The Data-Driven Community Meetup holds monthly webinars on business analytics and big data. Webinars are held on the second Wednesday of the month at noon (12:00 PM) Central Time via Zoom Webinars and will cover topics related to enterprise data management. Our goal with each webinar is to provide meaningful insights and actionable takeaways to simplify analytics so you can make better decisions.
We cover topics such as data strategy, data management, data warehousing, BI modernization, embedded analytics, and cloud migration and strategy. Learn how to build reporting solutions that drive your business demand based on your needs.
About the Topic
In this event, Celia Fryar, Training and Enablement Lead at XeoMatrix, showcased how Tableau’s next-generation capabilities can transform your analytics workflows and enhance decision-making.
Celia walked us through live demos of:
- The power of Tableau’s Semantic Layer for consistent, governed data across dashboards
- Dynamic visualizations and dashboards built with TableauNext
- Seamless integration with Salesforce, showing how embedded analytics can bring insights directly into your CRM
This article includes a recording, transcript, and written overview of the presentation on TableauNext & Salesforce Integration Demo (On-Demand Webinar).
TableauNext & Salesforce Integration Demo (On-Demand Webinar) Presentation Video
TableauNext & Salesforce Integration Demo (On-Demand Webinar) Summarized Presentation
The TableauNext & Salesforce Integration Demo session offered a comprehensive look at how Tableau Next is reshaping the analytics experience, both in its core functionality and through deeper integration with Salesforce. Led by Tableau expert Celia Fryar, the presentation showcased how the platform introduces a more modern, flexible, and structured approach to data modeling through its new semantic model. Celia emphasized that while Tableau Next builds on familiar principles from Tableau Desktop, it represents a significant step forward in usability, portability, and AI-powered analytics.
A central portion of the session focused on embedding Tableau Next dashboards directly into Salesforce to create seamless, workflow-driven insights. Celia demonstrated how to build a visualization within Tableau Next, configure Salesforce actions, and then surface the dashboard inside Salesforce using updated page layout tools. The result is an interactive analytics experience that responds dynamically as users navigate between Salesforce records, eliminating friction between data exploration and business operations.
The session also explored Tableau Concierge, the platform’s emerging agentic AI assistant. Using the Superstore dataset, Celia revealed where Concierge already performs well, such as calculating sales metrics, and where it still struggles, particularly with more nuanced logic or ambiguous queries. Strong data governance, clear business rules, and thoughtful semantic modeling can significantly improve the quality of AI-generated insights.
Looking ahead, Celia previewed new features coming to Tableau Next, including business terminology training, semantic model readiness checks, and a dedicated testing center for validating AI responses. These enhancements signal a future in which Tableau users spend less time configuring analytics and more time benefiting from automated, intelligent insights. Overall, the session positioned Tableau Next as an evolving, promising platform that blends modern data architecture with practical, user-driven AI capabilities.
Session Outline
- Part 1: Building a Tableau Next Visualization & Embedding It in Salesforce
- Part 2: Exploring Tableau Concierge (Agentic Analytics)
- Upcoming Enhancements to Concierge & the Semantic Model
- Q&A Highlights
Part 1: Building a Tableau Next Visualization & Embedding It in Salesforce
Introducing the Semantic Model
Celia showcased Tableau Next’s semantic model, a unified environment combining data lake objects, relationships (including Einstein-suggested ones), metrics, calculated fields, parameters, and logical vizzes. This space replaces the traditional “Data Source” page and adds interactive modeling and descriptive metadata.
In the demo, Celia included Salesforce Sales Cloud demo data and added an Excel file containing customer sentiment reviews. Tableau Next recognized it as a data lake object and incorporated it into the model. She previewed fields, explored the tabular data view, and pointed out similarities and enhancements compared to Tableau Desktop.
Drafting Calculated Fields with Einstein
Celia demonstrated the “Draft with Einstein” feature, which assists with formula creation. This feature is particularly helpful given some language differences between classic Tableau formulas and Tableau Next’s structure.
Building the Visualization
Celia built a visualization from scratch using the semantic model:
- Fiscal week on the horizontal axis
- Total number of reviews as the measure
- Sentiment as the categorical split
- Account name on detail
- Circle marks colored by sentiment
- Labels enabled and tooltips explored
In this viz, we can see improved color spectrum controls and how many interactions mirror Tableau Desktop while adopting more Cloud-aligned UI patterns.
Adding Salesforce Actions
A key integration step involved enabling Salesforce actions such as “Open the Record on Select.” This allowed users to click a mark in Tableau and jump directly to the corresponding Salesforce record.
Configuring the Salesforce Integration
Celia then demonstrated how to embed the Tableau Next dashboard into Salesforce:
- Editing the Salesforce page layout
- Adding a new Analytics tab and dragging in the Tableau Next Dashboard component
- Selecting the correct dashboard from the panel
- Adjusting component sizing
- Configuring filters so Salesforce passes the right context (e.g., matching Account Name)
- Saving and activating the layout
Switching between accounts in Salesforce dynamically updated the embedded Tableau dashboard, providing a seamless integration experience.
