Introducing: CustomerLake - Campaign Agents

Summary

CustomerLake is presented as the world's first agentic CDP, built natively on the Databricks Data and AI platform, enabling marketers to run campaign planning, audience segmentation, activation, and performance reporting using AI agents.
Campaign agents use Genie-powered natural language segmentation to build audiences from lakehouse data and activate them to channels such as Meta, Braze, and The Trade Desk via reverse ETL.
Infinity campaigns provide an always-on agentic loop that continuously refines targeting and messaging based on performance signals—opens, clicks, and conversions—fed back into the lakehouse.

Introducing: CustomerLake - Campaign Agents

Welcome to the next generation of marketing with the world’s first Agentic CDP – Databricks CustomerLake. this video introduces how marketers – and the data teams that support them – can run agentic campaign planning, insights and execution on Databricks and where AI agents use lakehouse data and models to decide what content to send, when to send it, and to which audience. Join us to see how agentic workflows close the loop between customer data, marketing decisions, and execution. In this video, you’ll learn: - How to build audiences for marketing campaigns using data in Lakehouse - How to connect to marketing execution channels via reverse ETL - How to map personalization variables and trigger one-off and micro-batch sends - How performance signals like opens, clicks, and conversions can flow back into lakehouse
Talk By: Cindy Jiang, Product Manager, Databricks ; Josh Braun, Staff Product Designer, Databricks ;

Chapters

FAQs

What is CustomerLake on Databricks?

CustomerLake is an agentic customer data platform (CDP) built natively on the Databricks Data and AI platform, announced at Data and AI Summit. It enables marketing teams to run audience segmentation, campaign planning, activation, and reporting using AI agents without moving data outside the lakehouse.

How does CustomerLake handle audience segmentation?

Campaign agents use Genie's natural language interface to let marketers describe their target audience in plain English, which is then translated into queries against the Unity Catalog customer data model. Audiences can be activated directly to platforms like Meta, Braze, and The Trade Desk via reverse ETL.

What are Infinity campaigns in CustomerLake?

Infinity campaigns are always-on agentic loops that continuously plan, execute, and optimize campaigns based on live performance signals. Engagement data such as opens, clicks, and conversions flows back into the lakehouse, enabling the agent to refine audience targeting and messaging over time.

Why do traditional CDP architectures fall short for AI use cases?

Existing CDP architectures fragment customer data across multiple vendor systems, requiring multiple data copies for different marketing workflows. This makes it difficult to deploy AI agents that need a unified, governed view of the customer to make accurate, real-time decisions.

