Agentic Marketing at Scale: Customer Lake and Enterprise AI

Summary

Databricks Customer Lake is an agentic customer data platform embedded natively in the Databricks Data and AI platform, enabling marketing teams to build audiences with natural language prompts and unify customer profiles without fragmenting data governance.
HP, Epsilon, Circle K, and the Databricks marketing team share how to scale AI-powered personalization by treating data foundation as a core marketing imperative and consolidating fragmented MarTech and AdTech stacks onto a unified platform.
This video covers how to manage agents alongside humans, distinguish low-stakes automatable work from high-value strategic decisions, and measure business impact rather than marketing activity metrics.

Agentic Marketing at Scale: Customer Lake and Enterprise AI

The most AI-ready function in the enterprise is marketing. With data-rich operations, always-on channels, and continuous experimentation, marketing can drive dramatic returns through agentic AI. However, realizing that potential requires unified, governed, and real-time context that sits alongside human judgment and strategic oversight.
In this comprehensive marketing forum, learn how CMOs and marketing leaders are treating data foundation as a core marketing imperative. Discover Databricks Customer Lake, an agentic CDP embedded natively in the data platform, and see real-world implementations from HP showing how to scale personalization without fragmenting data governance. Hear practical advice from Epsilon, Circle K, and Databricks marketing teams on building context, managing agents alongside humans, consolidating fragmented tech stacks, and measuring business impact rather than activity.
🤝

Chapters

FAQs

What is Databricks Customer Lake?

Databricks Customer Lake is an agentic customer data platform (CDP) embedded natively in the Databricks Data and AI platform, introduced by Michael Trapani in this video. It enables unified customer profiles, identity resolution, and campaign agents that build marketing audiences through natural language prompts without requiring separate CDP infrastructure.

How is HP using agentic AI in marketing with Databricks?

HP tested the agentic CDP approach to scale personalization without fragmenting data governance, working toward a unified marketing foundation on Databricks. HP's transformation also included organizational readiness work and employee enablement to prepare marketing teams for AI-assisted workflows, as described in this video.

How do marketing agents work alongside humans?

This video distinguishes between low-stakes work that agents can handle autonomously and high-value strategic decisions that still require human judgment and oversight. Speakers from Epsilon and Circle K discuss building governance frameworks for agentic marketing that maintain human control while allowing agents to handle tasks like campaign audience creation and performance analysis.

What is the build vs buy decision for marketing technology on Databricks?

The closing panel in this video examines whether organizations should build custom marketing AI solutions or purchase vendor products, with perspectives from Epsilon, Circle K, and Databricks. The discussion emphasizes consolidating fragmented MarTech stacks onto a unified data foundation rather than adding more point solutions, using Genie as a single interface to reduce tool sprawl.