Part 2: Exploring Tableau Concierge (Agentic Analytics)
Testing Approach
Celia used the Superstore dataset for her testing to avoid sharing private client data. She asked progressively complex questions, examined how the agent handled Superstore’s multiple-table logic (including the tricky returns table), and evaluated accuracy.
What Works Well
The agent performed well when calculating:
- Gross sales
- Returned sales
- Net sales
Celia also created supporting calculated fields to give the agent better guardrails. Over several months of testing, she observed a clear improvement in the agent’s ability to handle more complex queries.
Where Concierge Struggles
More advanced requests, such as identifying the top customer per region or questions requiring window functions, produced mixed results. For example, the agent could identify the top customer overall, but not per region. When asked for “top contributors,” it interpreted this as “most orders” rather than “highest revenue.”
When it couldn’t complete a request, the agent provided helpful guidance about what it needed—clearer definitions, different phrasing, or structural business rules.
Inconsistencies & Errors
At times, ambiguous language led to semantic query execution errors. Celia emphasized the importance of refining prompt clarity and building stronger semantic context within the model.
Upcoming Enhancements to Concierge & the Semantic Model
Celia previewed features coming in the next beta cycle:
Business Preferences
A new panel where teams can document:
- Acronyms
- Business terminology
- KPI rules
- Formatting preferences
- Internal calculations or definitions
This is designed to help the agent act more like a fully onboarded analyst.
Model Readiness & Test Center
The semantic model will soon include:
- Automated model readiness checks
- Instant description generation
- A Concierge Testing Center for validating answers and scoring accuracy before deployment
These features aim to support more reliable enterprise-wide AI analytics.
Discussion: Data Governance & Documentation
Celia repeatedly emphasized that good data governance—clear naming conventions, complete descriptions, well-defined business rules—is essential for reliable agentic analytics. Many organizations lack strong governance foundations, and Tableau Next makes this increasingly important.
Q&A Highlights
Key Q&A topics included:
- Licensing: Tableau Next access inside Salesforce requires correct permission sets, not just a Tableau Next license.
- Einstein consumption credits: Credits are evolving; the new digital wallet provides real-time tracking.
- Zero Copy architecture: Tableau Next uses referenced Salesforce data without duplication, reducing cost and improving governance.
- Performance considerations: Zero Copy helps reduce duplication and streamline performance, though more testing is ongoing.
Transcript
>> CELIA FRYAR: All right, a little bit about today’s agenda and meeting format and such. I’m Celia Fryar, and I’m going to be talking today about Tableau Next a bit. I’m a Tableau academic ambassador. I’ve been working with Tableau, in and around it since 2012-ish, ’13, but more of author meets first-hand and then instructor 2015, ’16. I’ve seen it change a lot over the years. I’ve been involved in their beta program for a long time as well. It’s always interesting to see at what point do they introduce you to beta programs, right? Is it kind of beta? Is it kind of often? It’s almost general availability. It’s been interesting to watch the scope on that over the years. I’m also going to be joined later in the hour by Blake. I misspelled his name. Blake Wade is his name. He is one of our analysts who has actually done, just completed a proof of concept for a client. We’re using live data production environment with Tableau Next. I’m glad to have him here. He’s offered to come jump in for the Q&A part of our meeting today. Lauren, my co-host, we have a slide here in a minute, but she’s usually the one who’s speaking to us on these meetings. I’m really grateful that she’s here to keep– She’s going to be watching chat if there’s any questions and also just help with any of the Q&A as well.
We are virtual. We have very much intentionally created this environment for it to be interactive. We’re very excited for you to please participate in the chat together as we go through the time. We will be having kind of a little bit more structured Q&A at the end because I have some things I really want to be sure and share and show, but if you have comments or questions as we go, I’ll do my best to catch them. We’ll definitely have a purpose time at the end for Q&A. We’ll have a link to today’s recording that will be distributed for anybody who is registered for the meeting today. It is being recorded for sure. I just double checked, so we’re good to go there.
As always, if you have ideas or recommendations or suggestions on topics that you would find very helpful to be covered, there is a space for that in the follow up information in emails. You can also just email Chris Monahan, who is our president and CEO, and he will most certainly– He sent Lauren and I a couple of specific examples of good things that people were asking for. I think [unintelligible 00:02:35] that come through last night. Lauren and I co-host this session together monthly and we really very much try to make it topics that you are asking about, caring about and/or that what we’re seeing our customers be super involved with.
