Accelerate Semantic Modeling with Data Pro and Built-in AI | Dreamforce 2025

Thanks again for making the time to be here and join us and learn about the agentic capabilities the semantic the Tableau Semantic platform has to offer. If you haven't done it yet, I highly recommend to go ahead and add the Tableau keynote session that's happening tomorrow into your agenda. It's going to be super exciting. My name is Daniel Alexandroni and joining with me here today is Neil Vaknin. We're both part of the Tableau Semantic Product team. You're probably asking yourself what is Tableau Semantic Team? Right, so let's dive into this. Let's start with Tableau Next. Tableau Next is the first analytics platform in the world which is an agentic platform. It allows you it's built on Salesforce and it's fully integrated with Data Cloud and AgentForce and it allows you great performance and scalability and anything that Salesforce has to offer. Within Tableau Next we have Tableau Semantic. Tableau Semantic is an AI infused semantic layer that builds within Tableau Next and fully integrated with Data Cloud, it provides great UI and allows you to create single source of truth for all your metrics and models and business context that you have in your organization. Within Tableau semantics you have a few agentic capabilities that help you to accelerate the workflow of the model creation that you will hear about today and you will be able to understand how with semantic layer you allow agentic capabilities to provide much more accurate answers and to make sure that you're covering and that you will have the same numbers across all the dashboards and the entities that you will create in your org. Within Tableau semantics you have the Data Pro skill and some additional built in AI capabilities that again will help you to accelerate the process. For example, within Data Cloud datapro skill you have the Create calculated field action that allows you to provide just a natua language description for the calculated field you want to create and get it out of the box. And you also have the auto SDM capability that will let you to talk like you're talking to a friend or a colleague, provide an Atom language description and get the semantic model fully created ready to use. So about these features, I'm going to show you some more how it works. So let's dive into a quick demo of this capability to make sure you can trust me. Okay, so in today AI's world there are two critical things that affect the speed and the accuracy of your data. First is the quality of your data and second is the context and traditional data modeling. Maybe I'll stop it for a second. Traditional data modeling, if you would like to create it manually from scratch, it requires you a few multiple tasks, such as selecting the data models which can be standard or custom. It might not be that easy. You will need to create relationships, calculated fields which are like formula. It's a technical task to do. You will need to know the syntax and everything. You will also need to create metrics. You will need to do the entire workflow. And then that small task of creating the semantic model might take a few hours or even days. But with auto SDM that we will see in a few seconds, this task drops to zero. So here I am in my Tableau Next workspace and let's imagine that for the sake of this demo that I'm a sales analyst. I want to analyze the sales performance. So I have here all the data objects that I decided that are relevant for my org and I'd like to provide a description for the agent for the model I want to create. Now I can type in something very generic, just I don't know, something very generic. But I'm an expert and I know exactly what I know, as I'm sure your teams are. So I'm going ahead. I will type in a very detailed description for the model that I want to create. I will tell him that I want to create a sales performance model to connect accounts and the contacts and opportunities and of course the products that are tied to them. So I can understand my sales performance and see which products, for example, drive revenue. And in just a few seconds it analyzes my intent and generate a plan for which semantic entities need to be created. And now I can review the model. I will say that keeping the human in the loop is crucial for that task because we need to make sure that the generated model is aligned with the main business needs. And also of course, we're making sure that this model will be AI ready, so it will be created automatically for agentic experiences. Then I will move on to review my model and I can see that the agent selected great data objects. I can see account, account, contact, opportunity, opportunity product, which include by the way, both standard and custom objects. I can see that it generated the relationships. That makes perfect sense. I can even see that it generated some calculated fields with the formulas and everything. So it's ready to use this model now is ready to use. I can share it, I can build visualization on top of it dashboard and of course ask concierge question about it. Maybe a quick note that Tableau semantic is now GA and available for any Data, cloud and tableau customers and that's it. This is auto SDM capability. It will be available soon and now I'd like to hand it over to NIR that will show you some additional amazing capabilities in this domain.