Full transcript

[00:08] All right. Let's go ahead and get started. Welcome everyone. I'm excited to be here. Uh thanks for joining us here. Um my name is Rick Schultz. I'm the CMO of Databricks. Um and it's kind of exciting for me. This is my 10th uh Data + AI Summit since I joined Databricks. And
[00:24] um and it's actually our second year doing a marketing forum. Uh so it's a relatively new program for us, but we had it so much interest in it last year that we brought it back and I think there's going to be a couple hundred of you here in the room during the session. We have a great lineup of uh customer speakers and some Databricks speakers.
[00:39] And I'm just going to open it up a little bit. Um and uh let me start by just getting a sense of like who's in the room. Um I just want to understand how many people work in marketing and how many people um don't work in the marketing department, but support marketing and work on marketing projects or what have
[00:56] you. So, first of all, how many people work in the marketing department? Show hands. Okay, so so about 50 60%. Okay, so I don't need to ask the other side of the equation. Um got it. Awesome. Well, cool. Well, I just wanted to set a little context for what we're going to talk about today
[01:11] because um as you might imagine, it's all about data and AI in marketing. But marketers have always been all about the data, right? Data has always been so critical, whether it's the statistical numerical data about campaign performance or customer data. Um and I think what's
[01:26] really this unique opportunity is um how to approach AI uh now in in the moment we're in in time. It's it's not only that AI is changing um dramatically in some ways changing buying behavior of customers, whether you're in B2B or B2C, uh it is
[01:43] fundamentally changing buying behaviors, but it's also changing the way you have the opportunity to do marketing. Uh and and that actually can can dramatically affect things like uh how effective and personalized your marketing can be to how efficient you can be and what things can be done by agents and what things
[01:59] still require humans and what have you. And how do you find that right balance? So, I think it's exciting, there's this massive opportunity, but it's also a challenge. It's like how do you actually paint the vision of where you want to be a one, two, three years from now? And how do you actually pragmatically start
[02:15] moving there quickly and getting toward it? And I think the way we see it is um that uh if you I'm assuming most of you attended Ali's keynote yesterday probably and heard us talk a lot about context over the last day and a half. And that that's the first point I want to make is how we think about it is that
[02:31] AI is like super smart already. It's it's brilliant. As as Ali said yesterday, AI does not have an intelligence problem, it has a context problem. It needs to understand your data, right? And specifically, what we think the two big types of context are
[02:46] that are most important are customer context and business context. And what we mean by that is customer context is who's the target, who's the person, who's the persona engaging with you, where are they right now in their journey, what channel they
[03:01] are they engaging with you on, what's maybe their loyalty status or what have you. Lots of and I could go on and on, but basically, it's all the the information about the consumer or your customer. Um but the business context is also important and that's what's your strategy, what does your brand stand
[03:17] for, uh how are you trying to approach this? Is it a new market or is it a cross-sell? What are your objectives for a given campaign? And so really having both those things is important and where they meet, the intersection of the business context and the customer context, is super critical
[03:34] to being able to do effective marketing. And uh what's cool is that when you think about what's cool for the people in this room is when you think about that, um marketers are actually kind of right at the center of it. Marketers are actually right at the intersection of
[03:49] the customer data. Lot times, marketers own many of the engagement channels, um they own the definition of the persona, they own the messages, they own the brand. So, they kind of own a lot of this or at least um play a pivotal role. Oh, wow. Now, I got to get Now, I want to get
[04:05] competitive and we need to like give it a big cheer, like a really LOUD CHEER. WOO! THERE WE GO. Um thanks for playing along with me. Um competitive person. Um but yeah, so we think there's this massive opportunity for marketers and that's the
[04:21] good news, right? Marketers absolutely should have a voice in what's happening. Marketers should lead the way in adopting AI. The challenge becomes like how do you do that? And we fundamentally think that the things that need to happen are the unification of the data and the understanding of the data. So,
[04:36] by unification, it's it's kind of the case in many organizations at least where you have some of your data in your customer warehouse or whatever marketing cloud you use or your ad tech stack or your social media um channels or all of the above. So, when you have the data all over, you kind of at least
[04:53] need a common view of it. So, you need a unified view and you need this context. Uh right? That you can provide to things like Genie ontology uh that was discussed uh in the keynote yesterday. And so, once you have that, it not only enables um you to build your
[05:08] own agents, but it enables things like the CDP which Michael's going to come up and talk about uh with customer lake. It enables you to run agentic campaigns more effectively and it also enables you to make all your all your people more effective and productive uh through self-service. Uh and an example of that
[05:24] is us. A couple of years ago, um Liz uh who runs our marketing technology and is going to be on a panel a little bit later on this stage. Um you know, her team actually rolled out We were the first department in Databricks actually to adopt Genie which is now the the thing we adopted is now called Genie 1. And uh we've been using
[05:41] it for 2 years and it's phenomenal. Uh it really is is awesome product because it understands our data and even someone non-technical like me, my my sequel queries are kind of rusty. So, uh some someone non-technical, which you can imagine is like 90% of our department is is kind of non-technical,
[05:57] let's say, uh can ask questions like, "What was the What were the top 10 performing campaigns based on total pipeline or pipeline on investment or forecasted pipeline on investment? Show me the top 10 campaigns over the last two quarters." Or show me the worst performing campaigns so I know which
[06:12] ones to kill or things like that. So, uh anyway, I don't want to go on too long, but those are the key points we want to discuss today are um the importance of context and be able to have an AI vision and AI roadmap, um the role marketing can play in that, and how
[06:28] important it is to unify your data and bring together context on it. Um so, with that, I'm going to introduce our first keynote speaker, uh Michael Trapani, and I'm super excited to introduce him not only because he's a fellow New York Knicks fan. Um shout out to the Knicks, but um he also is he's our head of product marketing um for all
[06:46] Agentyc applications, which includes a CDP, but he has particularly strong background in CDP having worked for several years at ActionIQ. So, welcome to the stage, Michael Trapani.
[07:03] Hello. Hi, everyone. How are we doing? Good. Excellent. Wonderful. You know, as a marketer, it is nice to be around some familiar faces, whether it's marketers or folks who support marketing. Listen, I'm I love meta harnesses as much as the next marketer, but you know,
[07:18] it's nice to talk about marketing. Um so, thank you, Rick, for for the time that you spent sort of setting the stage. We're going to be spending some time today talking about what's possible and what folks are doing right now, as well as give you some practical next steps on how to make all of this stuff real. So,
[07:35] I'm going to tell you a little bit about some of the announcements that we've made this week that are extremely relevant to marketers. Then, we'll be bringing some of that to life with some real-world examples from the team at HP who's going to be talking about how they are engaging their customers in uh really exciting ways.
[07:53] And then we'll be bringing it into uh how some general best practices of taking agentic marketing and making that real in your organization. Our friend Dan Morris is going to be talking about that. And then to bring it home, last but not least, our grand finale, our friend Scott Brinker will be joining us
[08:09] to host a panel with some marketing leaders on how they are sort of bringing in this next era of marketing using data, using AI, engaging with their customers on a truly personalized basis. So, um let's dig in into some of our uh
[08:25] presentations that we had this week. So, um for folks that don't know me, my name is Michael Trepany. I've been in the marketing technology space for some time. Um and like I said, it's great to work with some familiar faces. I've worked kind of on the intersection of
[08:40] business, marketing, as well as data and AI. And I'm here to talk to you a little bit about Databricks Customer Lake. So, for the folks that didn't see the keynote just recently, um or maybe who don't do not have a chance to look at some of the uh the the sessions that we've had so far, I'll talk a little
[08:56] about a little about it here and we can go into some some detail. So, let's start with why something like Customer Lake is useful, valuable. I think the main point that we're trying to get at is that we all know this, right? We're all supporting or adjacent to marketing.
[09:13] Businesses grow when they understand their customers. Right? Customer data, that said, makes it an extremely valuable asset to own. But as we also know, for the folks that are connected to marketing here, that extracting value from customer data is
[09:29] an enormous challenge. And there's a number of reasons why this is the case. First is getting that data that you have, wherever you keep it, from raw data to golden customer records, is extremely difficult. It can take months, it can take sometimes years, and then
[09:45] there's work to continue that over time. It's not a one-and-done situation. Then, once you have that data, maybe in a way that you're ready to use it for the business, it's really hard to get that data from data to action, right? Using that data and actually making
[10:01] sense of it in the campaigns or in the ways that you target your customers. It oftentimes rely on a pretty disconnected martech stack and adtech stack. And then, we all know that AI has this tremendous potential to automate work, to help us do our work, to personalize
[10:16] for customers, but the way we've seen it implemented so far feel a lot like sort of bolted on chat interfaces to existing SaaS applications. Now, the requirement to keep up with this, to make this real and to sort of get this
[10:33] data into action, is really important now more than ever. Rick talked a little bit about how consumer behavior is changing. I mean, agents are now being deployed on behalf of consumers to do a lot of comparison shopping for them. Something like 70 to 80% of US consumers
[10:50] have consulted with a chat like ChatGPT before they make a product purchase. So, this is really really important. And as a result of this happening, buying decisions are happening a lot quicker than they used to cuz a lot of that research days, which can take days,
[11:06] hours, or even weeks, is now happening in seconds. And then with buying happening everywhere, the proliferation of commerce becoming headless, buying is happening in every channel, right? So, these decisions and the requirements to keep up with the type of consumer that
[11:22] we're dealing with now is harder than it's ever been. Now, if we think about how we try to solve this problem, the way that we used to think about this was through the concept of marketing technology or even customer data platforms, more purpose-built solutions to solve this
[11:39] problem. But what we're finding is that none of these solutions have really been close enough to the data to make this work, especially if we're dealing in this new world where we want agents operating on the marketers' behalf that can keep up with the speed of agents operating on
[11:55] the consumers' behalf. We're just not close enough. We can't really wait for syncs in order for this to take place. And so you had the early days, right, of of maybe gen one SaaS where you had email providers pulling lists of data. Then you had a purpose-built solution