Lauren, she is a gifted analyst and designer of all things visual. I just have more of a logic brain and I am always in awe of her skills. She leads a lot of what we do at Xeo in terms of visual analytics and creative solutions for connecting the data and data prep and all of that. She’s been a marvelous partner for me on this and also just a delight to work with. I have also in some of my spare time, I’m a part time university professor at USF. That’s me being out in California a couple of days ago. Then I’ve been in data my whole career, but analytics the last decade plus, and something I very much enjoy, really like being involved in a data driven decision kind of ideas. Let’s continue, I guess, and jump into Tableau next.
Before I actually start talking about what I have prepared for you guys, just want to see, have any of you in the chat, if any of you have tried Tableau Next or been involved in the beta or any of that, I would love to see that in chat. A no is fine too. I’m just curious who’s in the room in terms of experience that we might have. Anybody? [laughs] Okay, Blake, I see you. Okay, awesome. That’s hilarious, sir. All right. Okay. Good, good. All right. Well, we probably at some point we should do a little bit more of a welcome to Tableau Next because there is language and labels that are different than Tableau Desktop and Tableau Cloud.
However, there is a familiarity to it that is very like a common thread, if you will. It’s not like starting over by any means, but there are some devices and ways of doing things. Please do keep in mind that it is general availability in its base form, but it’s– I remember when Cloud came out and how that we were all like, “Wow, that needs more features.” Well, they’re hard at work and bringing more features to it. This is the first iteration of it, of course, and it went GA with 2025.2, in that timeframe. It’s been fun to see it grow even in the time period that I’ve been participating in the beta test, because they introduced us to the beta in May, and I would say July and August were really big steps up in terms of stability and features and things that we have come to expect from the brand of Tableau.
This month, in fact, in five days, we’re starting a new beta internally. They are starting on their new internal beta, and they’ve invited Tableau ambassadors and visionaries, and I’m sure others, to get involved because they’re building up more of the features around the semantic model. To my mind, the semantic model is the big deal here, and the features around it. I do have some things to share about how it is now and then what some of the near future features that they’re adding in as well. Let’s go ahead and dive in. Okay. The two things I’m really purposed to share with you today are how we can embed a Tableau dashboard in a Salesforce environment.
If you’re coming at it from a Salesforce point of view, you’re using that to manage your customer relationships, I feel like there is just a quick win on being able to bring in Tableau Next and tightly integrate it into the environment. I think that that’s some of the intentionality that is in the design of the whole thing. If you’re using Salesforce, that might be a good way to really get started with it. Another part of it that has had a lot of marketing around it, a lot of conversations around it is Concierge. That’s that agentic analytics agent that’s going to do the conversational piece with us at some point. My testing has been around it a lot because I wanted to reconcile the data, validate the answers, and then figure out where we go from how I ask good questions and how I get good answers.
I was an early adopter with chat GPT. So I’m very much aware, and in my teaching, I’m always teaching students, especially my college students, to get involved with AI and do well with it because being a great prompt engineer is a big part of it. If you don’t ask the right questions, you get unwanted answers. Then trying to figure out what part of this is at play here with what’s happening inside of the Tableau ecosystem. With that, let’s go ahead and dive in. I went ahead and recorded a no words walkthrough so I could articulate on top of it what it is that was happening here, but I didn’t want to be distracted by any of the little things that might happen along the way.
The first part of this is about nine minutes long and we’re going to walk through how you create a Tableau dashboard inside of Tableau Next, and then how we connect it up so that it’s integrated over in Salesforce. This is the dashboard that you’ll see be built– the viz part is what you’re going to see actually be built. To be fair, they have several templated dashboard structures, so building dashboards here are pretty straightforward and very familiar in terms of the interface and that, and we probably will do another session some other time or we’ll be making some videos to make it clear how to assemble things because it is similar but different. Okay.
The idea here is when you have the account tab open in Salesforce, you would have potentially other things, but you would at least have the related and the details tab in this space, and you’ll be able to add an analytics tab where you have a dashboard that you’ve created that highlights things you care about. That’s the goal. It all begins with a semantic model. We’ve not seen it before. This is what it looks like. This is the sales cloud, this is a demo data set that Salesforce has made available around the whole concept of customer management and opportunities and pipeline and all that. You can see that we have–
It’s an interactive environment and you’ll see that in just a moment as well. In this particular example, I had uploaded an Excel sheet that had customer reviews that had sentiments of positive and negative and neutral. It was being added into this environment, and this is sort of like our data source tab on Tableau Desktop, but Cloud side, and then also with more, they would call this the semantic model but we also are under the heading of a workspace. That’s the hierarchy of words, use of terms is workspace and semantic model and then we have, of course, these other divisions as well. We have metrics and calculated fields that are separated out here nicely for us and parameters. Logical viz would be this piece right here and it’s always illuminated in orange.
Then we have our data objects. The semantic data model is made up of these data objects and they call them data lake objects, otherwise I would not have characterized the review worksheet as a data lake object, but the DLO label is what gets assigned to it. Regardless of what it is, it sits together nicely with all this when you get to decide the relationships how they are built. You also can turn on an Einstein feature here that will suggest relationships, and most of the times it’s right. I’ve not seen it be wrong yet. That’s also a possibility. I’m going to kick off this recording and then I’ll be speaking over it so that we can be seeing and– Here we go.
