Salesforce, Inc. (CRM) Earnings Call Transcript & Summary

Welcome to today's session: How Data Cloud Powers Generative AI with Your Enterprise Data, and thank you all so much for joining us today. Before we begin, I'd like to cover a few quick notes with you about our webinar platform. Today's webinar will be available on demand after we wrap up, and it will be accessible through the URL you're on now. It will also be sent to you via e-mail tomorrow. Please note the slides will advance automatically throughout the presentation. To enlarge the slides or any other widget, simply drag the bottom right corner to resize. Should you need technical assistance, click on the help widget located on the bottom left corner of your console. We've also added some additional resources, which are available through the resource library to the right of the slides. There, you can find some additional related content. We encourage you to submit your questions at any time throughout our presentation today using the submit a question widget. We'll answer as many questions as we can at the end of the presentation. And lastly, let us know what you think of today's presentation by sharing your feedback via our webinar survey. You're also welcome to share your excitement in the moment by using the emoji reactions on your screen. And with that, I am turning things over to Ryan to get us started.

Awesome. Ariana, thank you so much, and thank you all for joining us this morning, afternoon or evening, wherever you are. As Ariana mentioned, my name is Ryan Schellack. I'm a Director of Product Marketing here at Salesforce. I have a great crew with me here today, and we're going to talk about, as she mentioned, how we're enabling you to take Data Cloud, this foundation of unified data within your business and use it to power generative AI. We get this question all the time. How can I take my data, put it all together, even if it resides in another lake or a warehouse, and put it to work not only towards things like automation and analytics but also that critical [ aspect ] well, getting generative AI grounded in your business data. So let's jump into it. I've got some fun stuff here from my friend, Minhaj, that many of you probably have never seen before unless you were with us at World Tour in New York a couple of weeks ago. So I got to throw a forward-looking statement over here. Please make your purchasing decisions based on products that are generally available. But we've got some good stuff coming up soon, so if you make a decision on that, I won't get mad at you. Now I have a great crew, as I mentioned. You've seen me, so I'll just get it out of the way. But I want to introduce Stella, one of our incredible experts from Data Cloud you'll hear from in a moment. Stella, you want to say hi?

Yes. Hi, everyone. So I'm working for Data Cloud product engineering team. So I'm working with a lot of our customers to really make sure they understand the values and they will be successful and also getting their direct feedback into our product team, so making our product even better. Minhaj?

Minhaj Khan, I'm a product manager for Prompt Builder, and we're building the Einstein platform and working in partnership with Data Cloud to bring all of the cool stuff to you that we're going to show you in a few minutes. Yes, I've been at the company for a few years, and I was a software engineer previously as well, and we're really excited to show you all the cool things we're working on. Back to you, Ryan.

Awesome. So let's jump right into it. I only have one kind of a marketing problem slide here, so we're in luck. Most of this session is just pure technical content. But I do want to acknowledge one thing. Part of why we're talking today is that everyone here is being tapped on their shoulder, many of you are IT leaders, practitioners, developers and admins, and people are saying to you, "How can we deliver really rich AI-powered experiences to our customers?" And many of you already know in your bones, that's not really an AI discussion. It's, first and foremost, a data discussion. You can't really have a great AI strategy until you have a perfect prerequisite data strategy. It doesn't have to be too perfect but you have to have a thought. You have to have an idea of how you can bring systems together because we all know many of you have invested in different lakes, different warehouses, legacy data stores. This is why we say that 71% of company applications tend to be disconnected. And we've all put a lot of effort through ETL and integration pipelines and to bring them together. But as apps proliferate, there are still more and more sources of data that you'd love to get at to power things like great AI, great automation, relevant insights from analytics, but you've got to bring that data together. The platform we've been investing in to do this is what we call the Einstein 1 platform. And if you're looking at this and say, "Hey, what happened to the Salesforce platform?" This is the next evolution of the Salesforce platform built to deliver AI enterprises. What is that? What it basically means is an enterprise that can bring together their data, connected securely to a large language model of your choosing, even a fine-tuned model, as we'll talk about, and then activate AI-rich insights in the flow of work. The most important thing that we're going to talk about today are really the flow of work and the data piece. The models are critical, but really, you need to have a great mode of activation, a great UI where users actually are, where they're connecting with customers and delivering value, and you need to have that data foundation in order. So we've got to think of ways we can bring that together. Now what's changed here, to demystify it a little bit, is that we brought Data Cloud fully into our metadata layer. So we have this rich lakehouse architecture powering Salesforce, bringing together not only transactional CRM data but also capturing real-time web engagement data and also data that might include point-of-sale data from Snowflake or inventory data from Databricks. All of this can be accessed and got at with Zero Copy and then activated for those key use cases like AI. And to make AI really usable and palatable for enterprises, we've instituted this Einstein Trust Layer, brought generative AI into this metadata framework so you can use new capabilities like Einstein Copilot, our conversational AI assistant, in the flow of work, grounded, utilizing that rich Data Cloud data alongside your CRM app and your industry solution data as well. So that's it for me. I want to pass it over now to Stella to talk more around how Data Cloud works and powers this experience. So Stella, over to you.