[12:10] with customer data platforms, which were great to to try to address the problem, but practically speaking could not really get over that hurdle of making copies of the data that you have and shipping it off to a third-party product, right? They sort of existed in the world pre-data lakes.
[12:27] Then we had the composable CDP era, which is a much, you know, improvement. You can actually leverage data that's in the warehouse, except the practical realities of actually using it means you actually miss out on things like governance and security, even schema
[12:43] mapping that you're used to when you're working entirely within the data platform that you're using. And so now we have this agentic era today. So we need something new. We think a fundamentally different approach is required to meet this era of engagement. We need unified profiles
[13:00] that are built onto your data foundation. We want to eliminate the silos without losing the governance and security that the teams that support marketing and building our customer data have invested so much time and effort in building. We also want interfaces that are designed for marketers, for business,
[13:18] non-technical users, right? And not only do we want it useful for those users, but we want to give them capabilities that allow them to engage with agents and sort of deploy agents on their behalf in a way with new capabilities that they previously weren't able to.
[13:33] Ultimately getting to the stage where one-to-one journeys can start building themselves. Right? Where they can react to customer signals and engage with customers as they go. And so what we introduced this week is called Databricks Customer Lake. Databricks Customer Lake is an agentic
[13:49] CDP that is embedded into Databricks that allows marketers and data teams to be equipped with a fleet of agents creating the perfect customer experience a billion times a day. We'll talk about what this looks like.
[14:05] First and foremost, there are really two core components to Databricks Customer Lake. The first is called profile agents. Profile agents, think of this as the set of agents that supports data teams who support marketing
[14:20] to go from raw customer data to golden customer profiles. Right? So all that work that goes into resolving customer identities, the schema mapping, the data models, all that work that needs to be done um can be automated through profile agents.
[14:36] The second is campaign agents. And campaign agents are designed for marketing teams to use. You'll see what this looks like in just a minute. Um and this is where you're going to see a lot of the capabilities that traditionally belong to a customer data platform, a CDP, with a lot more. And then of course
[14:53] native connectors that can connect to your marketing technology and advertising technology ecosystem. Everyone knows here who works with marketing that none of these tools can live in a silo. So let's talk about these a little bit. Um embedded, I can't stress this enough. I mean, the realities of trying to do
[15:10] anything outside of the data platform once you're set up within it is really difficult. Um so having a native customer 360 um that is built into Databricks is really valuable. Now, there's going to be a lot of cool capabilities here like
[15:25] native identity resolution. I'll talk about that in a second. As well as a third-party data and identity marketplace that can hydrate actual, uh, profiles with third-party data. It's democratized. So, this is going to look nothing like any Databricks, uh,
[15:42] interface that folks might be familiar with. This is a brand new UI that's designed for business users. And not just for business users, but for business users to work with agents on their behalf. And then we've designed this fleet of agents that are embedded through the application throughout. I mean, if you
[15:57] think about when every CDP that came before this was invented, all of them were invented before 2022, right? So, all of this happened before the ChatGPT moment. And if you were to build an application today, you'd think of it fundamentally differently, um, when
[16:13] you're considering an agent as a primary user. So, we think Customer Lake is really valuable for a few reasons. First and foremost, it's the only CDP that is embedded into a data platform and right there with your AI, uh, capabilities as
[16:28] well. It is also built for marketers and also supported by IT, right? So, this is happening inside Databricks. No more data copies, um, no more sort of, you know, changing of the data schema to make it work with a, uh, with an external vendor.
[16:44] And then finally, you have this ability for marketers and agents to work together and I'll show you what that looks like. So, we talked about those two concepts, profile agents and campaign agents. I'll quickly touch on each of those and then I'll show you what it looks like. So, um, since we're at Marketing Forum,
[16:59] I'm actually going to start with campaign agents. In the keynote, we started with um, profile agents. Um, but here really this is work that, uh, this is these are agents that automate campaign work. Right? And so, what does that look like? Well, one of the big steps in building a campaign is building your audience. And
[17:16] building an audience is usually, um, at a fundamental level the SQL query. But now we're sort of allowing us to do that not with a traditional drag and drop interface, although that's available. We also offer the ability to create audiences through prompts, right?
[17:31] Build me an audience of this type of buyer in this location that doesn't subscribe to my newsletter, hasn't made a purchase in the last 6 months. You can get really granular with these. We also power if folks saw this concept of infinity campaigns, which are these
[17:46] always on evergreen constantly reactive to individual customer signals as they are engaging with your brand. And you can monitor and manage those, connect them directly to your execution channels if you're using folks like Braze or others.
[18:03] You can understand your customers with some new insights, audience information, seeing where audiences overlap, seeing which audiences are performing well within each campaign. And then of course native connectors across your marketing and advertising technology stack. Really, really
[18:18] valuable. Now, on the data side we have profile agents. And these are the capabilities that automate and prepare that customer data for use by the business. Right? So, a lot of capabilities here. I want to touch quickly on this identity resolution piece. So, traditionally
[18:34] identity resolution is done when you you have two profiles that you know don't necessarily have a customer match between them. And the way you traditionally do this is through rules, right? If they sort of match this, you sort of set up rules in advance, some deterministic matching.
[18:50] Later we moved into probabilistic matching, which is helpful, but usually after that there's this human review step that you sort of eyeball check what you're seeing, making sure that these rules and and probabilistic matching is happening in the way that you expect. What we've done
[19:06] is we've inserted an agent into that just before the human review, which will actually feed the learnings back into the model and determine if the rules that you've set are right or need to be changed. We'll make those recommendation recommended changes, the marketer can
[19:21] approve them, and it'll just start working again. And we found in our testing that this leads to much better match rates. Um of course, long-term keeping that data in sync over time, mapping it directly to um to the various components that you might be working with.
[19:38] And then of course, getting to that stage where you're creating golden customer profiles. Now, the agents that we've built are designed and we'll show you what some of those so some of those look like, but everything from audience agents, personalization agents, advertising paid media agents, um lots of capabilities.
[19:53] And because this is part of Databricks, we also have Genie built in. So, Genie is powering a lot of these uh GenAI capabilities, but of course, you have, you know, some fun capabilities like a mobile app that a CMO can go in as the example that Rip gave um and said, you
[20:09] know, "Hey, how are my campaigns doing? How are things performing? Um which audience is performing best? Which one is most likely to churn?" And we're also launching with a really elite partner ecosystem. Um some great integrations that are live on day one,
[20:24] um and we have dozens more that are in the pipeline to to produce. So, really really great capabilities, um and we'll show you some of the unique integrations that we have with some of these partners, too. And then finally, um a number of customers that we're very proud to be working with directly, um and so we
[20:40] thank customers that help us sort of co-build this application, um but also validate the approach, making sure that this is something that marketing teams are actually going to benefit from, are using, and um and you know, we're we're getting a ton of great feedback from this. So, we're aiming to get to this
[20:56] perfect customer experience a billion times a day. So, let me show you a little bit about what this looks like. All right, I'm going to refresh my screen here and it looks like we're good. So, I'm going to start from here in Campaign Agents and I'm going to start with a prompt. Let's say I'm a marketer, a campaign manager, and I want
[21:13] to build a new campaign. Usually, this is a process that happens in meetings, across a series of days or weeks, but I'm going to go ahead and and set up a prompt here. Let's say create a campaign that targets
[21:31] business travelers for summer vacations. I'll hit enter and it's going to start building out, try to understand all the context that it has for my business, pulling in what I might mean by different sets of of words that I'm
[21:46] using and make some inference there. Um and what the first thing it's going to do is actually generate not the campaign itself, but this is where the human-in-the-loop piece goes. It's asking me a series of clarifying questions. Now, these are all dynamic, generated in the moment for the specific
[22:02] query that I had and the context that I created. The first question is "What channels would you like to use?" Well, I'll select Braze here. Maybe I'll add Google Ads, maybe LinkedIn Ads cuz I'm targeting business travelers. Then it says, "How many messages should the travelers receive?" And I can make a selection here, but I can also say to
[22:19] for Genie to decide for me based on what it thinks is best. And then when should this campaign end? Um and it's giving me some options here. It's offering design, but I can also insert, you know, if I think, let's say October 1st.
[22:34] And now I'm going to draft a brief. And what's happening now is um Customer Like is taking all of the data that I put in, it's taking everything it knows about my customers as it's working, and it's taking all of the business context that I have, and generating a campaign brief that I can now look at and understand how it's
[22:50] thinking about those things. So, what we have here is a full campaign brief. It's generating a campaign summary, the goals, the audience that it's going to try to put together, the goals that it set, some first pass of messaging, the types that it'll deliver, the plan
[23:05] for the messages that it'll send, sequencing guardrails, assumptions, right? All the things you'd expect from a solid campaign brief. Now, I can go in and edit all of this if I want, but I'm going to go ahead and trust it for the sake of this um capability set. And what you're seeing now is, based on
[23:22] that campaign brief, Customer Lifecycle has built all of the elements of a campaign that I would need. It's created an audience. I can edit or change that audience, and I can view what it looks like. Now, I could have built the audience from scratch this way using the standard sort of interface that you might expect from
[23:38] a CDP. Um and I can edit it in the same way. Um and it also includes the exclusions that it has. Or you can see the goals that it's setting, and I can view those. The attribution window, the channels that it's using. Remember I said Google,
[23:54] Braze, and LinkedIn. Um here are some of the operating procedures, the guardrails, right? So, it's going to make sure that there's no overlap on the guardrails that I put, opt-outs, it's not going to hit those, some quiet hours it's going to recommend, and then you can save the draft.