If anybody is having any trouble seeing it or hearing me, please let me know in chat or let Lauren know and she’ll interrupt me, get me going. I’m just walking through some of the ad data pieces here, and I just mentioned that the customer reviews was something I had uploaded, just an Excel sheet that I’d uploaded. You want to take those data lake objects and add them into your data model. You can see that there was other Excel sheets or other data files that were available there. You can see the spread of the content of it here in the panel on the left, like the data pane. You also can see it tabularly presented across the bottom.
I’m just scrolling and showing you, we have very– you see this similar look and feel to Tableau Desktop, right? In addition to some of the things we have in desktop, we also can sort the columns independently– not independent of each other but choose a column and sort the whole table right there in this view, which I just did with the sentiment. All right, and then I can close it with the panel to the right. We’re back to our semantic model. Now we’re going to just look at– We’re going to need to build a calculated field that is going to be bridging the logic between those, the sentiment and the count of how many right–
This drafting with Einstein is something that they are strongly recommending, and they’ve taken it out of the consumption credit, so that’s really awesome. That helps you with some of the syntax because the syntax is a little different here than what we experienced over in Tableau classic and in Cloud. You can see my new field bottom left. Now we’re going to go and build our viz. I can choose to borrow from something that exists or just start with my semantic model, my data source, which is of course what I’m going to do here because I want to do something broader than some other viz that already existed in my space.
Looks like Tableau, right? Desktop and Cloud, right? We got that data pane on the left, and we’re going to be searching for fields that we’re going to be pulling to make that visualization that I showed you earlier. My horizontal axis is going to be fiscal weeks, but look how these filters are– the ways to filter dates look similar. Similar, little bit different, but some parts of this are very much the same. All right, I’ve got a filter to close it out at the end of that Q1 time period. Now I’m going to be putting things on rows and columns and also [unintelligible 00:14:09] color. I’m going to put those review date up first to create that horizontal axis of the fiscal weeks.
You recognize our drop down is similar, where we have control over the date part. It’s interesting that they went ahead and used the word date part. Date part is one of my favorite SQL commands, but you still can switch it from dimensions and measures, and how you present it, you can visualize it as discrete or continuous as we do often in tables. Then here’s my discrete presentation of fiscal week. All right. I’m also going to grab those total number of reviews to create my pattern of sentiment, and then I’m going to divide it up by the sentiment. Also, each mark will have its own connection to the account name that the sentiment came from. We’re doing both of those here.
Here comes sentiment. That splits it out into three different sections. Now I’m going to add that account name. On to detail and gets a lot going on. I’m a shift to the circle presentation of the mark. Then bring my sentiment on to color. Even though we are absolutely abundantly spoiled in the Tableau house with the colors, what’s happening underneath this color tile has been improving every time they do a release. Just like in other spaces, we can change the access labels. A lot of what you find here in terms of control over presentation is right underneath the pill. We’ve had other menus do that with Tableau, but more and more you may have noticed with Cloud especially, the controls you need are underneath the pill that you’re talking about. Very much so here with Tableau Next.
I’m changing my neutral to that in between gray tone, and then I’m going to make negative a little bit more of an alert color. When we first started the beta, we just had a few colors. So this is hugely improved, and also being able to get it where you’re pulling it off the color spectrum. We have account ID, account name on the label, so we can have that information. Now you can see as a hover, no control– I mean, limited control of the tooltip so far, but it’s there. It’s [unintelligible 00:17:04] provides information. Did you see that they turn on the button at the top and it opens up the Salesforce action panel to the right?
We had a Salesforce action and we also had a navigation action as the two choices there. I’m just scrolling through there to show that there’s a lot of different types of Salesforce actions you can choose. For this one, I chose to open the record, and it would be open the record on select, not hover. This is where we would open the record right here. This is the goal, but I didn’t show you how to connect that yet. So in the next little section here, you see how now we just have related in detail. Now I’m going to show you how to do the integration. You go to your setup menu, edit the page, and then we are going to wait for it to load. Okay [crosstalk]
>> LAUREN: Oh, and, Celia– I’m sorry.
>> CELIA: Yes, Lauren.
>> LAUREN: We have one question from Christian Anderson asking about if we have Tableau Next licenses, should we be able to access this in our Salesforce environment today?
>> CELIA: If you have Next and you have Salesforce, yes.
>> LAUREN: Awesome.
>> CELIA: The custom label I’m creating to make it say analytics, I’m changing the order of it just by sliding it around. Here’s where you add the component. Lots that you could add right there. Then I’ve chosen Tableau Next dashboard. You may quickly see that it’s not necessarily the one that was opportunity one, but when I click into this search feature over to the right, it shows me all the ones that are open and available in this environment. Then I’m going to change the height right there to give it more space because I know that my dashboard is a little more tall.