Thank you. I mean, first of all, we are able to bring our data into our platform in a very hyperscale fashion, right? We have an extensive library, a different type of out-of-the-box connectors native to Data Cloud, and also in partnership with our MuleSoft Anypoint Platform. And then once the data is in, you'll have the ability to prepare and further transform the data, whether it's using clicks or using code. Then the challenge the customer has is a lot of this data format is different, right? And this is why we provide you a common data model, allow you to be able to harmonize the data. And the result of this data model, then now you have true customer profiles. And this will become your single source of truth of the customer profile, then the real power of Data Cloud because it's, what Ryan mentioned earlier, part of the Einstein 1 platform. Now once you have established the customer profile, then you will be able to activate the data and profile into any of the applications within Salesforce ecosystems. And the next thing I want to talk about is the Zero Copy Partner Network, right, as Ryan mentioned earlier. So our platform, one of the 2 principles from a development perspective is we wanted to make sure it's always open and extensible. So what's new to Data Cloud is our Zero Copy Partner Network. So basically, we try to give companies really the flexibility and the choice to allow them to access data between different platforms and then further building AI-powered solutions, right? Salesforce introduced -- it's really considered an industry-defining moment that we introduced this partnership with platform vendors like Snowflake, Google BigQuery, AWS Redshift and then Databricks, right? So we are building a lot of the Zero Copy integration solutions. What is Zero Copy? Basically, you don't have to physically move the data and you don't have to worry about those developing complex integration programs. Then you can access the data and further empower your use case and applications. And we also recently announced our extended partnership with IBM Watson and also Microsoft Azure. And then the second part, as part of this Zero Copy Partner Network, is we introduced this concept of Zero Copy together with data kit. Now a lot of you might be familiar, right, data kit has become a very popular vehicle that our ISV partners has been using to help them distribute very rich, high-value data sets onto our customers' platform, right? The data will be landed in a pre-mapped fashion, so a customer doesn't have to do any transformation around it. And if you're asking what is a data kit, data kit basically is thinking about a small container, right, that contains the Data Cloud metadata components and allows you to package and then install from AppExchange. Then the data will be just landed in pre-mapped format. So with Zero Copy, together with data kit, now this just basically empowers a lot of our customers to be able to access our extensive network. For example, you're now able to reach access to partners like Dun & Bradstreet, Moody's and then others, so this data just coming to your platform and further enriching your customer profiles. And then the last thing I want to talk about, we also have a long list of leading industry implementation partners. They are in the field helping implement Data Cloud and also providing best practice recommendations about Zero Copy integration patterns. And then the second thing I want to talk about besides extensibility and open trust, now I'm focusing on the lakes. We also have an open model ecosystem approach that would enable us to meet you where you are on your AI journeys. For a lot of our customers, we offer out-of-the-box Salesforce managed models, that's probably just the right starting point, and then is able to satisfy a lot of your AI needs, right? From an LLM perspective, we lean on OpenAI here. We are rolling out some specific domain models that satisfy more advanced scenarios. But then we also introduce additional LLM choice, such as the use of different OpenAI models, right? For some of our other customers, if they are more mature in their AI journeys, and you wanted to choose your own developed AI models, we can support you on that as well. Either you can bring your LLM model API keys or we can host a model for you as well. And the same concept and principle applies to predictive models as well. So if you happen to have developed advanced models for Amazon SageMaker, Google Vertex and Databricks, right, you can use that. The idea is really to enable a lot of the predictive AI on your terms by using the data gathered from the Data Cloud platform. Now the next thing I want to talk about is more focusing on unstructured data, right? So generative AI is very powerful, right? It can largely improve a lot of our productivity and efficiencies. But it also has some limitations. It can be inaccurate because they are trained on a probabilistic view of the training data, and also the machine itself may not be aware of the most recent context of the information. So this is why AI needs more data. And guess what? A lot of the data actually cannot be easily stored in a typical column-and-row database fashion because it's unstructured. Think about all your PDFs or your transcripts or your e-mails, right? This type of data traditionally has been very difficult to search or to provide a good analysis on. So what Salesforce has done is now we are allowing for you to ingest a lot of those unstructured data into our platform and install in a vector database. Vector database, just think about it as something that's numeric, right, allowing a lot of the advanced search engine to perform semantic search to be able to generate more relevant results, serving the LLM prompt. Now let's have a double-click on the Data Cloud vector database, why we enhance the Data Cloud with this native vector database, right? So talking about the ingestion of the unstructured data, right, we provide you different ways either through the connectors or through ingesting APIs or you can even bring your own blocks. And then once the data has come in, there's a few more steps that's in the next, right? First, we do trunking. So what is trunking? The primary objective of trunking is to kind of break down a big object data into smaller, manageable, and then semantically meaningful trunks so that they can be accurately embedded. And this process is very crucial, first, to help remove some of the irrelevant tags like HTML text, right, so the noise can be minimized, and also to get the relevance of the semantic search, maximize it. And then the format of the vector database itself is just basically each individual vector is an array of floating point numbers. And the distance between the 2 vectors basically indicates the extent of how related they are. So when both structured data and unstructured data is transformed into vector embeddings, now you're able to provide -- support use cases like similarity search or provide additional context for the LLM groundings, right? And also on top of it, because this vector database sits on top of our Einstein 1 platform, we are now able to enable all these different business applications to further utilize the power of unstructured data, whether through automation or analytics or personalization use cases, right? And this is just a more detailed view of what happens when a user actually trigger a gen AI request, right? So when a user request comes in, such as an employee question or a customer question, right, this request is right away transformed into our vector embedding, using the same embedding model that we use to embedding your unstructured data. And then behind the scenes, there is the Einstein Copilot search that will start. The Copilot search basically will take this numeric kind of user format that has been transformed and run against a semantic search into the vector database. And then thinking about the initial request, and then the data is searched against, they're all in the same format. So behind the scenes, we can trigger some pretty powerful AI algorithm, for those of you familiar with AI algorithm like k-nearest neighbors, and so that is able to find the most relevant, semantically similar content to provide back to the prompt, right? And the results from those search service is now used to augment the prompt and sending to LLM. Of course, from a Salesforce perspective, because everything is trust layer, everything is passing through Einstein Trust Layer to make sure none of your data privacy will be violated, and then you get the most relevant answer coming from the LLM model, right? So zooming out, just kind of to summarize what I'm just talking about, right, this is really kind of an enterprise AI and data architecture from a Salesforce perspective. Like I mentioned before, right, we provide you a very holistic data foundation, so giving you different ways to ingest the data, right, whether it's physical copy or it's through the Zero Copy Network, right, allowing you to harmonize so you're building a single source of truth of the customer profiles. And once you've done that, then this will further empower the machine learning model. Again, the model itself is open as well, right? Either you can leverage the Salesforce native developed model or you can deploy your own model as part of this platform as well. And everything is going through Einstein Trust Layer. At the end, with AI and the model together, we can further empower a lot of the applications data in Einstein 1 platform through point-and-click, ease of configuration to serve a lot of automation, analytical use cases and AI use cases. Yes, going back to you, Minhaj.