[24:09] And so, this will take me to a dashboard of how that campaign is going. This one just launched, so there's nothing in here. Let me go ahead and open one. So, you can see what this looks like. And here you have a great view of this Infinity campaign running. Um we can see the activations, how they're happening,
[24:25] how many that it's hitting at a given time, what your volumes are. You can scroll down and even see things like, you know, the messages that it's receiving. And if I look into this, you can see what these look like. So, this is actually using an
[24:40] integration from Braze. Um Braze has a wonderful bidirectional API that it uses for its message creation, and so it allows us to see what these messages actually look like. This dynamically generated message just for this campaign, just for this audience that we just created. And And can see what that
[24:56] looks like here. And then of course you can see all the things that you might expect from a customer data platform, the syncs, all of the connection channels that you might use. You can go into audiences and create a new audience from scratch just like we said.
[25:11] And monitor your campaigns on a high level every day. Now, what's cool and the last piece that I'll show you is that you can attach context, right? And so the practical realities of marketing is yeah, we have all this data stuff, yeah, we work in all these tools. Sometimes we do campaign planning in PowerPoint, right?
[25:27] Maybe not sometimes, maybe all the time. Um and so it'd be great if I can just attach a PowerPoint. Well great, if you have a campaign plan for the year that you've built out, you want to attach that to your campaign and have that inform the brief that Customer 360 builds, wonderful, you can do that. And
[25:42] you can connect it to other tools like Google Drive, uh maybe if you're building out a dashboard or something like that, you connect it to those and goals that you've set from previous campaigns. So that is Customer 360. And with that I'm going to pass it over to Katie Yuan who's going to introduce our next session. Thank you very much.
[26:29] All right, welcome everyone. I am super excited to be joined today by two leaders from HP who are shaping the future of customer experience, Horacio Miranda, SVP of marketing transformation, and Kumar Ram, global head of marketing technology and AI enablement.
[26:44] HP has been a global technology leader for over 80 years, serving millions of customers in 180 countries across both consumer and enterprise markets. So, today we're going to discuss how HP is transforming customer engagements with AI, HP's vision of the future of work, and
[27:01] how Databricks and Customer Lake are helping HP take steps towards that future. So, to kick us off, Horacio, would you quickly introduce yourself as well as what marketing transformation means for HP? Yeah, thank you. Hey, good to see everyone here in the in the room. Great
[27:17] participation during the event. Uh some of you may know HP, some of you may not. So, yes, we've been in 87 years in business helping customers through many different technology transitions.
[27:34] Uh today at HP, uh you may have a teenager uh son or daughter that maybe using one of our gaming devices uh that perhaps convince you that it was a device for for work at school. Uh also, the the printers and PCs that
[27:52] you may be using at work, uh the collaboration rooms that you may be using for video conferencing your companies, workstations in uh with NASA as they took astronauts to the moon, uh 2 million dollar digital digital
[28:09] presses and 3D print printers that are changing the way uh companies do manufacturing. So, for many years we've been do we've been in the business of uh engineering the way work happens.
[28:25] And from a marketing perspective, uh well, we have a very large organization with more than 1,000 employees uh globally. And my job is to help modernize the systems of work.
[28:41] So, my job is to help with my team engineer the experiences, bring the capabilities so that our employees can do their best work at HP. So, so, so that's on
[28:57] on our agenda for our transformation. Awesome. Thank you so much for sharing. And Kumar, could you also introduce yourself and what your role encompasses at HP? Absolutely. Great to see everybody and my name is Kumar Ram. I head all things marketing technology and AI enablement
[29:14] under Horacio. Um what does that really translate to? We're the single one-stop shop for everything technology, everything data, and AI at least within within the context of marketing.
[29:29] Um underneath my team, I mean or rather within my team, there are five different functions that we support. And these are I mean historically, I mean we've had this all over the place, but one of the things that we've done in the last several years is kind of bring everything under one umbrella. And the five different functions are first data,
[29:46] infrastructure, everything that we do in the world of data. I mean that's why we are here at Databricks, right? Uh everything we buy, every type of data that we buy, whether it's from Acxiom, D&B, you know, Bombora's, etc. The integration of and maintenance of all of that data. Of course, the infrastructure that goes
[30:03] with that, whether it's AWS or Azure or whatever is that operating platform. So, managing that. So, all of that sits under Luis's one of our data engineering lead who's also here. But then, the second function that I support is all things web. Every piece
[30:19] of web technology, whether it's Adobe Analytics, Google Analytics, site tagging, media tagging, it all sits within the second function. The third function is truly martech platforms, which includes things like Eloqua, journey orchestration tools like Converter, etc. So, that's the third function.
[30:34] Fourth function, which is very new, is content, right? Again, we spend tons and tons of money with agencies in terms of content production. How do you scale that content production using AI, especially with tools from Adobe or Figma or whoever it is? And how
[30:51] do you kind of integrate that into the process? So, that's the fourth function, which is very new. And one of the programs that we're spearheading right now is the content transformation, which is grounds-up reinvention of how we think about using AI for content. And these are the four verticals. But then, I have
[31:07] one horizontal function, which is called as AI enablement, which stitches all of these four together to essentially make the journey of our end users as easy as possible. Thank you, Kumar. And because we can't have a marketing conversation without an
[31:24] AI conversation these days, um Araseli, I wanted to ask you. So, as you look across HP's marketing org, what are the biggest shifts you're trying to drive today with AI in marketing? Um I personally been in this uh roller
[31:40] coaster journey over the last 2 years. Because in our organization, maybe that happened in your organizations, the journey of of AI started with uh leadership mandates.
[31:55] Right? Hey, the technology is here. We need to use it. We need to move fast. We need to be behind our competitors. Uh let's give Copilot licenses to everyone. One, let's move faster. Let's reduce costs.
[32:13] Let's drive productivity. Let's automate jobs. And I think 2 years later, I just real- we just realized I realized that that was the wrong uh the wrong proposition.
[32:31] And uh because at the end, the work is done by people. So, the way we have shifted and pivot our strategy is we want to enable our teams to do their
[32:48] best job. Hey, we do a lot of external surveys on the topic of work relationship and we find out that globally across many organizations only 20% of knowledge workers have a positive relationship with work.
[33:06] Uh I would like to get a uh raise of hands. Who here knows the engagement scores of your organization for the current year?
[33:21] Very little. And are they going up or are they going down? In my organization, they were going down. Okay? And I think this is a very important topic because that is the big problem is technology moving too fast,
[33:39] but in reality the organizational readiness to move and and bring our people along is not that fast. By the way, I do hope that you have a lot of Gen Z workers in your org. Uh across the globe, 30% of workers are
[33:55] Gen Zs. But guess what? 50% of them has a side gig. So, their engagement with work is at all times low. So, we want to use this uh transition opportunity
[34:11] to use it on the advantage of how do we redesign the value proposition in and we are using a very simple fundamental question. What will our organization look when 100% of the employees can do their
[34:28] best work? So, So, are using that as our design center right now to evaluate what is the type of job and tasks where we can help employees uh uh leverage the technology and the
[34:45] capability. And based on that design, what capability we are going to architect, put in production in production, and deliver to our employees. So, it's a a very important shift that we are doing uh as we speak, but I think it's the right shift that any organization should be doing.
[35:02] Thank you for sharing HP's mission. So, Kumar, from your perspective, from the MarTech and AI perspective, what's required to make this transformation a reality? And where does that traditional MarTech stack start to break down? Uh absolutely. Uh that's a great question, Katie. So, I know Horacio hit
[35:18] up on a couple of key things around effectiveness and efficiency as some of the core principles of transformation, right? When I take a step back and look at this, part of my job is to kind of build the ground or solve for the ground truth, right? If as a technology
[35:34] implementer over the many years, finding that balance in terms of what Horacio's vision is and the leadership vision is within the organization, and being able to translate that into something real is always a struggle for a lot of us. I'll I'll give you guys a little bit of history, right? I moved to the Valley
[35:52] back in '92, '93 uh coming out of grad school. And when I moved, um I've been jumping jobs every couple of years, every 3 and 1/2 to 4 years, roughly. 4 years is my average kind of job going from one job to another. And now you've been at HP for 10.
[36:07] 15. 15. Okay, good. This is my 16th year. So, the thing is I when I joined HP, I my objective was like, "Okay, I'm going to spend 3 years here, and I'm going to move on, and I'm going to go find the next big thing, right?" Interestingly enough, that didn't happen. Why?
[36:23] Most importantly, what I realized at HP, working with people like Horacio and other leaders is shared vision. Shared vision is such a calm such a core component that's needed both laterally as well as vertically. And why is that important, right?
[36:38] Uh when I look at my my conversations with Horacio, we're always fully aligned. Not necessarily in terms of how we solve the problem, but at least in terms of the strategic vision on the big problems to go solve for, right? And Horacio aligned with the CMO and that
[36:56] you know, vertical alignment is one of the core things. Then, on the other side, the horizontal alignment with my peers, which includes IT. You know, having that shared common vision with IT, having that shared common vision with like finance, these are all core
[37:11] constructs of what drives an effective, efficient, you know, technology implementation in any organization. And that's something that I found as being sort of unique to my my years at HP. But the second thing is empowerment,
[37:28] right? One of the things that we preach and I always while I lecture my friend Luis all the time on how he needs to think about the problem, right? He's empowered to make a lot of the decisions on technology implementation when it comes to data infrastructure, etc. So, essentially that and Horacio
[37:45] empowers me to do my job. He gives me the freedom to go execute. I think that is sort of the two key takeaways. The shared vision and the empowerment are really really core to any successful technology organization in the valley.
[38:00] So. Awesome. Thank you for sharing. That alignment the importance of that can't be understated. So, Kumar, I wanted to continue and ask you. So, Databricks announced Customer Lake yesterday. It's a new Agentyc CDP that's built natively in the data foundation. We heard a lot about it from Michael. Um HP has
[38:17] actually been testing it in the recent weeks. So, I wanted to ask you um, stands out to you about Agentyx CDP and how do you think about it as the evolution of the category? Absolutely. And I'm going to take a few steps back in terms of how we're thinking about this problem, right? Um, we've had a CDP for several years but
[38:34] Michael, you know, um, and and team essentially when they came to us, one of the first things the common argument was there is a different way of looking at this problem. So, B2B was one of the big problem areas for us, right? When we look at it from a CDP perspective.