Full transcript

[00:09] Hi everyone, welcome to today's session. You might have just heard a sneak peek from Ali in the keynote that data bricks is actually launching an agentic and native CDP. Well, you're at the right session to find out more about it. In today's session, we'll give you a glimpse of our road map and the product
[00:22] ahead of us, the keynote tomorrow to show you what what exactly we're building for Agentic CDP. Before we get started, just a little bit about myself. I'm Cindy. I'm a product
[00:34] manager at Data Bricks working on how to democratize data and AI for a marketing persona and bring data intelligence to your customer data. And I have Josh here with me today. He's the product designer on customer lake, one of the brains
[00:47] behind all the awesome tools we're building for marketers. We all know that the best companies really differentiate themselves from delivering the best customer experience by really understanding who their
[00:59] customers are and designing a personalized experience for them. However, we're still hearing so many feedback that delivering great customer experience is still very difficult. For most organizations we work with, their
[01:12] existing data architecture actually get into the way of achieving their marketing goals. Let's look at a typical enterprise architecture to see exactly why this is still failing for them. Your data might
[01:24] be fragmented and have multiple copies sitting in different vendors. You might have the source of trion data lakeink but you will have to export multiple data sets to downstream systems for your marketing team.
[01:37] Moreover, we're also heard feedback that these customer data are usually locked behind technical screens. Usually the data team has to come in support any analytics on your customer data and as a result um time to insight can often be
[01:50] in days or even weeks missing opportunities to engage your customers at the best timing. Last but not least, all these problems um really compound in the age of agents
[02:02] and AI. Despite the global hype around AI, achieving this transformative use cases is still incredibly difficult. As of today, you always have to think about, you know, where to deploy your AI and agents. Do you want to have your
[02:14] agent closer to your application and end use cases such that the agent can be immediately useful or do you want to have your agent sitting right on top of your data so your agent can get all the
[02:26] context and be power and to build a powerful agent? As of today, we have not actually seen a good architecture to maximize the value of AI and agent. Now over the years marketers have tried
[02:39] to use technology to solve some of these problems. First there was the era of email marketing where tools could pull list of customers for targeting. This was useful but it it was really the technology from the 90s. Then we had the
[02:52] package CDP. These are purposeful build tools to unify and activate data. But CDPs were really invented before we had the data warehouse. So getting the data into a CDP means that you have to make
[03:05] multiple copies. And again this was a governance nightmare for your data team. Now we have the composable CDPs which are supposed to be kept in the lakehouse with the data and the marker friendly
[03:17] interface on top of it. While you're still exposing your customer data and this was really built before the agentic error. So it's honestly limiting when you think about deploying AI and agents on top of your customer data. So none of
[03:31] these solutions have been built for agents from the ground up and none of them have been really close enough to the data to make agents really work at scale. So to solve all the enterprise challenges we just outlined, I'm super
[03:45] excited to announce that data bricks has launched customer lake. Customer link is the first agentic and embedded CDP that is sitting directly on top of your data lake. This is where customer 360 in the lakehouse actually
[03:59] meet agent intelligence. Customer link is is also the only CDP embedded in the data AI platform. For the first time, you don't have to compromise on data governance for your most important data. Security and trust
[04:13] are built in within the Unity catalog with no data movement or no data copy. Customer link is also democratized and built for marketers trusted by your data and IT team. With our products being
[04:26] having marketer friendly campaign interfaces, marketers can actually self-service and get access to this enterprise data to personalize campaigns. Last but not least, customer lake is
[04:38] built for marketers and agents to work together. This is where our agent can come in, understand all your customer data and personalize one-to-one interactions at scale for your customer data.
[04:51] So with being embedded, democratized, and autonomous, marketing teams no longer have to operate from siloed data or static customer strategies. For the first time, we have a customer data, marketing application, and marketing
[05:04] agent actually side by side and next to each other in a unified platform. We believe that this is actually the architecture for your marketing agents. We heard so much feedback that marketers want to run campaigns with onetoone