Thank you so much, Stella. Thanks for walking us all through that. Hi, everyone. In case you joined a little bit later, once again, my name is Minhaj Khan, and I am a Product Manager for Prompt Builder. And we're very excited to show you a little bit about Prompt Builder and how it works well together with Data Cloud to enable amazing generative AI experiences in your org. Cool, so let's dive right in. Some of you might have used some of these generative AI tools, for example, ChatGPT, and trying out prompts and getting responses back and seeing how that magic happens. But the challenge is that prompt engineering is especially difficult for businesses. And one of the reasons for that is that it's hard to bring in your personal business context and your customers' data and your business data into the language model without having to manually copy-paste that information into this third-party tool. And that, in turn, introduces a new challenge, which is that you need to protect your personal data, your customers' data and your business information. So you may not want to be putting that into the open web where you may not have a contract and agreement with this provider for the language model, where your data might be going into a place where you may not feel completely comfortable and trust how that data is being processed. And lastly, we know that these cool new experiences with generative AI are only going to be the most powerful when they are integrated directly into the flow of your work inside of your CRM, without having to have this [ sole ] share experience between your CRM or your flow of work and going into this third-party tool. So with these three challenges, not being able to bring in your business context, needing to protect your data and integrating this AI experience into the flow of your work, we feel that it's going to be a hard challenge for the businesses to use this new technology. So how do we solve this problem? This is where Prompt Builder comes in, where Prompt Builder enables you to activate prompts directly in the flow of your work and the flow of the work of your CRM business users. Firstly, we enable you to build trusted generative AI with ease. And this is where the employees and the users you have at your company are going to be able to finish their tasks faster, by leveraging prompt templates that you have created that, for example, summarize or generate content with simple clicks. And the trust element of this is really important because we have something called the Einstein Trust Layer, which ensures that all of the data that is being sent over to the language model is going to be secure and it's going to be handled in a trustworthy way. For example, we ensure that your data that contains sensitive information is masked. For example, social security numbers, health records and so on, we mask that information, we make sure that doesn't leave the premises of your company. And then we also ensure that the models that we work with, for example, the OpenAI and other third-party models, we ensure that they don't have any data retention of your data and they're not training their models and doing anything else with that data. It's just to generate that creative response with your information and give that back to you. Secondly, and this is the really exciting part about Prompt Builder is that we have a really cool tool, the Prompt Builder, to enrich your prompts and your prompt templates with your CRM and customer data. This is where you can pull in the information of yourself or your customer given, for example, a sales lead or a service case, and use that information to have a tailored and unique response and a unique creation for your prompt usage. And we're going to show you a little bit about this in a second. And lastly, all of this cool new technology is embedded directly into the flow of your work inside of your CRM. This is where the prompt templates are going to be available to be used, for example, in your Lightning pages, in your Record experiences, in flows, and even in Einstein Copilot, where you can have a conversational interaction with the language model that is powered by prompt templates in the background, all coming through the Prompt Builder. So that's a little bit about how the platform works in the background. Let's talk about the specific examples of prompt templates that are available today in GA and you can use them. The first one is the sales e-mail prompt template. So starting off with this example. The sales e-mail prompt template allows you to create a personalized and tailored e-mail for a specific customer with just a single click. So what happens here is, when we have this prompt template, it's going to pull in your information and your data, which is your identity, your brand, your voice and tone and, for example, the salesperson's identity, and your product catalog and other opportunities that you may have. And then it's going to pull in the information of the customer that you're interacting with, their industry, their company, their role, what opportunities they may have open, and maybe, for example, what products they may have purchased in the past. And once we've grounded all of this data into this prompt template, we're going to be able to create a tailored and unique response for every individual customer interaction using the sales e-mail prompt template. And this applies to a variety of scenarios. For example, we have a template for a cold lead where you want to initiate contact. We have a template for introducing yourself. We have a template for following up on a warm lead that you already have been chatting with. And all of these templates are coming out of the box and you're able to clone and then customize them as well if you want to provide more grounding data or you want to change the instructions for how this data is generated and this creative response is generated. So that's the example for a sales e-mail. The second one I can share with you is the field generation example. Field generation prompt templates are when you create a new custom field on your CRM objects, for example, account. This new custom field is powered by generative AI, and it's able to create new content or summarize content for you directly in that record. So for example, let's say you have an account with 10 open cases and you want to get a summary for that account's open cases. You can create a new field, a custom field, on the account object. And you can create a prompt template that will power this new field and the prompt template will pull in all the open cases for that account