[38:50] Every time we had to go build a segment, we had to bring in newer types of data. One of the first things that we had to do is go write newer types of SQL query because that was fundamentally one of the broken constructs of any CDP. I mean, why do I have to write SQL queries every time I bring in newer forms of
[39:06] data? That's one of the first things that we saw with the Agentyx world. The second thing, I think Michael talked about profiling, right? And profiling was yet another problem, especially when you talk about B2B audience profiling. There are so many different factors of
[39:22] data that we need to bring in like technographic, um, uh, you know, firmographic data. You then you mix it with the demographic data on the B2C side from the consumer side. When you mix all of these different types of data, the flexibility of being able to do this, right at a prompt, is sort of what
[39:39] drives the change, right? So, the being able to do the profiling and do the campaign setups on the fly. And then when you truly have to do personalization, writing SQL queries is not the way. And that's one of the core constructs that we found working on the customer lake to be really really beneficial. Something
[39:56] that used to take us weeks, you could get it done in less than half a day. Of course, we're still testing it. I think there's a long way to kind of go, but, um, that's just progress in the path. Awesome. So, having that stronger foundation of data and AI is super important, but Horacio, where do you see, um, that foundation of technology,
[40:13] um, creating the most meaningful value for HP's marketing organization? I have these very simple two by two framework in my mind that that we're trying to use as we evaluate the opportunities.
[40:29] And again, I need to take it back to the employees, right? And it's like as we look at the jobs and the tasks of our teams across content uh activation, we have a communications team, we have marketeers in the markets in the
[40:45] countries. I'm thinking about two categories of work. Uh the first category is what I call the low stakes uh work. And over there, we just needs need to clean the path.
[41:01] Hey, we are hiring a lot of folks on the promise to do the value add work, but I'm sure that happens to you. Big chunk of our work is also the low value, high friction, repetitive work that needs to get done. And if you have a lot of that work every
[41:16] day, you don't have the time to think the value add work. So, that category of work that is the repetitive, the the time sensitive, I want to see use cases there in terms of how we can do substitution. Cuz
[41:33] that's the work that employees don't want to do. Okay? The other category is the value add work, right? We are hiring marketers to do the amazing work of creating great stories, impacting business, growing businesses.
[41:48] And that's the area where we want to raise the bar. So, on the raising the bar, we can also use AI. But over there is not to substitute the human labor, is to augment the strategic thinking.
[42:04] Right? Because over there is not about saving time, is about driving impact. So, the way I'm using that framework with Kumar and another team members is that's where we evaluate the type of work that we want to go after. Now, the question is where is technology ready
[42:19] now to do value realization soon versus where we are still exploring uh identifying partnerships or or capabilities that we need to go and test against that framework. But, I think to me what was very important is to shift
[42:35] from what I call candy shopping across the the tech stack and all the beautiful things that uh many companies are creating to really go to with a very precise list to say, "We want to help employees on this area. Show me how we can do that
[42:52] together and and create those outcomes that we look for for our employees." So, so that's where we're going right now. And I think in the context of of this Agentyx CDP, I I I think you can see the value both on helping with the repetitive tasks
[43:08] but also helping with a strategic thinking. And yes, I don't want that marketer to just copy-paste, push the button on Jenny to go to campaign execution. I want that marketer to think hard if that strategy is the right strategy and is what we need to be doing
[43:23] because the big mistake that we can make is to outsource the decisioning and the accountability of the employee to achieve the business results. That cannot be outsourced to AI. That needs to continue to be uh um on the center of the job of the
[43:40] employee. Yeah. Makes a ton of sense. And on the topic of automating busy work, so Kumar, I've heard you say the phrase before from zero copy to uh zero code. So, what does that mean in practice? There's a trademark to that, so just as an FYI. So, I've already trademarked it. So, uh the
[43:56] I'll take a few steps back. Practitioner view of things, right? When Luis and I originally had the discussion around building the data lake about 3 years back, was it? Right? Roughly? Every conversation that we had was about TCO, total cost of ownership. I'm going to be able to reduce AWS cost. I'm going
[44:12] to be able to reduce all the Redshift cost, la la la. We had 22 different data assets. Most of it around Redshift. Massive amount of money being spent there, right? So, the argument actually became, okay, at least the initial pitch to the
[44:29] leadership team was around TCO. But then, I was also running the data sciences function at the time. So, one of the first things that we did was to go and survey all of the data scientists. We asked, okay, hey, you're an expensive resource for this organization.
[44:45] How much time do you spend collecting data? How much time do you spend cleaning the data? And most importantly, how much time do you spend actually analyzing it? So, guess what the first two were?
[45:00] Roughly 70%. 70% of their time collecting data because we had 22 assets. Go pull, go pull, go pull, and bring it into one place. Mix the data, clean clean it, and apply your own set of rules, and then finally get your models or get your algorithms
[45:17] built on it. Now, fast forward, that 70% of the cycles have gone away. We're predominantly now with every data that we use from a marketing perspective sitting within um Databricks. Those two functions we were able to
[45:33] eliminate right out the gate. But here comes the next problem. I go to my analysts just last week, and I ask them, okay, where do you spend a majority of your time? And I'm not talking about data scientists. Data scientists are a slightly different breed. They're still going to be building algorithms, etc. They're still going to be writing custom
[45:48] code. But I went to the analysts. They're spending 60% of their time writing SQL code and converting that into PowerPoint. So, this is sort of my whole thesis of going to zero code with something like
[46:04] Genie, that's the step in the right direction for us to be able to eliminate this whole idea of writing SQL queries the way to go versus asking something like Genie to be able to produce the insights that you want. And again, as an extension to this,
[46:20] this is not just about the structured data that we get from media or dot com or whatever it is of the conversion data. We're also planning to bring in a lot of the research work, the qualitative research that goes in. Bring that into the same infrastructure, so you can now combine these two in the
[46:35] context of the questions that you want to ask. Goes back to the context discussion that Michael hit upon in the in the customer lake. Yeah, that's awesome. And to close us out, Horacio, what's one takeaway that you would like to leave with leaders in the room that are thinking about their
[46:51] own transformation journey? And before I say takeaway, the Luis in mention is Luis Alonso, part of Kumar's team, who is leading our data uh strategy and capability. But I think to me that's a great example of of the value realization economic of it all, right? Let's reduce the repetitive work.
[47:08] That's time extraction so that Luis can go into the value-added side, apply that time to figure out the identity AI CDP that that we are trying to deploy in motion. That's about value creation, revenue impact.
[47:24] That's the way we need to do our plan. So, I just have like three takeaways that I wanted to leave with the room. And and this is of course based on my experience and what we are doing. You need to make a choice. Your organization needs to make a choice. Right? Do you want to use technology
[47:41] to eliminate work, automate work, eliminate jobs? Or are you using technology to get business advantage, elevate the the value of your workers, and the ability to further do the best job? I think that's a choice that each one of
[47:57] our organizations has to make. So, I have three principles in the way I'm going about it, in the way I'm we are designing this discussion both with the CEO, the CMO, and the leadership team. One is employee at the center of the design.
[48:12] We have to have a design principle. Uh second, the the work is to redesign the work itself, not to deploy tools. Of course, we're going to deploy tools and agents. But, the principle needs to be the
[48:28] redesigning of the fundamental way these teams needs to work. Okay? And final, uh I think we need to measure the value, not the activity. Right? And and I think that's the hardest part because
[48:44] if we own these transformation initiatives, we need to not only define that value, but be accountable to the delivery of the value, right? That we are going to be extracting for our organizations. And I forgot to add a fourth one that I think is super super important. We Of course, context is is is is is
[49:01] important. Data's a foundation. Without that, we cannot move into experimentation and scaling of these capabilities. But, I really believe that the operating system is the culture. If you have a broken culture in your
[49:17] organizations where team don't talk to each other across different groups, that's not going to be solved with agents or with these tools. Okay? If you don't want to lie to face the sales organization because you're afraid that they're going to say, "No, your plan is wrong."
[49:32] You're not going to solve that by working alone with Genie. So, I really challenge And and this is something that I'm thinking from our organizations' perspective, what is the real culture that we have right now? And is that culture the culture that we need to be successful as
[49:48] we take these giant gigantic steps to modernize the way we work. So, so you know, I I really invite you to think about that both whether you are on the marketing suite or all of our partners you are in the job of enabling marketing
[50:05] to do their best job, right? Their best work. So, I really challenge folks to think about those three areas. Yeah? Thank you so much for sharing your insights. Thank you both Roncio and Kumar for being here and thank you all.
[50:31] Okay, please welcome to the stage Dan Morris, head of go-to-market for customer lake and industry solutions. Hi everyone. So, Dan Morris here, the go-to-market for customer lake and marketing solutions.
[50:48] And today I'm going to talk to you about identity marketing in practice. As an industry, we often hear some common refrains, right? We're going to be managing a team of agents, we're going to be focusing on strategy, but seldom do we ever really talk about what this looks like in practice on a
[51:03] Tuesday morning for you as a marketer, for your team if you're an executive, or your customers and stakeholders if you're in data and technology. And so, I'm going to unpack this today going a little bit more on a technical level what this actually looks like in
[51:18] practice and what's needed to bring it to life. So, to do that we have a subscription video on demand service that we're going to use as a walking scenario. The goal that we're going after is that we want to win back lapse subscribers and the audience is anyone who's churned
[51:35] in the last 90 days, whether it's from a recent price increase, failed credit card transactions, or just engagement being low and not being too surprising. If this were happening today, you know the drill, right? You'll start off by building your audience. So, you'll pull
[51:51] some data, you'll analyze some churn signals, you'll iterate a little bit, and then you'll move on to creating the brief. So, at this point, you'll then go and create some cop some copy, you'll work with design on some variants, and then finally, you'll
[52:06] move on to building and launching the campaign. At which point, you'll then work with your analytics team. So, data will come back in, you'll get some insights, you'll iterate, and hopefully capturing some lessons along the way. Now, all of these steps are absolutely necessary.