[05:17] personalization and we believe that this is finally the architecture that can empower marketing teams to do so. So with that vision in mind, I'm super excited to announce the two products we're launching in Kushner Lake. The
[05:31] first one is profile agents where we're making ETL simple and agentic for data teams to transform from the raw bronze table to the gold tables to prepare the customer 360 as easy and as quickly as
[05:44] possible. And as part of the profile agents, we're also launching a net new identity resolution capability. Next, we have campaign agents. This is where we are powering the marketers to
[05:56] understand their customers, plan and execute campaigns with agent. Today we'll focus on the campaign agents and I'll walk through our upcoming road map while Josh will give you a glimpse of the product in his demo.
[06:10] With campaign agent, our goal is to help data bricks customers better understand their customer data. I'm excited to announce that the product is already in private preview and as part of the private preview we have a NoSQL business
[06:22] userfriendly interface to help marketers create segments on top of their customer data. It also supports using natural language to build complex audiences.
[06:34] And with every audience you create, we have an out of the box inside panel that helps understand your customers immediately. For example, we'll help you break down your customers against existing values, compare your audience
[06:46] against other audiences, and last but not least, would actually let you tell us what's your most important metrics for your customers and marketing team, such as lifetime value, propensity score, and we'll display them per
[06:59] audience out of the box. And once you have created your audiences, we'll then help you activate your audiences and insights. I'm excited to announce that Data Bricks is building managed reverse ETL connectors to a
[07:13] variety of CRM, campaign automation, customer engagement, email services, and advertising platforms. The connectors you see on the slide are the ones on the short-term road map,
[07:25] while we plan to expand to many dozens by the end of the year. Moreover, for all the channels we're activating to, we'll also have delta sharing or ingestion connectors to help bring the measurement data back such that all the
[07:38] campaigns we run on data bricks will have closed loop measurements. And I'm also excited to share that upcoming capabilities on the road map beyond the product preview such as a
[07:50] profile API which is essentially a cache of our c customer data served through our lakebased offering. This is essentially a real-time servant layer for customer attributes and custom audiences to support use cases like
[08:04] inapp or web personalization, call center optimization or trigger messages in real time. Last but not least, I know we're here today to actually hear about how to run marketing campaigns on data bricks. So, let's dive in. So, before I
[08:18] get into the details, let's first see how marketing campaigns are actually run today. Usually campaigns are built through a waterfall process. You know, marketing does the strategy, data engineering puts customer 360 together, marketing ops
[08:32] have to build the audiences. Now, creative team comes in to build the content. And finally, marketers actually activate a campaign and analyze its performance over time. Well, but with the age of agent and AI,
[08:46] we've heard constant feedback that this whole campaign process is broke broken. It's complex, rigid, and slow. it takes weeks or even months to launch or just adjust a campaign and these campaigns actually rarely change let alone like
[08:59] adapting in real time. So really even though industry has been talking about onetoone personization for more than a decade at this point the reality is that we had not had the technology to do anything other than sending the same
[09:12] content to millions of buyers all at the same time. Well, but what if we could personalize every interaction for every customer in real time continuously? To
[09:25] do that, we really need to think differently and invent a new type of campaign. One that doesn't have a start and doesn't have an end. One that can personalize audiences of one versus audiences of millions. One that can
[09:39] respond as fast as the buyer's agent take actions. And the only way to do that is to replace the waterfall with a continuous agentic loop with agents that work onetoone with
[09:52] every customer that's always on and respond in real time. Let's look at an example. Imagine we had agents embedded in the lakehouse and what those agents are doing is to actually process the latest signal from every single customer
[10:06] and then the agent is evaluating what is the next best action and the ideal personalized content for each individual customer. This superhuman personalization then
[10:18] actually reach back to the same audience that interacted with previous content, take actions again and from there the loop would continue in this infinity-like shape. Evaluate how your customers reacted to the outreach you
[10:31] know bring back the signal which feed into the continuous agent loop with additional signal and context. Now this would achieve actually for the first time true one-to-one personalization with our marketing agent and this is now