and then summarize them. It will read through the context for those cases and then create a summary for those open cases generated, and then it will save that into that field. And this is the cool power of generative AI where it's making your life easier, making your users' lives easier, where it's doing a lot of this heavy lifting for you, reading the content, generating the content, summarizing the content and placing it directly into the flow of your work, so that the next time you go into the account, just by perusing through the account, you can see, for example, a summary of the open cases that are on this account. And of course, there's a variety of use cases that you can apply for your business with this technology. The third one is the record summary. And this is a really helpful tool when you're wanting to get a quick snapshot of what's happening with a specific record. And so today, you can use Einstein Copilot and by asking it a question about a record saying, "Hey, can you summarize this record for me," it's going to use the record summary prompt template, and it's going to read the information of that record. For example, its fields, its related lists, the information that's available on the page layout. And then it's going to basically create a snapshot summary for you for that record and just give you a high-level view of what's going on. The last one, and even more excitingly, is the flex template. You can think of the flex template as a custom prompt template. In the previous three examples, we have out-of-the-box templates, which take in a specific set of objects, for example, an account or a contact for the sales e-mail. With the flex template, you choose what objects you want to pull into the prompt template, whether it's, for example, a field service asset or products from your product catalog, or any other custom objects that you may have or industries-related objects that you may have. And once you pull in these objects to your flex template, you can ground with that information and then you can provide whatever prompt instructions that you want to give to do something with that information and generate new content. And this is available in the GA today as well. So now that we've talked about the Prompt Builder platform, some of the out-of-the-box prompt templates, we want to dive a little bit deeper into grounding. And this is going to connect with how Data Cloud comes in and partners well with Prompt Builder to see how we connect with the Data Cloud data as well, as we'll show you in a few minutes. So just to give you three high-level overviews here for what does grounding mean. So the first one is you're able to ground your prompt templates with fields. This is where you can pull in fields from your objects as well as related lists from those objects, for example, a set of cases on an account or a set of contacts on an account and so on, and using this information enables you to enrich your prompt templates. Secondly, you can ground your prompt templates with flows and Apex. This is where, if you want to have a little bit more rich and complex data sets that you want to pull in, or you want to have a conditional logic or a custom set of information that you want to pull in, we enable you to use flows or Apex invocable actions to ground that information directly into the prompt template to generate your content. And thirdly, and most excitingly, what we're going to show you in a little bit in a demo today, is grounding with Einstein Search. Einstein Search is what allows you to retrieve relevant structured and unstructured data from your data cloud, including external and third-party sources that can be ingested into Data Cloud. Now what does it mean to use Einstein Search? I'll talk a little bit about this. Einstein Search uses a technology we call vector search, which is a little bit different from what you might be used to with keyword search, for example, with the Google search engine using keyword search. In keyword search, you're mapping with keywords that match and you're able to find documents and pages that have those similar keywords. With vector search, what's happening is we can take a large set of content and then we can do a mathematical operation on it to find similarity with other large pieces of content that you might have. So for example, if you have a long conversation on a case or an e-mail engagement with either a service scenario or a sales scenario, we can actually vectorize that content and then we can use vector search or Einstein Search to find similar content to that. So if you have a previous case that was already closed, and you have a current case, you can actually do a similarity search and find cases that were in the past that were closed to your current open case. And when you surface that information, that gives you even more rich context to ground your prompt templates with. And we're going to show you what this means and how it works in a few minutes. Awesome. So I want to make this a little bit more visual, and we've talked a lot about grounding prompt templates. So I want to use an example here to show you what does it really mean, if you haven't seen this, to ground your prompt templates and how does the content get generated. So starting on the left, we are looking at a prompt template that you're going to be seeing in your Prompt Builder inside of your org. And this prompt instruction says, "You're an agent at this organization, your client is this contact and they're at this company, and here's the profile of the customer." This is what you're typing in as a dynamic merged field inside of the Prompt Builder. Now for a specific customer interaction, for a given user and for a given customer, all of this data is then hydrated in the prompt and then it's sent to the language model. So when you look over to the right, in this example, now this data is available, it says you're an agent at Cumulus Financial, your client is Denise Martinez, and she's working at this company, who has been a customer for this much time and so on. So all of this information is then hydrated, which enriches your prompts and your prompt templates with unique data from your CRM, so that it gives you a more tailored and creative response for what you were trying to get out of the language model. Going down below on the left, we have the example of grounding with flows. You can say, "Hey, the customer service agent response is in this conversation." And then using a flow, we pull in that set of data for that conversation. And then below that, it says you can use these knowledge articles that