[52:21] The challenge is that in many cases, you're working across five different teams. You have handoffs that take several days at a time. With agentic marketing, this looks a bit different, right? So, as a marketer, you're still going to start off by defining the outcome. So, in this case,
[52:39] we want to win back lapse subscribers within 30 days. And you're still going to define the audience. So, in this case, you know, customers who have lapsed within a certain time period for all the reasons that I previously mentioned. But now, in this example, we want to use
[52:55] agents to help us build the copy or write the copy that we want to use to engage with each one of our lapse subscribers. Now, instead of going and handwriting that copy, handpicking some images, instead, what we're going to do is we're actually going to find a campaign
[53:11] rubric, very similar to what you saw Michael show a little bit earlier. And so, what this means is that we need to define in very specific terms what good looks like. So, what type of emotion should we invoke with our messaging? What type of claims are are
[53:26] we allowed or not allowed? What type of tone is on on brand? And once we have this, we have our outcome, we have our audience, we have our campaign rubric, this is where we can now lean on agents to help us build and execute our plan.
[53:43] But now let's talk about what this actually looks like. So in this type of scenario, we have three types of agents that need to work together. Each of these agents specialize in different things. We have a planner, a grader, and a writer.
[54:01] The planner is going to take all the different contexts that we we provided. So it's going to go and take the campaign rubric that we discussed. But it can also take something like an offer policy. So what type of offers are we allowed to give to different types of segments? What type of caps? What types of constraints? We might also have our tagged asset
[54:18] library. So what are the images, what are the hero moments, what are the reference titles that we're allowed to use that have been already approved by brand communications? It's going to take all this different context and it's going to generate a plan for each cohort that's identified
[54:33] within the broader audience segment. Now what this plan is just a document, right? So it's inspectable, it's debatable, it's ultimately yours to approve. But the onus isn't just on you. Because you took the time to define your
[54:50] campaign rubric, you can now leverage a grader agent who's going to go through every bit and piece of that plan for every single cohort within that broader audience, and it's going to do checks. It And if anything doesn't pass that check, it's going to kick it right back to your planner to try again.
[55:06] And so what this means is that you as a marketer, you don't have to deal with first passes or incomplete passes. Everything that you get will have already passed the rubric that you defined. And if it doesn't pass the test, uh or if you need to go back and refine it, you can.
[55:26] Finally, we move on to the writer agent. So the writer agent is now going to go and take that plan, and it's going to apply it directly. It's going to use the plan, it's going to use the subscriber data, and any other context that you provided to write those personalized messages at scale for each
[55:42] and every one of your lab subscribers. Of course, this all gets surfaced to you for that to to then approve uh before you go and launch the campaign. And so, at this point, you've kind of gone through this process. It's pretty
[55:57] neat, right? Cuz we've gone and wrote these personalized messages for all, you know, million users that we want to interact with. But, the beauty of it is that all of this will be based on templates. These templates are designed to the cohorts that are identified within your
[56:13] audiences. And what this means is that you as a marketer don't have to go and review a million wildly different emails, but at the same time your customers are going to get truly personalized one-to-one uh messages that they're going to be that much more likely to respond to. And should you want to inspect it a
[56:28] little further, of course, every every uh pass, every every piece of it is going to be logged, so you can go back and fire away with questions as you need to. This is cool cuz we've now scaled a pretty heavy part of our process, and you can see how this applies to pretty
[56:44] much any other part of the process, and you can you can do that as well. But, if you ask me, there's actually a part that's even better. So, at this point, data is going to flow back into Databricks just like it does today. But now, we can have an analyst agent.
[57:00] That analyst agent can now churn through all the different signals that we're bringing in. So, it could look at it by segment, by offer type, by message type, by timing, and it could go and find all the different things that should be taken into consideration and basically write it down into another document.
[57:17] And these findings can then be encoded into other documents that you ultimately use when you were planning your your campaigns. So, again, tying it back to what Michael showed you, these become additional pieces of context that you can provide. And so, you might have a messaging
[57:32] playbook, for example, that accumulates all the lessons learned over time in terms of what works and what doesn't work. And similarly, the offer policy that I mentioned earlier, you might have recommendations such as maybe not provide a free month to people who have
[57:47] a higher return risk. But there are some insights that require looking across campaigns or really require a longer time horizon, right? So if we're thinking about something like a win-back campaign, you might have some
[58:02] some customers who come back, maybe they accept your free your free month, uh but then they churn, right? And so in this case, you get your conversion, but it's not really durable recovery. And so that's where you want to have these longer horizon agents who are able to come in, they're able to ask
[58:19] questions that are independent of that initial analyst. So are we are we really optimizing for long-term? Are we optimizing for the right metrics? And as it does this, it can now self-correct, right? It's going to find things that just weren't possible to find with a shorter time horizon, and
[58:35] it'll correct that for the next go. And this is really important cuz what it means is that when you're using agents to move faster, your gains are going to be very linear, right? You're going to save an hour here, a day a day there. But when you're using agentic systems
[58:51] like this, your results are really going to compound. All the different artifacts that you have is going to provide better context, which ultimately will uh you know, exponential gains in terms of your your metrics. And so now you might ask, how do you get started? So whether you build or buy,
[59:07] the foundation that you need is is the same, right? So as you've heard, I think a few times today, everything ultimately comes down you need that unified customer context that you're able to to build, right? So your behavioral data, your transaction data, your demographic data, need to bring that all together.
[59:23] But you also need your business and marketer context. So things like brand guidelines, but also when you get into marketers, things like judgment and taste and all of all of your different experience needs to be encoded in documents that you can ultimately use as context.
[59:40] And then lastly, governance. None of this is possible if you don't have proper governance in place. So, you need to ask, what type of guardrails do we need? Where do we need humans in the loop? And once you do this, of course, you can then accelerate a lot of your agentic marketing initiatives.
[59:55] If you do want to simplify it a little bit further, a lot of the complexity that I went into with the agents, that's where I'd encourage you to check out Customer Lake and Embrace Infinity Campaigns. So, with that, I will now hand it back to Michael.
[01:00:37] Liz Dobbs, the head of marketing technology at Databricks, and our panel moderator Scott Brinker. Thank you.
[01:00:59] All right. How's everyone doing? So excited for this closing session here. Have some amazing marketing marketing technology leaders talking about this journey to agentic marketing. So, we have Liz Dobbs, who's the AVP of MarTech data and growth. Sounds like you got all the really
[01:01:14] interesting things at Databricks. be interesting. Now it's gotten a lot more interesting. Chris Wisinger, who's the chief product officer at Epsilon, and uh Jay Malalapadi is the global director of data science, customer, and marketing strategy. So, welcome. Um
[01:01:30] just to help people get a feel for you and your backgrounds, I'm curious, what was your first job? It professionally or just Okay, yeah, cuz actually when I asked Chris this, he's like, "Well, I did grocery stocking at 14." So, yeah, let's
[01:01:47] let's go with first professional job. All right. Um I would say my first professional job was actually just a marketing coordinator at a software company. Um and, you know, one of the first things I was actually tasked to do in my internship was uh actually manually enter cards from events into
[01:02:04] our CRM, which was Salesforce at the time. So, literally they'd come back from an event and hand me a pile of business cards, and I would manually enter those into the CRM, and I've been on a journey ever since to never do that again. That's the old-fashioned data ingestion. Okay.
[01:02:21] Well, same for me. So, I'd rather talk about my first job. I think it was uh stocking grocery stores when I was 14 or something. I think I had that in common with you, I believe, or you're a bagger. I was a bagger. I aspired to stocking. Oh, okay. Well, it wasn't as glamorous as it sounds, let me tell you. Um yeah, first real job, I would say uh
[01:02:38] so I was at a consulting company, and we were working on a variety of things. One of the more interesting things, totally non-marketing scenario, but doing uh back in the MP3 days, if any of you are familiar with like audio sample rate conversion processing. So, I was much more on the software engineering side at
[01:02:55] at that time, and had a lot of fun uh in the heyday of the uh beginning of uh digital audio. Okay. I don't have such interesting stories as the other panelists, but my first job was a data scientist itself at a this transportation company called XPO.
[01:03:12] And one of the interesting problems that I was trying to solve was um at XPO we used to get all these bills which were handwritten bills with address information, customer information, all this. We were trying to scrape that data to match it into what we have in CRM.
[01:03:30] That was Thinking back now how the technology has evolved and all and the type of problems that we were solving then versus now it's like we have come a long way. Yes, indeed. All right. Well, actually Jay, let me start with you. We'll go back the other way. Um
[01:03:45] What has been your experience building marketing around a universal data layer like Databricks? Yeah, uh definitely. I think the think the phrase that Rick was kicked off in the starting where he talked about the unified customer data foundation. I think that resonated a lot with me.
[01:04:02] Especially when we started out, we built our marketing platforms or most of the workloads around the tools that we were using or the services or the agencies that we were working or the channels that we are targeting around email or push or anything. Now that completely flipped where we
[01:04:17] have the data first and the intelligence and where we focus more on unifying all our customer data sources and then we go out and serve across these different platforms. So, that that has been a really empowering experience for us and the way we go forward.
[01:04:34] Chris, you've seen a few of these. I have. Yeah, so if you're not familiar with Epsilon, we build martech and adtech solutions and and platforms for our clients to optimize their marketing outcomes. And I'd say in the last couple years, um you know, working with with Databricks in particular, uh we've been
[01:04:49] trying to move the marketing execution operation, our identity, our data, our activation capabilities closer and closer to where the data originates or where the data is stored as a source of record. And so, uh that fits very well with the concept around customer lake in