[10:45] within reach. We thought that this is the new approach to marketing campaigns and we are calling it the infinity campaign that you'll hear more about it in the keynote tomorrow. Last but not least, we have contextaware
[10:58] genie embedded in our CDP. This is where you can truly ask any questions about your customers and campaigns. for example, looking for your highest lifetime value segment or looking for
[11:10] customers that might have high churn risk in the next quarter. Now, with that in mind, let me pass it to Josh to give us a demo and a glimpse of the actual product.
[11:26] Good morning everybody. I'm Josh. I'm a product designer here at data bricks working on customer lake and I'm really excited to get to share an early look at what we've been building with campaign agents. Um campaign agents is a separate app or
[11:40] experience inside of kind of the greater data bricks ecosystem and it's built just for marketers and the data teams that support them. With campaign agents we can learn more about our customers.
[11:52] Um, we can also activate those customers in marketing engagement or uh advertising platforms downstream. And like Cindy mentioned, we can even run full agentic campaigns that we call um
[12:04] infinity campaigns. For the sake of our demo, let's imagine we work at an airline called DBX Airways. Um, all of the kind of use cases you see are going to be grounded in this airline travel
[12:17] data set throughout the demo. So, with that out of the way, let's uh take a look at um the product. So, we're going to start out on the campaign agents homepage. And the homepage is sort of
[12:29] this like command center for marketing teams. Um like I mentioned, it's a separate experience in inside of data bicks and you can access it right from the app switcher up in the top right from any of our other data bricks products. Right from the homepage, we
[12:43] can ask Genie questions about our customer data. So, we might ask something like, you know, what's uh what's contributing to this increase in marketing email optouts this week? And Genie can help us kind of uncover the answers there. As we scroll down, we can
[12:58] um observe some of our live campaigns. We've got quick access to some of our most important audience segments. In this case, I saved an audience called business travelers. I'm keeping track of our marketing optouts and uh a couple
[13:12] other audience segments as well. And then down below, I can see some of my most active syncs as well as some of my fastest growing audiences. If we click into one of our live campaigns, we can learn a little bit
[13:24] more about it. And I'm going to click into this post sale upsell campaign. And in this campaign, we're trying to increase premium revenue by offering some upgrade upgrade paths from people
[13:36] who've recently purchased a ticket. Now, campaigns in uh customer lake are a little bit different than other CDPs or engagement platforms, and that's because they're orchestrated by agents. What I
[13:49] mean by that is that each of the profiles or customers inside of our target audience for this campaign are going to be evaluated individually by an agent to help determine the right channel, the right message, and the
[14:01] right timing to activate them. we can sort of see some of the insights of those activations right from the homepage of our campaign. So, we've got the number of evaluated profiles, the number of activations over time, and
[14:14] even the number of conversions. And as we scroll down, we can see this cool heat map view that's going to show me the number of activations per day since we've launched this campaign. We can even see some interesting insights here,
[14:26] like the most active day of the week, uh, which looks like it's Thursday. Um, and we can see the most active month as well. So, it's back in February when we launched this campaign. Another really cool thing about Infinity
[14:39] campaigns is that we can run uh cross channel activations. So, here we we're activating across three different channels for this campaign. We're starting out with an email blast in uh from Braze and here we're sending out
[14:52] those offers via uh via braise and we're pulling in some of that performance uh data from Braze so that I can observe right from the campaign page itself. So here I can see the number of emails
[15:04] sent, the open, clicks, conversions, typical things you would expect to see in uh a marketing engagement platform. But we can also observe some of the metrics coming from our uh from our meta
[15:16] audience sync as well as another sync. We're pushing profiles to the trade desk to do some additional retargeting across the web. One really important thing to keep in
[15:28] mind when agents are orchestrating a campaign is that we want to be able to set up guardrails for that agent. And we can view some of the guardrails right from the configuration tab on the campaign. So I'm going to scroll down to the guardrail section. And here we can
[15:40] see some of these guardrails that have been set up globally across all of my campaigns. Here we're doing some audience suppression. So, we're telling the agent that it should never reach out to anyone who belongs to the marketing
[15:53] optouts audience, the anyone who's had their latest trip disrupted or who has an open support ticket. And this allows us to sort of defensively uh prevent the agent from activating certain groups of
