was just it was such a a beautiful dream and then I got out in the real world. And I learned that data science is
[26:58] really only about 10% modeling and 90% data engineering. And I see a few head nods. Yes, you know. And I just I found that so interesting. And so when it comes to activation, we can't find the right person if we don't have
[27:13] the right model outputs. But you won't have the right model outputs if you don't have data you can trust. Trust. You're going to love And so it's not the the sexy work, but it's the work that powers all of the insights that you use. And then when it comes to identity, you know, without it,
[27:30] you are personalizing to a a session or a device. You're not personalizing to a person. Okay, personalizing person. And so um Um Okay, so think about the the Silverado example that we walked through earlier.
[27:46] So let's say Bob goes to our website. And what if we don't know that we also sent Bob an email 3 weeks ago? Or that he came to the website 6 months ago and was configuring a vehicle. Or that he's part
[28:02] of a household that we have been targeting with media for the past 4 months. If if all of those data signals stay siloed, we're just we're building noise. We're not we're not building something that's actionable. And so identity resolution, it it helps us to
[28:18] break down those silos and and we can build propensity scores that actually reflect a person's true purchasing journey. And personalize content for them that's going to resonate. I mean what what makes anyone engage with an app?
[28:33] What makes What makes me engage with an app? Unless I touched it. Uh what's making What makes me? Um if you if you know me, you know I love dogs and I'm a huge foodie. And so if I'm on the Silverado website, yes, I want to know that it can
[28:50] handle mud, and that it can handle fur, and that I can clean up, and I can feel really good about myself rolling up into the valet to a Michelin star restaurant. And so, you know, with with identity resolution and and the strong data foundation, you can
[29:07] you can, for example, take like a a 70% model accuracy and turn that into double the lift and response rate. But without it, you're just you're just trying to look really cool with some modeling work that's not going to drive any business results. Yeah, I see.
[29:23] I really want to take that Silverado to the Michelin star restaurant. I do, too. Yes. Zachary, this this one's for you. Um plays off of the question for for Jenny. But from your perspective, what breaks or becomes impossible when identity
[29:40] isn't done well? Well, everything breaks. Um yeah, so that one thing, everything will break. Um and Tell us more. Yes. Well, the way we look at identity is first you you have to understand that when you talk about data data, the data
[29:56] in this context is our consumers, our humans, our our persons, right? Um and the the goal or the output of identity is to is to create a person level profile. So, a persistent profile that we can
[30:12] understand and then we can address, right? So, it's the then you look at the the the various functionalities of identity, which are sort of the ingesting all the raw disparate data. There is then the uh matching all of that raw disparate data
[30:28] to a to a graph, such as Actions graph. We we bring in a authoritative graph ID, a real ID, to then provide the basis in which to then to deduplicate and stitch. So, you get into ingest, deduplication, stitching. This is all like tactical and can be very messy,
[30:45] right? But the intent of all that is to create a persistent profile to which you can then have understanding and then you can have a conversation through a continuum of a journey, right? And that's really the the basis of what is identity, right? Understanding
[31:01] and ability to communicate, to address, to to to reach, right? And then you get into its tactical minutiae of uh first-party data and third-party or enrichment data. You get into match rates, you get into addressability and targeting and scale and reach and right?
[31:17] All that is just tactical minutiae relative to the core which is persistence, understanding, and and and addressability, and and communication, right? So without that that underlying foundation then you the data scientists don't have
[31:32] the the basis of a data foundation to to build additional context on top of that, right? The indices, the propensities, the the scoring um whatever model you would want to build relative to context we just can't do without that that persistence, right?
[31:47] You cannot uh reach them, you cannot communicate them without the construct of addressability as a as a goal, as an as a end point of of resolution. Like I need to address, I need to reach, I need to have a communication that's ongoing relative to where they are in their life stage and
[32:03] their phasing, right? So then you have the orchestration on top of that underlying identity foundation, the context layering, the agentic layering, and then the orchestration layering. That's where you get the marketers and the agents working together to create
[32:18] personalization, to create experience that's relevant. That's that's with precision, that's with intention. That's um uh that is in the in the context of who they are and where they're at in their their staging and what they're into and what they're not into and and whatnot
[32:34] and whatnot and whatnot, right? And then of course the ability then to capture all that the um um the output of that customer experience, the exhaust, and then bring that back in. How do you bring that back in? Well, you need a common identity foundation. You need a persistent profile. You need an identity key that allows you to join that back.
[32:52] Right? None of that would exist without strong identity, right? I could keep I can go on and on, but uh Without identity, it all breaks. It all breaks. That's right. That's right. I can't think of one thing that doesn't break. Right? Like none of
[33:08] the magic of identity CDP would work without this. Right. Cool. So. Thanks. Right. And that's really the core of the value we bring to to that top line. Thanks. Yeah. Uh this last one, Jenny, is uh for you. And its theme really is making customer
[33:25] intelligence actionable. Okay. A lot of companies and a lot of people build great models. They build great models. But struggle to operationalize them. You do. Yeah. How is GM actually activating these insights across channels and customer
[33:41] touchpoints? Ooh, okay. I love this question. It's very close to my heart. Um I did it again. I need to stop doing that. So, uh I'm going to touch a little bit back on the modeling. Building models is so much fun. All right. So, you built a model. You're done nodding. You know what I'm
[33:57] talking about. You just you get into this zone and it's all about exploration and it's trial and error and you're again, you're you're tuning your parameters and and you can just get lost in it for hours and hours and hours. And the whole day goes by and it's just this this incredibly beautiful thing to do.
[34:13] But then But then the model's built and the the fun part is over because now you have to operationalize, right? You have to put governance in place. You have to be optimizing for your compute costs. You have to be production ready. And that work is hard and it requires a
[34:31] lot of multiple teams working together. Excuse me. And so, you know, what made what made MMM successful and what makes us successful today is now we're in-housing our media mix models is we
[34:49] looked at activation as a systems problem to solve. And so, yes, we had our data scientists working through with the logic. At the same time though, we had our data engineers building pipelines. We had our test and learn team building out a scaled parity
[35:04] approach. We had our leadership out in the front lines gaining buy-in. Good question. Excuse me. And you know, we had a lot of teams working together and and rowing in the same direction. We were doing this over
[35:19] time. We didn't try to do it all at once. We were in-housing small pieces each time, vetting them, solidifying them, and moving on to the next one. And so, again, if I can if I can harp on this one more time, that cross-team alignment
[35:36] is is really what helped us operationalize. And it's the hardest part, but it has the richest payoff. And now we have our propensity scores, we have audience definitions, and we have um
[35:52] our identity graph all in Databricks, all close enough to the business so that we can actually run at the speed of the business. I I think that's amazing, Jenny. Thank you. And uh everyone that joined us today, thank you so much for your time.
[36:09] We really appreciate it. Uh we will not be doing any open Q&A. Um but feel free if you are in market for a GM vehicle, you can come see me. I'll throw in that 5%. Uh Zachary, thank you. Jennifer, thank you. I've been
[36:26] thinking about this for three weeks, So. so

Learn more about the Databricks Data and AI platform.

The information provided herein is for general informational purposes only and may not reflect the most current product capabilities or configurations.