[01:05:05] particular. So, it's been uh it's been a good partnership. You have the home field advantage on this. I do. Um and you know, I think it's it's the transformation we've really tried to go on in in our marketing technology journey upon the ingestion, everything's in Databricks, is really how we enable
[01:05:20] ourselves and our marketers to interact with the data. So, I think as Rick mentioned on Genie, something that it's not just about getting the data, activating the data, it's I think it's getting deeper and deeper context and understanding of not just where the data's coming from, the possibilities of how we could use it, and really where
[01:05:36] the next best action is coming from. And I think, you know, we've been talking about this space for so long, this next best action thing, that it actually feels tangible now for one of the first times in my career. Nice. Nice. All right. Well, then let me start with you on this. How has that led to any changes in your
[01:05:52] marketing tech stack? So, we build on Databricks. Big surprise there. Um, no, I mean, I think it's you know, I think we I I It's funny. I I joined Databricks about 6 and 1/2 years ago. And we had so much marketing technology
[01:06:08] for a company of our size. You guys would honestly never believe it. It would never happen today. Um, and so really really haven't like fundamentally changed our tech stack dramatically over that time. I think we're actually just getting into this new frontier now, which is like, you know, the SaaS versus app
[01:06:23] debate. Does like, when does it make sense to bring in a SaaS provider to solve a business problem or business pain? When does it make sense for things to build ourselves now that the cost of building is is so cheap? Um, and so I think what we're kind of on this new frontier now of of the software to app,
[01:06:40] what is that right friction point? And where can we find things in our process that are kind of isolated, that we don't have multiple dependencies on, that we can start experimenting and piloting there? So, I would say to answer your question, not dramatically different quite yet, but I think where we're thinking about going next is going to be
[01:06:56] the really interesting piece to move forward. Right. Um, so I would say the biggest change has been our focus on uh enabling choice, additional choice for for our customers. So, if that choice means multiple clouds, if that choice means
[01:07:12] multiple database platforms or lake platforms, um we're we want to try to be where they are. Um the other thing I would say is that building on, you know, you you heard I think I mentioned this to you last week when we were talking is metadata's cool again or context is is
[01:07:27] now uh is the the the the better way to say it. Um but there's so much work that either has been or will be put into improving building context. We saw the Genie ontology uh releases this week and, you know, definitely excited to see what that can
[01:07:43] can add. Um but we want to be able to build upon that and not reinvent that. So, in terms of, you know, leveraging uh better marketing outcomes, understanding more about your data, that's really what we want to enable. Yeah, this is uh especially
[01:08:01] I guess much more beneficial for us. Uh for most of you who don't have much familiarity with Circle K, we are again a global company that's spread across 27 countries operating in 30 business markets, primarily rolled up as an M&A acquisition, so it involves
[01:08:16] a lot of different sub markets with different tools, marketing uh services, and there is localized operations as well as global operations. So, you end up with at some point 100 different tools. And similarly data sets and all. So, now
[01:08:32] with the whole Agentyk era with the data set and the Databricks coming into this picture, we can now see a direction where we can consolidate or optimize all of these tools. And the best part is the business or the stakeholders are also very interested to move in that direction.
[01:08:48] Have the champions from them to adopt and able to leverage these capabilities. And we've also seen a shift in how we treat our data. Earlier it was all about this individual project or how we approach a particular problem, but now we are focusing on building data
[01:09:04] products that can serve not just one channel or one, you know, market, but as a whole all the 30 markets that we have. Hello, I'm going to come back to you because you prefaced my next question here, which I was going to ask, you know, you This is Yeah, it's pre-loading, pre-fetch. Um
[01:09:21] your take on build versus buy and sort of how has that changed or is changing? What sort of factors do you consider in making that decision? Yeah, I mean, I think early on I think everyone in this room wouldn't be here if you didn't love building. I think this era we're in
[01:09:37] right now, the AI era, it's for the builders. And I think the next era of people who step into the marketing technology field are people who are going to be curious and want to build and get their hands on keys. And I think it's such a fun time for that. Um but you know, as my time at Databricks has
[01:09:52] grown and the company's evolved, we've moved much more from the build to scale. And I think what you had mentioned here is the interoperability of our data, the platforms we have to support, kind of the complexity is much higher. So, you know, as much as I think our team and myself have a passion for building, the
[01:10:08] you know, the the buy and the scale and the you know, support we need to run an enterprise company of our size, obviously there are tradeoffs. So, um I think the the potential for build is is huge right now and I think we're just starting to scratch the surface there, but I think, you know, in the in the
[01:10:25] businesses and the company we support, scale and how we can kind of rely on these systems are still just as important as they've ever been. Yeah, I think that's a great point around building. I mean, I think that's that's a natural fit for me personally in the role. Um many of the teams,
[01:10:41] that's, you know, that's where many of you I'm sure are going to going to fit in um as well. I would say though, I I'm not a subscriber to everybody's going to vibe code every SaaS application ever made. Um I don't think that's the best use of opportunity cost and and time. Um so,
[01:10:56] I'd say, you know, focus on building the things that are differentiating for you, you know, strategically differentiating for you or part of your core competency. And those you should definitely go about building. Um, so that's what I would focus on. I'll echo whatever Chris mentioned, right? That That's where our focus right
[01:11:13] now is building the things that act as your competitive advantage with leveraging the technology and AI and all the platforms like Data Bricks that we have right now. That's where our focus is. While buy anything that can be commoditized. It can be campaign execution or it can be simple journey
[01:11:29] orchestration, any of those tools. But building like core capabilities around intelligence to be understand the customer, where does all that fit? That's more on the build side. I think just as a quick follow-up to that, like, all right, so you largely have these other the systems that have been in your tech stack, still
[01:11:45] absolutely using them. Wondering if the way in which you interact with them has changed a bit though. How much of it's through the application interface versus how much are you now accessing it through other AI or Data Bricks genie interfaces, things like that? Yeah, you know, I I think it's a great
[01:12:00] question. Um, and you know, if you don't have an MCP, you should really start building one. Um, no, but I mean, I think it's interesting cuz I do think, um, you know, when we had Hershey up here talking about HP and their journey is like marketers do spend a lot of time context shifting between systems. Like
[01:12:17] just the tool, the tickets are in Asana or you know, Workfront, whatever tool you're in. Then you have got to go to this system to find the data. Then you go to this system. So I do think we are still, you know, spread very thin when it comes to getting our jobs done every day and especially for marketers. Um, that I think a lot of this evolution
[01:12:33] will be can we consolidate the interfaces? So the AI interface that a marketer comes to, can we bring the tasks and data to them versus expecting that person to go find and know where to go for all of these things. So I think that's exactly right. I don't know if the technology, you know, is
[01:12:49] fundamentally going to change. I think a lot of this stuff is commoditized. Like there's not a huge value in building a ton of these things. But, I think how we use them and how we interact with them as marketers to get our jobs done, I think that's where AI will have a huge competitive advantage to start building these roles-based applications that
[01:13:05] make, you know, someone who's writing comms or someone who's doing design have a very different day than what they have to do today. Anything to add? Uh yeah, maybe. I mean, we we've been focused more on uh an intimate internal use cases, I think uh around what you were talking about. So, instead of our
[01:13:22] our products and platforms that you go uh to market directly to sell, and we've been working on, you know, getting the contextual alignment correct and and getting, you know, the all the various ontology-related items uh set up. Um it's not an easy uh outcome. It it definitely takes time to go through
[01:13:38] that. And so, we've more recently started using Genie for some internal use cases to understand some of our uh own enterprise data that that we generate ourselves. And it's it's been a uh enlightening process this far, so.
[01:13:58] Thanks. Yeah, for us uh it's been more practical use case, actually. Earlier, um all the marketers or any of the campaign people, they used to raise a request, "Hey, analyst, can you get me this particular data set? I want to understand if there is a set of audience I want to target in particular region." And that goes back and forth 10
[01:14:14] different times with all the tweaks and all. But, with the introduction of Genie last year and some of the work that we've been able to do, build some data foundation, and get this in the hands of our uh loyalty campaign managers or marketers, they are now playing with it as if I
[01:14:31] would play with the ChatGPT or any of the things, right? They're asking like questions back and forth, cutting down the time that it takes for them to get the information that they need, and able to make the decisions much faster and more empowered in that case. So, that's
[01:14:46] how we've seen the whole text stack change both for them as well as us. Uh we focus more on getting the data ready and they focus more on making decisions with the availability of information. Cool. Well, I'd love to actually get concrete. So, I'm going to ask each of you to share one example of like an
[01:15:04] agent or an agentic marketing use case in your company. Chris, I kind of feel like you lose out here in the middle, so you should start this one. Let me borrow that again. Okay, so let's see. We started Well, Epsilon's been around for 50-plus years doing
[01:15:20] marketing execution, marketing optimization, identity data, etc. for a long time. And we have a fairly large strategy team that's focused on how best to create various marketing strategies to get to the type of KPIs that you're
[01:15:36] interested in achieving. And so, last year we we spent some time saying, "Hey, you know, we've got a ton of this assets that we've built up over lots of years of doing this for many, many clients. Let's start to analyze that and start to put together
[01:15:51] a strategy agent." We started with focusing on the loyalty program strategy in particular. That's one of our focus areas from a from a platform standpoint. And it's been interesting so far. We launched it earlier this year. We've got a handful of clients that are starting to to use it.
[01:16:07] And it really tries to work with you with a the marketer to get to some kind of creative outcome of a loyalty program, whether it's you know, base rules of the program, various offers that you might want to create, reward recognition, surprise and delight. There's a variety of different areas
[01:16:22] that it's kind of focused on. And so, we're really excited to see, you know, larger and larger use of that over time. Jay?
[01:16:38] Yeah, I don't know how many of you all are from retail background. So, at Circle K we have, like I said, 30 business markets spread across 17,000 stores globally, right? And every 2 months, we have a promo cycle promotion cycle. Which essentially involves all our category managers, which is like a