[16:05] people we don't want it to. Uh we also can set up things like frequency caps or quiet hours to make sure that the agent is only uh triggering activations when we want it to.
[16:17] Um, another like interesting view of um, the observability of our campaign is that we might want to like take a look under the hood and see um, a really detailed view of how the agent is making its decisions. And we can do that on the
[16:30] agents tab. So, we'll click into one of the recent uh, batches of our agent evaluations. And we can actually see um, how the agent is um, evaluating the profiles in this campaign. So for this
[16:43] batch of profiles, we can see that it activated uh 22,000 people. Um and we can see the breakdown of those activations across the three channels here. It's primarily activating in braise. And that makes sense because the
[16:56] email is our our initial trigger for our campaign. And we can also see the breakdown of campaign overlap across all the profiles in this campaign. This is really important because we want to make sure that the we're not over um
[17:09] activating profiles across multiple live campaigns. And we can see that you know the majority of our profiles here have a low overlap meaning they don't belong to any other campaigns. And then there's a few uh a smaller group I guess that has
[17:24] a high campaign overlap. We can actually dig one level deeper and view a sampling of the profiles that the agent evaluated as well. Um here we can see that we some of these
[17:37] profiles have a low overlap, a medium overlap or high. And we can actually see the activation that the agent uh decided to to trigger. So for this profile here, uh we have a high campaign overlap
[17:49] because they belong to three other live campaigns and the agent deferred any action during this evaluation. And I can actually click in to this reasoning tab to see a summary of the agent's
[18:02] reasoning to make this decision here. The agent actually deferred any action on this profile because it prioritized another live campaign that this profile belongs to. We think this is like a really powerful way to help orchestrate,
[18:15] you know, multiple campaigns at the same time and prevent them from kind of colliding with each other. We can also see um kind of inspect when the agent does actually activate a
[18:27] profile and we can see how it's going to personalize aspects of of this activation as well. So here we have Miles who we activated in Braze and I can click preview to sort of see the the
[18:39] personalization email that he's going to receive. Miles has an upcoming trip to Tokyo and so we've personalized this uh for him. And then we've we're recommending two upgrade paths for him and the agent is choosing two pieces of
[18:52] content that are the optimal upgrade for him. So here we're suggesting a checked bag upgrade and a uh pre-onboarding access pass. And this is unique across all of the profiles. So if I click into
[19:05] another one here, we can see that we've used what we know about this profile to customize this message. Marcus has an upcoming trip to LA and we're recommending two different pieces of content.
[19:17] So, that's kind of a quick look at infinity campaigns, but I want to take a step back next and show how we can sort of build our way up to one of these infinity campaigns from the ground up.
[19:29] And earlier I mentioned that campaign agents is a separate app, but it's important to note that it's still built on top of the lakehouse. So when we set up campaign agents, we're going to bring in data sets from Unity catalog to build
[19:42] what we call a customer data model and we can view that data model right from the admin tab from the main nav. So here I can see some of the data sets we've pulled in to help build this data model.
[19:55] Um we have a customer profiles table which um if I click into this I can see some of the columns on this table. It's things you would probably expect to see in um a customer 360. uh customer name, email, phone number, and some other
[20:08] traits about the customer. Um, but we're also pulling in some additional data sets to make our data model more rich. So, we have things like app sessions, booking transactions, ancillary purchases, customer support tickets, and
[20:22] even flight activity for all of the profiles um in our data set. We can use this data model to build what we call customer attributes. And if I navigate up to the attributes tab up on
[20:34] the left, we can see some of these attributes. Attributes are basically just facts about our customers that um we can use to sort of build audience segments later. It might be something like total spend for 2026, latest
[20:48] booking date or the loyalty churn risk score. And if we click into one, we can learn a bit about it. So for this total spend for 2026, it's defined as uh the total airfare and ancillary spend from
[21:02] confirmed or completed bookings this year. I can see some highle data quality metrics about this. So I can see the number of missing values and the uh min and max value for this attribute. And
[21:15] this helps me know if I can actually trust the the data in this attribute. Um, similarly, I can see this really cool distribution across this, which helps me know if I can trust it or not, too. Here, it looks like most of our