There are additional configurations to make sure that I’ve got the right filtering happening between them that are going to come up next, but you can see that already this is going to be– to land with your account open and your visualization up top would be a win. Here’s where I’m making sure that the semantic model from the Tableau Next side is connecting correctly to this Salesforce key part. Making sure it’s talking to account name and actually matching the record value, not in these match exactly. That’s what we’re doing right here. Make sure it equals it. [silence]
All right. You see the dots resolved to just the ones that are relevant to global productions. Then we save. Two steps left, save and activate. Now, the activate is going to be different on my video than what it was the first time I went through, but I’ll just try and reproduce it as much as possible. There’s three ways to activate stuff. Org wide, app default. A lot of this has to do with what sort of form factor your viz or vizes have been set up for. It’s asking me to remove [unintelligible 00:20:22] org default because that’s what I had it set prior to this moment. In any case, it would give you a little bit different confirmation management. It’s your first time through.
If you change things, it’s going to ask you about forms and formats of different apps and what they’re going to go toward. Then let’s see, I’m going to close out of here and show you the accounts I can switch between and see the relevant data shown for all these particular accounts that are here. That would be like one of the first ways if I was help– This would be one of the first ways to really get it going. I think that let’s take that question for Blake in a minute because he’s– If you mean creator license and site admin, most likely, yes, but I definitely can find out for you. Okay.
>> LAUREN: Yes, I just sent him a message. So we’re both–
>> CELIA: Okay. [unintelligible 00:21:29]
>> LAUREN: I was trying to save that one.
>> CELIA: We’re going to need to look in that a little bit more. Very much with the beta [unintelligible 00:21:38] we’ll focus more on the how to part, a little bit less on the, exactly what licenses I need because the beta they give you what you need, right? We did also need to have Einstein enabled. There’s some enabling that has to happen for– Frank, great question. Very intentionally, they’re setting it to be interoperability. If you watch any presentations of any exact Salesforce at Dreamforce that will come up this winter, what’s happening at DataFam in London beginning of December, interoperability is a big focus and commitment from their side.
The exact methods are going to need to be flushed out as we do some more implementations. That is the part that I wanted to show you for sure on the Salesforce side, and I do have it all loaded up in the background. [unintelligible 00:22:41] when we get to Q&A to hop around and show you some more over there if you’d like. I’d like to spend a minute and talk to you for a minute about Concierge. It seems like from a Tableau side of the house, that’s a real popular topic to talk about, and certainly has been demoed a lot starting at TC 25, at least, if not before, with lots of potential there.
My own observation about my own self, my testing methods and questions that I come across. I’ve noticed that its ability to answer complexity in my questions is improving and there’s a definite progression there. Then also there being a need for me to step up with increased completeness for descriptions and business rules. I’ll tell you a little bit more what I mean by that. Just to set the stage for it, though, the method I have used, pardon me, Superstore for lots of demo things. When I need to restate something in a public way without revealing any private data. Lots of us will recast things with Superstore. I use Superstore for this. Then I also was trying to progressively build complexity into the questions I was asking.
First of all, I would ask about top customers with successful sales, and then I wanted to have return sales. If you can just do a add reaction on your icon on Zoom chat, if you’ve ever used Superstore, I’d like to know if there’s a little bit of us, a whole lot of us. You can run the Zoom window across the bottom. You should be able to add a reaction. Some. All right. One of the things about Superstore is that we have three tables that are the core information. It’s e-commerce, online, office supply. We have the orders table that is essentially our sales table. It records all the transactions.
The way they decided to structure this, not trying to defend it at all, just saying how it is. They have a regions table that tells you who the regional manager is and the region name. They also have had historically over time a returns table, which in some educational circumstances, I have changed. That return table just has a order number and the word yes. If you do a left join with it, you end up with the ones that were sent back. One of the artificial structures about it is there’s the construct of if the order number appears over here in the returns table, then all of them went back. Regardless of all the logical flaws there, that is the structure of what that table is on.
That represented an opportunity for me here because that’s really pretty odd. Except for that lots of data is separated into smaller tables. Salesforce has lots of small tables. Testing that relationship and being able to fish out the net sales was where I was going with this. I first started talking to the agent and building a complimentary set of things that should support the conversation about successful sales and the return sales, and what was the value in the amount. Then moved into net sales. Gross minus the return. That wasn’t possible in the midsummer, to be fair. Then it started being more possible mid-August. The next question is some things that represent those compound questions.
I wanted to see top customer within each region, what their successful sales or what their most– who are our frequent flyers on returning stuff for a lot of money. That’s the kind of business question I’d be wanting to ask, but let me pause on that. This is my line of thinking, what I’ve tested quite a bit against, believe it or not. Let me talk to you a little bit about what I’ve learned from them recently about the upcoming features and this beta that I’m going to be participating in starting in five days. They’re looking to do an expansion of the semantic features. They’re busy doing lots. They wanted us to think of the agent as something like a new hire.