are relevant for this customer and then a search retriever is used to pull in relevant and semantically similar knowledge articles for a given customer conversation. And on the right, all of this information is hydrated. So these screens are available in the Prompt Builder where you can type in your prompt template, bring in the grounding data that you want, and then you can see the result prompt, for example, a given customer. And then you can see ultimately the response that you get back for what you wanted to generate with the language model. All right. So we've covered the Prompt Builder. We've covered the grounding scenarios and how it looks and works in production today. One more thing I'll leave you with is the concept of data retrievers. So all of the cool enriching and grounding elements we've talked about are powered by these data retrievers, which is a kind of a component or a tool that's used to enrich your prompt templates. On the left, we've talked about CRM fields and doing field retrievals. We've talked about flow retrievals. And I'll introduce to you a couple of new retrievers here that we're working on. We have the data graph retriever that's going to be launching as well. This is where, for example, you have your data cloud instance and you have third-party sources for your customer. For example, you have a customer that has shopping card engagements or past purchase histories in a Shopify service that you're running for your business. Data Cloud is able to ingest this information from a third-party source like Shopify and then create a 360-degree view and a unified profile for an individual customer that includes not only your CRM data that you have in-house but also this third-party data that's been ingested with the Data Cloud. And then once you've created this 360-degree unified profile, you can bring in this data graph into Prompt Builder and into your prompt template and then you can provide it whatever instructions you want to provide to work off of this information and generate content based on what you've retrieved from all of these different sources. And then, of course, we have the MuleSoft as well where it's an API hub, which enables you to interact with any kind of third-party APIs and bring in that information. And lastly, but not least, at the bottom is the search retrieval. We've talked a little bit about this where you're using vector search to grab either structured content or unstructured content, for example, from chat histories, e-mail engagements, previous cases and so on, and then using vector search to bring in the relevant pieces of content and then use that content to ground your prompt templates. And we're going to show you a demo of how this works in a few minutes. All right. So this is a high-level view of how does vector search or Einstein Search work in the context of Data Cloud, in the context of Prompt Builder and the Einstein Trust Layer. We won't go too deep into this, but I will share a couple of things starting from the top left. So there's a few steps that we're going to show you in the demo, and I'll just kind of lay them out in front as well, which is, firstly, when you want to use Einstein Search, you first ingest the data that you want to search against. This could be third-party information coming from the outside. Once this data is ingested, we go through a process where we create embeddings, vectors and indexes from this ingested data. And ultimately, what you get is a vectorized index that's stored in the Data Cloud vector database, and this is ultimately what's used to do the search operation. Now all of this stuff can be done in a single click as we're going to show you in the demo as well, where you don't have to actually go through the process of figuring out the math and all the embeddings, and all of this is done for you, as well as giving you the flexibility to do more creative customizations, if you would like. Now looking on the left here, we have an engagement from a service agent with Einstein Copilot. And the agent is saying, "Is my customer eligible to upgrade their card?" Now in this specific example, the ingested data could be a set of e-mails, where you have a database of e-mails that are being pulled in and you want to do a semantic search or a vector search against that to solve for this query. So once this user query comes in, it's going to be embedded and vectorized so that it can be compared against the corpus of e-mails that you've ingested. And this is where the Einstein Copilot search comes in. It's going to use the user's query, and it's going to use the ingested and indexed information. And it's going to combine that and do a search to find the most similar content for the user's query from the database of information that it has found. Now this grounding context has been included into your prompt template and then your prompt template has been augmented. And as we mentioned, with the Einstein Trust Layer, all of this is done very securely where we make sure that we mask any sensitive information and we give you some controls to do that masking as well, more coming in the future for those controls, and then we work with some of the models that we have coming out of the box, for example, the OpenAI models, other third-party models as well as models that you may be hosting yourself, if you would like. And we make sure that there is no data retention with those models, and then you get back the response. After you get the response back, there is toxicity detection where we make sure that the response that's gotten back is safe. And in the future, we'll add citations as well to give transparency on where the answer came from, for example, with Einstein Search. Then we save that request in the user's utterance and the response in the audit trail. And then ultimately, you get the response back for your conversation in the Einstein Copilot. And in this example, the answer is, based on prior e-mails, this customer is eligible for a card upgrade and here is the source for how that happens. So I know that was a lot to take in, starting with Prompt Builder, the framework, grounding information and then partnering with Data Cloud to ingest and then connect with Einstein Search to enrich your CRM users' interactions. So with all of that said, we're going to tie this up with a demo that shows you, stepwise, how all of this is working. It uses Prompt Builder as well as Data Cloud altogether to enrich a CRM user's experience, and we're really excited to show you how this works. So we're going to play the demo now, and we hope you enjoy. [Presentation]