[21:28] customers have spent zero dollars this year. And then that number kind of trails off as we get higher. Attributes can be defined in SQL or they can be brought in as any of the columns on our customer data model can be
[21:41] converted to an attribute. Um, we can even build attributes in the lakehouse in a notebook or a machine learning model and bring them in that way as well. The important thing about attributes though is that they act as this sort of
[21:53] building block for personalization and audience segmentation later and the data team can centrally build and govern them and then marketing teams and agents can use those as building blocks throughout
[22:06] the rest of the platform. Next, I want to show you how we can use them to build audiences. So, if we navigate to the audiences tab, uh we can see some of the pre-built audiences that um we've already built.
[22:18] Um and audiences are basically just groups of profiles that have been grouped together by um customer behavior or different traits. And they might be something really broad like active
[22:30] business travelers. Um but they could also be really narrow and focused like high value Lake Tahoe summer weekender prospects. If I click into an audience, I can learn a little bit more about it. And here, our active business travelers
[22:43] audience. We can see the total size for this audience and how it's changed over time. We can also see some interesting insights about this audience like the average flight frequency, the uh bag uh
[22:56] the bag add-on propensity, ancillary spend over time, and we can even see at the bottom how this audience compares to other audiences. We can see how this audience is defined on the definition tab. And here our
[23:10] audience is fairly simple. It's just a couple attributes, but we'll um we'll see that this is defined as the primary travel purpose equaling business. And then we're qualifying it with days since
[23:23] last flight is less than 180 days. Um pretty simple, but we're also doing some audience suppression here as well. So, we're saying we don't want to ever include any profiles that belong to the marketing optouts audience or anyone
[23:36] who's had their latest flight disrupted. One of the coolest things in my opinion about um customer lake is that it's really easy to build and experiment with new audiences um without having to write
[23:49] SQL or having to submit like a support ticket to the data team. Marketers can jump in and build audiences in a point-and-click interface or uh use natural language to build them themselves. And it's a great way to kind
[24:02] of experiment in a self-s served way. And I want to show that next. So, we navigate back to the audiences tab. We can build a new audience by clicking new audience up in the top right.
[24:15] Let's imagine that, you know, our airline has recently launched a new route from San Francisco to Lisbon, and we're trying to capture the summer vacation market to uh to Europe. There's one challenge, though, and that's that
[24:27] the the new route is somewhat underperforming what our expectations are, and we want to launch a marketing campaign to drum up some new interest in this new route. But, um, we want to make sure that we're talking to the right
[24:39] people, right? Um, we don't want to just blast this out to, you know, a huge audience. It's a great use case to build something a little bit more targeted. And we can start by just clicking this add filter button on the audience builder. And we're presented with a menu
[24:52] of all of those attributes that we saw earlier. So we can start to compose these attributes together to build the audience we want. We can also build with other audiences. So maybe we want to start by targeting um a broad group of
[25:06] vacationers um as like a starting point. But there's actually an even easier way to do this and that we can just prompt Genie to assemble those attributes together for us. So maybe we're going to um type something like include customers
[25:20] who might be interested in uh direct routes from SFO whoops to Lisbon for
[25:33] for summer vacation travel. Um we also want to exclude some people though. Let's exclude anyone who um has uh booked a similar flight recently.
[25:50] Um and we'll have Genie try to build this audience for us. Great. So, Genie is going to um let us preview the audience that it generated. It wrote a quick summary and we can even see the size of this audience and we can
[26:03] see how it interpreted my prompt with the attributes that uh it has available to it. So it's um it's using this home airport attribute and it's set to San Francisco, but it also actually expanded our reach a little bit to include home
[26:16] airports in San Jose and Oakland. Um, it's also looking at um some of our web search activity to find people who have recently searched for trips to Lisbon or um related cities. Um, and then it's
[26:30] looking for a vacation signal. So, we have two attributes that it's looking at a route affinity to Europe or a summer of travel intent being true. Lastly, it's also doing some audience
[26:42] suppression down at the bottom. So, it's looking for booked destination in the last 90 days. And it's looking for anyone who booked to Lisbon or some nearby airports in Europe. This looks