You have somebody come into your company, is on an internship or whatever. Even if they have some fundamental domain expertise, they’re not going to know the way that we do business and the business lingo, the business rules, this is the way they’ve been trying to communicate what needs to happen on bringing the documentation up to help support the agentic conversation. This was a slide that was also on a presentation that I saw recently about grounding agents being a new responsibility. Well, for better or worse, what they’re pointing to is a lot about documentation that really is just classic data governance.
For those of us who’ve done data governance or who have had somebody in our ecosystem that is documenting the business rules, the way we do business, the words, what is the acronym and what is the words behind that? What’s the calculation behind that? Anybody who’s spent the time and effort to have that documented inside or outside of your data environment, this will be no big deal for, but the reality of it is, for the agent to read the data’s tables without any kind of descriptions or accurate labeling, it’s really difficult and not reasonable.
Onto that, this part that I have highlighted in yellow dashes is the instructions that they’re really hoping that will– they’re all going to bring new tools to support and they want to highlight to our attention that this needs to happen. They’ve done a good job of a starting point of giving us spots to do it as well. The instructions for business context, giving the data model labels, the short descriptions when it’s helpful, and then some data, calculated fields or metrics that go alongside of that. That’s going to empower the rest of it. For it to be able to execute, this is part of the foundation.
This is what the screen looks like for the business preferences. This is something that’s coming soon, and it’s in beta this fall. That is applying some sort of guardrails during creation. I’ll show you what it looks like right now. Business preferences is a way to feed in those nuanced vocabulary kind of things. In fact, they would like for us to think about it from three points of views. Something that is about interpretation, those acronyms, the way that a person who’s not a data person might ask the agent to present information about something. Every business has its nuances and how it uses its language.
Then you also can specify things about output and some of the defaults that would be in play there, as well as being able to add in some comments about formatting that you want to have, because this part about when we put KPIs together, we need to not have it have so many digits that it doesn’t fit on the space. Things like that. If you’re fortunate enough to be in that kind of space with a big business. Along with these business rules are some maximums about how many rules you can have and how many characters you can take. That’s another reason I think that they are also improving some of the tool suite to support the effort from our side.
In the midst of working with one of the engineers that was involved in the business preferences coding, this became apparent and I shared this at our internal lunch and learn not too long ago. There is a tendency to take things that really belong in calculated fields and put them into business rules. That really kind of defeats– That’s not the purpose. That’s not where the agent can see it best. If you have something that is a calculation and needs to be in a calculated field or a metric, and both of those devices, when you create them, do have spaces for documentation to happen.
I will gladly share this deck with you guys so that you have these as well. Some of the optimization and new features that are coming out of [unintelligible 00:31:36] in short order are going to be regarding things that are model readiness. Right now we have two spots on the semantic model page where one is for business rules, one is for model readiness. Right now it’s more of a checklist and they’re going to turn it into more of an interactive conversation with the users. That’s coming. Right now we also do have, though, the ability to test the model.
This is a screenshot from literally something I pulled up and did last night so that I can– If you have especially calculated values and you know roughly what it should be and you have the ability to validate that going into this can be an important part to do. The return status was a logical Boolean field I put together, and then I wanted it to match up against sales and then give me this output to test it. This is one of the things that is now and then is just going to get better as we go. Descriptions of general semantic fields is a big part of– Just about every place within the ecosystem, there’s a place for description.
The more you fill in, the more the agent does well. It just does, especially if your data labels aren’t informative on their own, but they’re going to do– I noticed over in the Salesforce data Cloud side, there are several things about data governance and about sufficiency and completeness of things being fully filled out. This is now coming to our side as well. Being able to identify what’s missing content and descriptions, and have some support along the lines. They’re going to attempt to do some description generation and see how that goes. If things are named well or if they’re coming out of Salesforce where the naming is more structured, there’s a good chance that that could be supportive there. Even if it just did a baseline and you had to adjust it, that would be a lot less work than starting from scratch, I imagine.
Any place where they have Einstein helping with the interpretation of things, as you’ll see, this is one of their signature things right now on helping you with the formulas and fill-ins. The Concierge testing center will also be new after probably Q1, but it’s in beta this fall and half 2. That’s where we’re going to have more of a place where we test things and validate and give it feedback, and come up with accuracy scores. This is one of their slides for this. Again, back to that conversation with the agent, as opposed to it being– this is a much more complete kind of conversation to preview the answers and validate them before it gets released to your users who would be making decisions on the basis of them.