All right. Welcome back. I hope you guys enjoyed that. I know that was a lot of content in 8 minutes, and we started with Data Cloud pulling in this third-party data from Amazon, indexing it and then creating a search retriever and then using that in a prompt template, and ultimately tying it all together with a flow and then powering that CRM user's pain point and enabling them to do their work much more easily. And we're going to be releasing resources over time that are going to give you the same previews that we've shown you so that you can understand and figure out how you can use this technology for your business. So I really hope you enjoyed that. I'll wrap up with one more thought here. So this is the release road map for Prompt Builder. So again, subject to change, and we're going to be working on this stuff diligently over time and see how the themes shift as well. So you can take a look, there are 4 or 5 categories that we have, or the themes that we have, for our road map. There's the core features for Prompt Builder at the top. Then we have the increased flexibility and use case coverage, and then we want to expand grounding as well as enhance trust. So I won't walk through every single item here, but just to call out a couple of the pieces. So now what is currently GA is, for example, having inactive or active templates or bringing your own language model, right? In the increasing the flexibility, we give you flex templates that we've talked about. We have metadata API support as well as we have basic localization support as well. And then for grounding, we have today for you the flows in Apex. You can access some data cloud objects directly today with the enrichments framework, and we have the pilot for the Einstein Search with the Apex as well. And the solution we've showed you is the GA that's coming in a few months. And then for trust, of course, we have the sensitive data detection and masking as well. In the future, what we're building for you and what's going to come next is, starting from the top, we want to improve the error messages that you might get back for a variety of scenarios. We want to give you the ability to put in flexible strings or text directly into the prompt template, so that you can use any user utterance from anywhere in your org and directly put that into the prompt template. And of course, we've talked about retrieval augmented generation or Einstein vector search grounding that's being worked on. And something we're going to launch for you also is the record snapshot grounding, which is the summary of a given record directly into the prompt template. And we want to increase data masking and localization capabilities as well. So that's a bit of the future road map we're excited to work on and show you. So please engage with us in any of the other channels, and we're happy to hear your feedback, and we're happy to answer your questions as well. And with that, I will hand it back to either Stella or Ryan. Thank you.