[26:54] pretty good. So, I'm going to actually accept Genie's uh proposal. And I can see um if I wanted to, I could start to modify this audience that Genie gave me. I could add additional web uh web search
[27:06] intents or expand or or shrink it. Um, but Genie also actually generated some insights for me right in the sidebar here. So, I can see how this new audience breaks down by home airport. I
[27:19] can see how it breaks down by our loyalty tier, next trip window, and even how it compares to some maybe similar audiences. In this case, we're comparing it to our vacationers audience, and there's quite a bit of overlap here, but
[27:32] I can see that um our new audience is much more targeted, and that's exactly what I want to see. Um, this audience looks pretty good, but there's one thing I noticed going through here, and that's um, we we have about a 100,000 people in
[27:46] this kind of ultra premium diamond tier, and we want to talk to them a little bit differently than we do the rest of our customers. So, I'm going to have Jeie exclude this segment from our new audience. And Genie is going to go ahead
[27:58] and rebuild and show me a preview um, of the before and after. So I can see how the audience size has changed and how that definition of the audience has changed uh from the previous state to
[28:10] the new one. Here it added a new loyalty um new loyalty tier suppression uh attribute. I'm going to accept these changes and I can see right away in the
[28:22] insights panel that now our diamond tier has uh is at zero. Genie also automatically gave my audience a nice name and a description. And I'm pretty happy with this. So, I'm going to go ahead and save it.
[28:36] Now, my new audience is um ready to start activating. I can use this audience um or I can activate it in two different ways inside of uh customer lake. I can build a sync, which is this
[28:48] kind of classic reverse ETL use case. I can push all the profiles from this audience to some downstream tool. Um and then I can also set up an infinity campaign. And I'm going to show you both of them, but I'm going to start with syncs. So, if we head over to the syncs
[29:02] page, I can see all of the active syncs that uh we've already set up. And if I click new sync, I can, you know, quickly configure my new sync here. So, um I first need to select an audience. And
[29:14] I'm going to select the one we just built. And then I need to select a destination. And um in our private preview phase, we support a couple dozen of, you know, the most popular, uh destinations. And we expect this number
[29:27] to grow pretty rapidly as we um continue to build this out over the next uh few months. Um I can kind of browse some of my active uh or set up destinations. And in this case, I'm going to quickly uh set up a sync to meta as a meta custom
[29:40] audience. I can run syncs manually or set them up on a recurring schedule. I'm going to leave it set to manual for now. And then each destination has its own set of configuration that we'll need to
[29:52] set up. And I can have Genie help me with that, too. Um, I'm going to leave everything else set to the default and click create sync. And just like that, we've set up a new sync from um the audience that we just
[30:06] built to meta as a custom audience. Um, if this was set up on a schedule, it'll run automatically or I can click run now and get a quick preview of this um of the sync. Here we've got the estimated
[30:19] rows of 150 or 850 uh,000 profiles that we're going to push to Meta for retargeting. I'll click launch run and it's going to kick off a run. This is going to take a little while, but once
[30:32] it's complete, I can actually start re-engaging with these customers in uh Meta as a custom audience. Um the second way of activating uh our audience is with an infinity campaign
[30:44] and it's the last feature that I want to show you. So if we head over to campaigns, we saw earlier like in the beginning of this demo how we can observe and use the agent to um the agent can help us activate a campaign,
[30:57] but the agents can also help us plan our campaigns as well. So if I head over to the new campaign button, um let's try to finish out that task that we started out um earlier. we have this underperforming
[31:09] route that we want to try to drum up some new interest for and uh hopefully sell more tickets for. So, let's say help me um increase bookings on um my
[31:24] new route whoops from SFO to Lisbon this summer. Um and I can attach additional context to this prompt as well. So, I'm going to add an
[31:36] audience, which we'll use the audience that we just built. Um, and I can also add things like marketing briefs, dashboards. I'm going to add a brief that my team and I worked on ahead of time. We'll click start planning, and
[31:51] Genie is going to take um the prompt plus everything it knows about our customers and my goals, and it's going to build out um it's going to attempt to build out a campaign plan for us. And if we left some things out, it's going to
[32:03] interrupt us and ask us for a little bit more context. So here it's asking some follow-ups about which channel we want to use. Um, in this case, I want to use braise and meta ads. Um, it's asking us