I’ll give you a little preview of some of the results that I got from my testing that I did. I did get through these. Anything that was overarching for my whole data set, I had good results with. Gross sales, returned sales, even net sales, I was surprised, pleasantly surprised, but I did go ahead and make a couple of calculated fields to ease the path on that. It wasn’t like it was having to make that up. I also was very careful with my prompt engineering to be very clear about the relationship between the returns table and the orders table and how the presence or the absence of an order number in the returns table, that meaning of that I expressed in my prompts that I was saying.
When I got around to trying to ask for the top one per region, the most offensive returner or the highest flyer on my sales, I had a little bit more difficulty getting it to do that. To be fair, on the SQL point of view, that would be a window function, most likely something like that. The little green list on the right was the things I was moving through to try to give it more support on getting that done. It had mixed results. I got an answer at one point that I thought, “Oh, gosh, that’s how I asked the question,” because this one– if I asked for a certain region, I could get it, right? That was easy and clear. This is my Tableau Next dashboard that I was using. I was asking questions on the agent panel over here, and I was validating my responses over here on the side.
Then the next one was the one that I was thinking was kind of humorous, because I was asking, and you can see it in the questions, can you show me the number one customer for the most net sales for each of the four regions? Well, there’s some nuance in how the words are put together there. I meant group by each region. What it did was it identified that Sean Miller was the top across everything, and here’s the contributions that he made to each of the regions. While this answer is not wrong, it’s not what I intended, but there’s a nuance in how we were using the words that made this not quite it.
If I named a region, I got the answer. It was interesting how that when I said some of the top contributors, that it instantly went to the top number of orders, not the top money related orders, but the top quantity. It did tell me that Ken Black was my top returner. It gave me a cute bar chart. Not about the money, but about the volume. That was on me because I needed to clarify that. What it did when it couldn’t calculate, it was twofold in the current instance that I’m working with. One is it would offer me requirements. If it told me it couldn’t calculate it, then I would say, “What else would you need to get this done?” Then it would give me a list of things to do.
I thought that was pretty good. That’s a good place to start. That’s come a long ways from what used to happen. That can help you think through how you’re asking the question. It can also help you create a calculated field or a metric that would support the process. Then also, occasionally I would get an error message that would say failed to execute the semantic query. Then I was like, I’ve got to understand what this is. I asked it specifically what it was. I also researched it a little bit online as well. These were the answers that seem like fair game on the kind of answers that it could give me back. Some data limitations, the complexity of the query, and back to use of language. The prompt engineering hat needs to be executed well in terms of use of nuances of how we use the languages that we’re using.
Then ambiguous questions is not going to work. I think everybody who’s using a generative AI, any tool in your world will understand that that’s going to be a problem. I did ask ChatGPT because in my house we are big ChatGPT [unintelligible 00:38:57] This was its recommendations on being specific and explicit. To be fair, I think that the answer straight out of the analyst agent was sufficient to prompt my thinking on a lot of it. We hope to have an update for you guys in Q1 of 2026 as we do more with the agent. The vizes are beautiful and you can tell that–
Here’s the workspace for this data set that I put in. You can see how it’s divided out. Let’s go back in. All right. There comes my workspace again. Nuances and new icons, they go with these things. Here’s my data model for a superstore. We already took a look at that. The pop ups for this is great. Having the calculated fields on the side here is awesome. I had built two dashboards, one with and one without regions, because once I was trying to make it tell me about the per region, I ended up adding a bunch of stuff onto my worksheet to just like, “Maybe I need to just assist you.” I ended up adding my list over here on the right hand side. I went ahead and put explicitly in here the segmentation of region.
Then I also realized that another complexity of this data set is that customers buy in different regions. It’s not a clean break there either. Then I also had my calculated field for this. I went ahead and counted up how many returns, and I was trying to ease it along the way. This is what our visualization looked like without the other stuff going on. You can do the– easy to do the total number as a dimension and then also have my bar chart with– earlier in the year, May, June, when I used a Boolean set up for orders returned on the color, it did not see that at all. Whereas now it actually wants to have– It will have conversations with you about that feature and instantly be correct.
This is the analytics agent right here. Einstein does have to be enabled for this to work. The questions could come down here. You’ll notice that when I’m traveling through the different devices, this agent does not come up if I’m in an individual worksheet. It needs to be a member of a dashboard or I need to be on top of the model. Then I can have conversations over here as well. This is where the business rules are. You can see I have a few in there right now. Then this is where the checklist for agent readiness is. There’s some here, but this is one of the things they are really adding more features to. So it will be more feature rich going on in short order.