So what happens when you already had a semantic model? What happens when the model that Danielle just showed us got here? The SDN is still required to be AI neck Well, we want to design for agents for sierge and analytics doesn't necessarily know what to do. The all used to design this semantic model for dashboards pretty clear the methodologies we are all aware of. But to make it ready for AI and agents, it's a new job that they need help with. After the research we got feedback from many analysts that they are looking for guidance and instructions to clean products. And this is where SDM the IRP editions we talk about a flow where user can develop that at the end or come back to it when they develop what they know what they need to do and then the view and to make it ready for AI agents. No worries. Can you hear me better? Thanks. I'm not a singer so you will have to apologize to you and come back and forth to this one. Okay, so here we have the optimized model panel. I opened it and I can see the model health. The model Health provides me a good indication of how ready I am for an agentic use and a progress bar that reflects to me how many tasks do I need to do add to address in order to complete to make it ready. Important to add that this feature is scanning the semantic model in advance. Sorry. Oh good. Scanning the semantic model in advance, checking all the relationships, checking the semantics, the models and making sure that it's ready for AI and surfaces all the issues in here. The first issue that we can see here is missing relationships. When we developed the dashboards in the past, the dashboard and the dashboard designer knew what they need to do and this is why they already assumed they can allow no relationships in this case. Okay, so I will click the suggest relationships and I can get here a recommendation for what I should add. Without this relationship I won't be able to ask questions between the product catalog table and the sales table. So I'm going to apply this recommendation for now and move on. The SDMI optimization is being updated on the fly and we can see that model Health is up to date. The next rule that we are going to take a look at is the missing descriptions. Descriptions so far took a very secondary job in the analyst role. They used it in Documentation, other analysts got that documentation or to that data catalog to learn more about this semantic model and how to use it. But but our agent they need this information in the semantic model and the importance to highlight that the missing descriptions that we are seeing here is only for custom objects because for standard objects I already have that knowledge. Based on the LLM training. Let's take two examples. The first one is compliance code. Can you tell what is compliance code and what we are referring to? With this field I assume that you can guess or assume what it means. But for my business and for this example it has a more specific purpose. We are going to use here another hidden feature description generation that will allow me with a click of a button to generate the description. It's processing industry knowledge, the semantic model itself, all the semantics that it already includes and proposes me a description for me to use. Now I cheated here a bit and it's already incorporated the business context that I was referring to which is the EU reach and FDA standards. And for this use case I'm good. I move on. Another example where I need descriptions is where I have ambiguities. In this example I have two tables of products, one of them product and the second one is product catalog and I need to instruct better the agent on how to use those tables. So in here as well I will generate that description. As an analyst I have to review and make sure it aligns to my expectation save. We'll do the same in here and then I can move on. The SDMAI optimization is up to date. I got a high score so and I am ready for starting to using it with agents. Oh, I forgot to mention that descriptions can always be generated through the properties pages itself. This feature is available in many other areas of the product. The second next feature that I'm going to show you is business preferences. What happens when the the logic doesn't exist on a specific field or a table? What happens if that logic is somewhere in the user utterance? In the business question itself, Let me show you an example. I will jump into a dashboard open analytics agent and I will paste my first question which is show me the best opportunities in previous year. Pretty straightforward business question. I expect it to answer only. I'm asking here on best opportunities and a previous year. But as it seems the agent was not able to provide me response. The reason is it's asking for clarification on best opportunities which makes sense. The agent doesn't know what I'm referring to when I'm saying best opportunities. I can Move on. And I'm sure that if you will see more demos, you will see that on typing in here my intent, I will get a response. But I don't want my users to every time type in their intent when they are referring to best opportunities in my business. And in this example I want to make sure that every time that a user is asking about best opportunities, it will do get the same response. This is where business preferences comes in. So I will simply type in here an instruction for when a user asks about best opportunities. Return to me the top five highest opportunities by annual revenue amount. The annual revenue amount is one of my measures in my semantic model. Now let's check again the agent response based on the new instruction. I just replaced my question. Every time I'm sending a question to the agent, it's taking all the semantic model a context that it has at the moment. In this case it will take the new instruction and I am aiming that it will address it in a few seconds. Usually it takes much shorter. This is the risk of live demos. Okay, takes more time than expected, but at least it returned a response. Cool. So I got a response. I have here a top five opportunities from the previous year ranked by annual revenue amount. This is exactly why what I provided in the instruction, which means that my instruction went well. I can double check that in this resources and and make sure that the fields, the right fields were used. Amazing. So what we saw right now are. No, not shell. What we saw right now is the power of Tableau. Next, with datapro and built in AI capabilities, we started with auto SDM creation, the starting of the journey of the analyst where they can input with a natural language a request to generate a model and then get in a few seconds a suggestion to start with. Then we saw the SDM AI optimization which is a feature that accompanies the user like an assistant on the fly and helps them making sure that their semantic model is ready for AI. Of course we saw on the way the description generation another hidden AI agent that is helping us set typing text faster. And we concluded with business preferences. That helps us tie the dots between the different objects that we have in the semantic model and the third dot which is the user itself and the user jargon and context. If you would like to hear more about our features, we have much more to offer and much more to show you. So please visit us at the campground. Our team is there to show you more, to answer questions and to refer to your feedback. Thank you very much everyone.

Thank you.