[32:16] about the number of messages we want to activate in this this campaign. I'm going to do a launch message plus some social retargeting. And it's asking us about the timing of our campaign. So we're going to do a 90-day campaign. And now I'm going to
[32:29] have it draft up a brief for me. So Genie can pull more context out of me as we're building out this campaign. Ultimately, it's going to create what we call a campaign plan, and it allows us
[32:41] to review Genie's understanding of our goal and our customer um customer data. So we can see that it defined a goal for us, a conversion target. It understands the channels we want to use and deeply
[32:55] understands our audience segments, the messaging plan. Um, and it even wrote out this this sort of two-step touchpoint plan for us with some sequencing rules. So, here we're going to send a launch email to eligible
[33:07] customers um for this specific route. And then after launch, we're going to sync only the converted uh the non-converters to meta ads for some social retargeting. It understands our guardrails and timing. And it this
[33:20] allows us to really kind of um ensure that Genie understands the intent of our campaign. I'm going to click build campaign. And now the agent is going to take this kind of unstructured plan
[33:32] document and turn it into a real structured campaign. If we wanted to at this step, we can modify some things. We could add additional audiences or or like we have full control here yet. Um we can see
[33:45] that it used the audience that we gave it up front. Um I can always preview the audience definition right from this campaign page. Um, it also set up and built a brand new audience on the fly to measure our conversions. So, it's it
[33:58] created a new audience of San Francisco to Lisbon converters. And again, I can preview that audience right from this this view. It set an end date for our campaign and even set up two syncs, one to Braze and
[34:11] one to Meta to um do our activations. Another really cool thing I want to call out is this campaign operating procedure. So the agent is going to kind of write out a set of rules to help
[34:25] determine um when to activate people and what message or which channel to activate them across when it does its evaluation loop later. The thing worth calling out here I think is that um
[34:37] after we trigger our first launch email, we want to sync only the non-converters to meta ads for social retargeting. And this is basically just a rule that the agent is going to use to um help when it evaluates profiles.
[34:50] Finally, we can see some of the kind of globally set guard rails for this campaign. So, the agent knows never to reach out to these audiences or um it will enforce these frequency limits right on this campaign.
[35:03] Um this looks pretty good. I'm not going to make any edits. I'm going to click save draft. And we can see that our new campaign is sort of staged and ready to go. I'm going to click launch campaign.
[35:15] And once the agent starts actually evaluating profiles for this campaign, we'll see some of that performance uh metrics uh flow into this and we'll be able to kind of review the campaign as it goes just like we saw in the
[35:27] beginning of the the demo. So that is my whirlwind tour of um the new campaign agents. Um it's in private preview today. Um, and we think that by
[35:39] kind of embedding a CDP inside of data bricks, we're unlocking this new era of marketing tools where marketers and data teams and agents can all collaborate together. And it really unlocks this future of onetoone personalization at a
[35:53] massive scale. So, I'm going to turn it back over to Cindy who's going to talk a little bit more about the private preview and close us out.
[36:07] >> Thank you, Josh. Um the era of agents powering true one-to-one precise experiences is finally here. Um as we have been developing customer lake, we have been honored to collaborate with an elite set of innovators. These are data and marketing teams that are pushing the
[36:20] boundaries of customer experiences by combining agents, customer data and worldclass marketing strategies. For example, we hear from Genet by Sentender that strong merchant
[36:32] relationship are at the core of how they grow. As a global payment business operating across multiple markets, channels, all the customer needs, and the CRM is actually super strategic to
[36:44] their growth. With data bricks customer lake, we're excited to help get understand their customers better, engage with their customers more meaningfully, and create more consistent experiences at scale.
[36:56] And similar as our customers, we're lucky to have an elite ecosystem of partners that jointly serve our customers in terms of data, content, advertising, activation, and services. Thank you. And I know that this is only
[37:10] the beginning of the summit and you'll hear more about the official announcement at the keynote tomorrow, uh, which we're all super excited about. But our campaign agent, as Josh mentioned, is in private preview and please reach out to us, our account
[37:22] team, as we're constantly looking for feedback. Well, I know it's at the beginning of the summit, but if you only take one thing away with you for the rest of the summit is that Kush customer lake is embedded, democratized, and autonomous. Thank you.

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.