All right. I think that I saw Blake come in the room. I want to be sure and give some time for you guys to ask some questions. Blake has just been involved in a project in-house for a customer with production versions of all this. That’s why I wanted to give a chance for him to answer and be asked some questions as well. Any questions. [silence] Let me put it back to you guys. Are you guys likely to try Next? Who’s interested in trying next? This is a technical person to a technical asking a tech question here. Who’s curious about it? [silence]
Okay. Frank’s asking about the basic credits for agent. There’s a couple of things I’d point out about that. You’re getting it next week. Very good. Okay. Couple of things about that. I have heard that there’s some changing of strategies on exactly how the consumption credits are going to work. There’s a digital wallet that’s just been released. I strongly recommend that you implement. It’s supposed to give you real-time calculations on what the consumption credits are that are being used. Yes, I think that there’s been– What I have heard and read is that there is a bucket that is allowed or given or credited up front. Then also there has been a change in some of the way that they are going to calculate it.
The thing that I was the most concerned about was how metrics were going to be calculated, if it had any level of complexity behind it and where it could be– The math would be a lot. My understanding is anything involved in Einstein, they have restructured how the charging is going to go. That’s as far as I know. That would be a better question for another person. That has been reconsidered. Anybody else other than Frank getting it anytime soon or curious about it? [silence] Okay.
I have just a couple more minutes left. I want to just ask you, what else would you like to see or know about Next or its integration with Cloud or Salesforce? Feel free to type it in the channel. I just would like to take some notes for whatever else you guys might like to see. Yes, Scott, I hear you. I did hear and see that there is a specific purchase order number where Next was bundled. If you have a Cloud and such, you might check. The connecting them to– Oh, really good points. Let me just show something here that– Okay, let me log back in to the other part. Zach is making a point about zero copy in use and how important that is. Let me go back to my– I want to show you on my sales cloud and this part of this demo part, I’m going to–
>> [PAUSE 00:45:56]
>> CELIA: There we go. My sales cloud workspace, you’ll see that– That’s not quite what I was looking for. Let’s see if I can go there still. Let me just make the point that he’s mentioning about zero copy. Zero copy is really important. What they’re trying to [unintelligible 00:46:36] more and more is that we leave things in place in one spot and we reference them. Zero copy is pretty well documented also on the Tableau Salesforce website. I would encourage you to dive into that a little bit. It keeps you from having extra charges and it makes it so that you can reference the data.
I can’t speak to performance issues because I haven’t had a chance to do that in a way that’s meaningful, but I just know that when I built these things for the demo about integration with Salesforce, I did not have all the data set and have a copy of it. I was able to reference it. You can see there’s a little arrow on my semantic model referencing the core of that. Any other tables I needed to add to it, like schema builder, kind of a little bit of overlap and how that works. Zero copy is important. It’s important. It’s also valuable. I’m sure all of us will have performance questions about that. Yes, zero copy.
Maybe we’ll be looking at that as another topic for discussion. Any other questions, anything what you guys want to ask Blake while I’ve got him here fresh out of the field from doing the work with–
>> CHRISTIAN ANDERSON: I’ve got a quick question.
>> CELIA: Thanks, Christian.
>> CHRISTIAN: I know we have Tableau Next, and I’m looking for it in my org right now. It looks like there’s permission sets that probably need to be assigned. Is that all that’s needed? I don’t have Tableau as an option in my apps menu, but I’m wondering as far as setup so that I can get to the Tableau screens that you were just showing.
>> CELIA: It just have to be enabled. There’s like two or three things that have to be enabled, and I have heard that there are some steps involved there that are not necessarily super well documented. Blake, can you speak to that a little bit? I know that we had another team member who also did that.
>> BLAKE WADE: Yes, as far as the data Cloud and specific permission sets on the Salesforce side, that was handled by another team member, but I agree, Christian, if you’re not in the Tableau Next specific permission set, you won’t see it in your app launcher on Salesforce. Again, that doesn’t mean you don’t have it. It just means you’re not in the correct permission set.
>> CHRISTIAN: Yes, it looks like I can assign myself those permission sets. I’m an administrator in the org here, but I just wasn’t sure if there was other setup steps, and sometimes there’s other things needed, obviously, the license, but maybe it’s as simple as just assigning the permission sets. I can do some reading too and try to find [crosstalk]
>> BLAKE: I wish I could answer definitively. I know there are more setup steps, but I wasn’t the one to have to do them.
>> CHRISTIAN: No problem.
>> CELIA: Christian, maybe we can try to follow up with you with any resources we can send your way on helps or guidelines on that. I’ll make a note.
>> CHRISTIAN: Okay. Well, thank you.
>> CELIA: All right. Anybody else? Any other questions? Was this helpful? Are you guys going to give me a thumbs up if this was in the zip code of being interesting? Okay. Oh, good. Okay, good, good. All right. It’s a work in progress and these guys are really pushing so hard to make it very feature rich. If you get frustrated with it, just remember where Tableau Cloud was in 2018. Thank you very much, everybody. Blake, thanks for being here. Lauren, thank you for cutting out to be here as well. I appreciate you all. We’ll be back with more in Q1 about it, no doubt. I’m excited with what they’re going to do with the semantic model. I feel like it’s a game changer, and the portability of it is going to be incredible. All right. Thank you for being here, team. Thank you so much.