Thank you, Minhaj. So here, I'm going to talk a little bit from a Data Cloud road map perspective and what different types of road map themes and innovations the team is working on. So one thing I wanted to focus in here is the real-time data graph, just like what Minhaj just mentioned earlier, right? For those of you who are not really familiar with Data Graph, Data Graph really is just, before the generation of Data Graph, if you wanted to retrieve data from Data Cloud, you'll have to make individual queries per call going to different tables. But the concept of Data Graph is to allow you to define a graph that's more relevant for a particular sales scenario or service scenario or marketing scenario, right, so you can have a data graph, including customers, sales order items and products and everything. What a system does behind the scenes is having those joints relationally calculated and then sit in the cache. So when you wanted to retrieve information for a sales purpose scenario, you can just call in the Data Graph, the latency for such retrieval will be much, much faster. And then in our most recent, which we're going to announce in the near future, real time, we're talking about end-to-end scenario, like if you're browsing something on the website and you can immediately have this information available on your service console, for example, right? So that's in Data Graph. We talked about open extensibilities. And then another thing, from an enterprise robustness perspective, for those enterprise customers, a lot of them have been asking for more object level, field level securities, and these are on our road map. And then we have something pretty soon coming out to address those, right? And then our team is also putting a lot of focus, deep focus, on platform integrations. Really, the idea is we want to make sure -- we're constantly talking about Data Cloud and that it's part of the Einstein 1 platform, so everything is exactly the same metadata service approach, right? We want to make sure our customers is able to access the data the same way as how they are able to access any of the CRM data. So that's our North Star goal, and we are taking incremental enhancements to make it happen for our customers. Yes, I think that's pretty much I wanted to cover from a Data Cloud road map perspective. Ryan, going back to you.