[01:16:54] particular I don't know, cold dispensed beverage or a coffee category manager. All of them talking to their respective vendors and coming up with different promotions and deals that goes into the whole marketing campaign. What this typically involves is there
[01:17:09] are each business unit, let's say there are 50 of these category managers trying to fill in these bunch of Excel sheets working with macros and pivot tables and all trying to input simple information as what is the promotion I want to introduce, how should that visual look
[01:17:25] like, and what can they do about it? So, this process typically takes multiple weeks from the source of a campaign category manager figuring out what they want to do till a signage, which is a creative that goes out. So, what we initiated a project called
[01:17:41] as creative services automation or internally we call it Prism. It includes a bunch of agents, about six of them that tracks across one, collecting all this information then verifying for the quality of it, then assessing whether it can be
[01:17:57] modified to a certain creative automation service that can go to Adobe products or any of it. And then finally marketing teams. So, this has saved a lot of time in terms of this being multiple weeks to now less than a day or two. And we also have a saying in the company where there
[01:18:13] are some things which are done 30 times across all the 30 business units, but there are some things which is done one time. And this is one of those opportunities that we were able to capture using agents and all. I think I'd probably be remiss if I didn't use marketing genie as our
[01:18:30] example, so we've named her Marge by the way, if you haven't heard about that, but um uh so, Marge is uh you know, when we went on this journey to bring Genie to life, you know, 2 years ago, uh, you know, the thesis that Rick mentioned is like we really think that we will get better usage of our data if
[01:18:46] we allow a better interface for marketers to use the languages they know, the acronyms they're familiar with, and we can actually show the data in a way that they're able to make decisions off of, not that they have to do like the mental translation to what does this mean to marketing? So, I think we we've made a a ton of
[01:19:02] progress on the journey. We're really happy with Marge, but I think um, you know, then this where does it go next? What is what happens next? And I think where, you know, Genie and you saw, it's still this interface that you have to go, you have to type your question, right? It still very requires the marketer to go in, have an idea, have a
[01:19:18] thought, go in. But where Genie's going and what we're excited about is how do we bring the data to the marketer, not the marketer to the data. And I think with that, where AI is really helpful is the role-based. Because while marketing is, you know, a part of every business,
[01:19:33] every sub part of marketing is incredibly unique, and the data that they need to run their business incredibly unique. So, what we're kind of thinking about now is how do we start using Marge to send Slack alerts to people in their business with data that's relevant to them. How do we set up email
[01:19:50] notifications and dashboards that come into their inbox without them asking for it every day? So, I think that's where we're trying like, you know, the agent we've built is Marge, and I think she's really great when a marketer comes to her. And what we're trying to do next is make it even better when we go proactive out to all other marketing stakeholders.
[01:20:07] Great. So, one of the things we keep hearing with all this amazing AI technology is how it accelerates things and it takes out friction. Inevitably though, like when you remove one set of bottlenecks, you find, well, what's the next constraint? I'm curious
[01:20:22] for each of you what you see as the next constraint that you're facing in taking this further, and any thoughts about how you're going to address it? You want to start? Besides all the headcount we don't get.
[01:20:39] Um no, I mean I think it's it's it's really interesting and it's a great question because um another journey we've been on and that's just tangential to marketing is how we're working at Databricks to reimagine uh scaled prospecting, which is how do we do outbounding, how do our SDRs and BDRs work, how does the outbound motion work
[01:20:54] and like with humans and agents together versus just humans alone. Um and on the surface it seems like yeah, no, great use case. Um but you actually realize trying to pull the marketing data with the sales data for an application that's real-time like an agent is incredibly complex and you kind of realize how much
[01:21:11] human middleware and human taxes are BDRs have been playing to fix bad routing rules and bad like account hierarchy ownership that, you know, maybe it was a ticket that got forgotten years ago and was never fixed to this BDR has been doing. So I do think it's
[01:21:27] it's like as soon as these agents move out of one domain and our cross-domain agents, more and more of this kind of human tax will be coming to light and I don't think it's unsolvable, but I do think the time to value is going to be slower than people want it to be because there's just going to be a lot of
[01:21:43] drudgery to get through that that kind of, you know, skeletons in the closet before these agents that work across and between teams will be as powerful as they could be. Fair. Chris? Um so show of hands, any attorneys in the audience here?
[01:22:00] All right, good. I'm not going to offend anybody. I don't I don't see a single one. Uh so yeah, that, lawyers. Um so and that's not to say that, you know, data privacy and security is not important. I don't mean that, right? Um but we subscribe to a human in the loop process for any approval of uh various
[01:22:18] marketing outcomes that an agent might create uh prior to execution. But even with that, it still strikes me as a major limiting factor that when legal gets involved to review whether it's, you know, data use standards, AI standards, etc., it just everything
[01:22:33] grinds to a halt. And there are lots of review sessions, and there's lots of analysis, and people ask a gazillion questions. And by the time you get to the end of it, the market's changed already again. And so, you kind of have to redo it. So, that would be one area that I think we constantly struggle
[01:22:49] with. Unfortunately, I don't think it's going to get a whole lot easier with the increase uh continuously increasing expectations around customer privacy around the world. Um maybe we need a lawyer bot. I don't know. Uh we'll see. Someone from Harding in the room? Yeah, could be an idea. Yeah.
[01:23:05] Yeah. As a user, um what we have seen is at this point, everyone talks about AI, not just externally, but also within the company. Even our store operators at this point are like happy to use them, you know, anything on the screen that's
[01:23:21] the Gemini or whatsoever to ask a question. Our executives are playing around with So, everyone is adopted into this whole AI culture. But the real challenge that comes down to us is the governance around it. So, what's the biggest constraint that is uh right now for us to take this even
[01:23:37] further? Is we as an organization trying to come up with the right set of governance rules. And it's exciting to see the Unity AI gateway. It kind of helps us lean on to set up this kind of foundation to enable or empower all our employees across our
[01:23:52] functions, right? Not just marketing or finance or even stores, for that sake. To take it to the next level. Right. All right. Well, this is uh a very timely one. Although, uh Liz, I'm going to exclude you from this one. Uh I just want to get uh from each of you like kind of your hot take on customer
[01:24:08] like. What do you think that means? I can go first. Um I think Kumar was mentioning this earlier where traditionally all CDPs were the platforms that uh all the companies were trying to get into, including ours, to consolidate all the customer data in one place, right? Whether it's all the
[01:24:25] profiles or the real time and everything. The real value that agent CDP or the customer lake that we see drives is CDP is not just a platform anymore just to store your data, but also action on it and execute on it and leverage it across different functions
[01:24:41] in your organization to understand your customer across cross platform, cross channel, as well as drive your initiatives and it also end up even even evaluate the performances, too. So that's one of the exciting things that Dan and the others were showing off. So that's what I think is the hot ticket
[01:24:57] of it. So we've been a big partner of Databricks since 2018-19, so for for quite a while. In fact, Liz even pointed out that I'm wearing Databricks shoelaces at the moment. Yeah, yeah, yeah, they were nice. Some swag.
[01:25:18] But you know, I I would say that we're definitely interested in improving the value that our clients can use with the data that they have. So to the extent that customer lake improves a lot of the data wrangling capability, data access, single source of data, reduced data copy, all of those things that
[01:25:35] we've been hearing about yesterday and today. We're we're tremendously on board and we're very excited to partner around the extension for activation, our data expansion and and identity expansion from there. Terrific. I kind of assumed you were Yeah, team
[01:25:51] customer lake. Okay. Last chance here. Any closing advice to the folks in the audience here who are What was the title for our session from you know, MarTech stack to agentic marketing? How to think about getting going on that journey?
[01:26:14] I would say do the non-fun stuff first. There are so many shiny objects and so much potential with AI, but you know, as Chris mentioned only ontology is so important. Ontology is built on taxonomy. It's like the non-sexy parts of marketing that are going to be running the future of context. And if you just kind of skip go and collect
[01:26:30] $200, these things can come back and kind of bite you later on and you don't want them to bite you when you're so far down the path that you lose trust from a lot of your stakeholders or your customers. So, I think my my recommendation is a lot of these things where we we all I think we're all
[01:26:45] aligned that context is going to be the next most important thing that marketers can own and manage in their company. The value we'll provide in the future that context is built on a strong, you know, data foundation. So, don't ignore the non-sexy parts of marketing cuz I think that will be what propels, you
[01:27:02] know, the good from the great. And that's definitely more sexy than having to like type in business cards and yeah, so it's gone better. that much more. Chris? Yeah, so I think it's really easy to come to a conference like this, hear about all this cool stuff that is being
[01:27:17] built, and feel a little bit of anxious of hey, where am I in that journey? We're behind. We want to be, you know, moving faster and all that. Um but I think what's interesting, you know, we've never seen this uh speed and acceleration of speed in in
[01:27:33] change that as we have in the last few years. Um and you could look at that and say, hey, that's kind of scary what's going on. I think it's also a bit of an equalizer. So, if you are maybe a little behind where you are, it kind of, you know, uh aligns everyone or pulls everyone
[01:27:48] down to a more consistent playing field. And so, I think that the biggest things that I would focus on is, you know, be willing to be flexible, adaptable, and be curious uh for your consumers. All right, Jay, you get the closing word.
[01:28:04] Yep. Um a lot of pressure, but actually, the one thing that earlier Horace you mentioned that we don't talk about often is the culture part of it. I know we all as individuals, technologists, or marketers, we are passionate about what we do and
[01:28:19] everything and we want to make the best in our roles and all. My take is having change management in organizations is hard and you need to find people in your partner teams, whether you're marketer, find your data
[01:28:34] scientist, or a data engineer, or technologist who's as passionate about you to make this change. As Chris mentioned, the technology is changing at a, you know, sometimes, you know, week or even day basis. So, they they your partners also have to be equally
[01:28:50] passionate enough to explore, ideate, experiment, trial and error, and then go make those changes. So, yeah, I would say that that would be a biggest thing to make any impact in your organization broadly overall. Perfect. Thank you. Can we make that room jealous
[01:29:07] again and give a great round of applause to our Agility Marketing panel here.

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.