Thank you so much, Stella and Minhaj, and thank you both for just spending some time with us, getting right into the details. We really appreciate it. Just a quick call to action for the folks here. If you want to get going now with this data and AI journey, realizing some of what we've discussed, first and foremost, hit up sforce.co/freedatacloud. It's free Data Cloud. You have a $0 SKU you can provision to actually begin getting started with Data Cloud with a couple of prebuilt credits. And credits mean a lot to you that can enable you to start experimenting with this. So you don't need to wait. You can actually start using Data Cloud today, seeing where it works for your business and you can extend it so that it meets your needs. And then go look up our new einstein.com web page to learn more about generative AI and also our trust layer, which is how we implement security guardrails that make generative AI palatable for all of you in the enterprise, and I know we all are. Finally, scan this code. We have a really good new Data Cloud trail. That QR code will take you to that trail and get hands-on with some of what Stella and Minhaj have discussed here. But now with just a few minutes left, I want to turn it over to really the group of you all here with some questions that we picked up from the audience.

Stella, my first question is for you. There was a question around if Zero Copy, the Zero Copy pattern, can work with Databricks. Can you speak to that?

Yes. I think we already have Zero Copy integration with Databricks. I believe, and Ryan, you can keep me honest here, I think it's coming GA in this June, right?

That's correct. That's my understanding.

We actually have a lot of customers who have been trying the pilot together with Databricks. So we have a couple of pretty big enterprise customers who has been using Zero Copy with Databricks and then bring the data over on to Data Cloud platform.

Awesome. Thank you so much. Minhaj, this next one is for you. This is a Prompt Builder question. Someone here was saying they use Data Cloud. They've used Data Graphs, which are these kind of de-normalized collections of records, for those who don't know. They were asking if there's a road map for using Data Graphs within Prompt Builder. Can you speak to that?

Absolutely. And that's a great question. Thank you for asking. Yes, we have on our road map to include Data Graphs directly into Prompt Builder, a first-class UI where you can retrieve the Data Graphs through the resource clicker and then include that in a single click. That's going to be coming in the next few months as well.

Awesome. Thank you so much for that. One more, too, for you, Minhaj. This is a question around flows. You mentioned dynamic grounding with flows. And you might have spoken to this as you went, but we have this question around can you use flows that pull resources from Data Cloud when you're dynamically grounding a prompt. So can you create a flow that pulls, let's say, some point-of-sale order information from Snowflake into prompt? Can you talk to that?

Yes, absolutely. So that is something that works today with flow as well, being able to use flow to retrieve the Data Cloud information, whether it's from Data Graphs or DMOs and then using the flow to ground that information. That's the current solution to enable Data Cloud grounding into prompt templates. And as I mentioned, in the future, we're going to have a first-class UI where you can directly engage with Data Cloud resources inside of Prompt Builder without having to use Flow.

Awesome. Stella, this is a quick tactical question for you. You mentioned the vector database, and there was a question around when that will be generally available, if you can speak to that.

I think we are going to announce the unstructured data vector database GA in June this year time frame.

Awesome. And there's a question for me, which is can we get these slides. You can get these slides. We're going to share the slides and the recording, so don't worry. If you took aggressive screenshots, that's okay. But we'll give you some good, high-res stuff in a minute here. I've got one last question over here that I'm seeing. And this might be a question for you, Stella, that kind of goes into a road map, which is can we connect the third-party data sources in the future like ZoomInfo?

ZoomInfo. I do believe ZoomInfo, as I recall, is part of our Zero Copy Partner Network. So I'm pretty confident we'll have a pretty solid partnership with the company in the future, yes.

Yes. Girish, I think you asked that question. Absolutely. If you look at our World Tour New York press release, creating a Zero Copy data kit for ZoomInfo, among many others like Moody's and even Workday, is something that we're investing in, and we'll have more on that later this year. Let me just see here if any other questions came in while we were chatting. Minhaj, I see you answering a couple of them. I think that's pretty much it, folks. So we can give people a couple of minutes of their day back as they say. But on behalf of this crew over here, we really appreciate your time and these very thoughtful questions, and just stepping out here to learn more about Data Cloud and AI and how these are working together. As I mentioned, the slides and the recording will be available to you. Please check out the resources available in the webinar, learn more about Data Cloud, learn more about AI and get started. We'll talk to you soon. Bye, everybody.

Thank you.