From prompt to platform: connecting an AI agent to Tableau through MCP

0:00 Chris Williams: Good afternoon, everybody. Welcome to the August edition of the Portland Tableau User Group.

0:08 Uh, my name is Chris Williams. I'm the, the owner of the group, and it's good to talk to you again. We have everything figured out from yesterday, so there should be no confusion or no shenanigans with any laptops. We have Kanwal here today, but before we get to Kanwal, I'm gonna go ahead and start running through sort of what's going on with our, you know, with everything Tableau.

0:30 So, uh, sometimes… With this is what we're gonna talk about. We're gonna talk about, you know, give, you know, a little bit about me, but you already know this pretty much. Um, who's hiring? If anybody is hiring or looking for, um, you know, a job, please post it in the chat, um, and then we can go ahead. If anybody's looking for anything, that, that's one thing I like to, to pretty much put out there.

0:56 I'm gonna go over some of the events, some of which are coming up, some of which are, uh, already happened, and then give you some co- uh, some product updates. Then we'll go right into, uh, our guest speaker, Kanwal, who will talk to us about Tableau NCP and AI governance. Uh, and then the next meeting t- will be at least on October 1st. I may do one in September, but I'm also gonna be at Dreamforce, so I may do something shortly after Dreamforce, so we have some more meetings back-to-back.

1:29 Uh, I kind of want to put something out there just because I don't want to go through all of September not doing something, but since… We'll figure that out. But there's definitely gonna be one on the 1st of October.

1:43 So if your company is hiring, please go ahead, unmute, let us know, or post it in Slack, or you can, you know, use our, uh, Portland, uh, area to, um, talk about it. So let's figure out what, you know, if you're needing something, or if you need us, or if you need the job, or you want to post, you know, post a job, please let us know. If you don't want to do it in this setting, please feel free to drop me a line at, uh…

2:08 And I'll put my, uh, address in the chat if you just want to do it more of a one-on-one situation. Okay?

2:15 So let's get into the events that are going on. This is one thing, uh, that's pretty cool. We have an AI analytics virtual workshop that Kyle Massey's leading. Uh, I know him, you know, pretty well. This is gonna be a fantastic, uh, effort. We are… Be able to go ahead and connect Tableau to Slack. Uh, so if you would like to register, please take a look at the lower right-hand corner.

2:39 You'll see the QR code. Go ahead and scan that guy in, and we'll be able to figure out how you can get registered and go from there. There's also a URL at the very bottom of the screen right here on the left-hand side, right under Kyle's name, that you can also register as well. Okay?

2:58 Dreamforce is upon us. It's mid-September. I will personally be there. I'll be talking a- about Tableau and trying to educate the masses at Dreamforce about the strengths of Tableau. It's gonna be in San Francisco, and you can catch it online. Um, you can also go there in person. So September 15th to 17th, it's going to be up in, uh, basically the Moscone Center in, uh, San Francisco.

3:26 Um, if you would like to go and you're worried about how much it's gonna cost, uh, please let me know. I'd be able to get you a pretty significant discount. Um, but if you really want to understand a little bit more about it, go ahead and scan this QR code, and that should take you to the registration. But, um, but if you're really interested in going and really are kind of curious about that, please let me know and I can probably get you a much more significant discount than what that's probably going to offer.

3:53 Okay? It's pretty big happening. We had, um, close to sixty-eight thousand people last year in San Francisco. It was pretty amazing.

4:02 Um, you… I know we focus primarily on Tableau in this, in this user group, but they are starting to talk more about Tableau within the Salesforce and, uh, you know, clientele, which is great because, uh, all we want to have is some, you know, be included in the conversation. Uh, and a lot of people don't realize how strong Tableau is. I know Salesforce does, otherwise they wouldn't have purchased it, but not just Tableau Next, but Tableau as well.

4:33 So it's been a minute, but Tableau Conference was pretty cool.

4:37 We had devs on stage. We had these three being able to connect and compose, being able to, uh, mobilize and unlock, and navigate and chart. Uh, these are some of the new things that were talked about on stage. And so as those get rolled out, uh, I'll keep you posted in this, in this venue when these guys are gonna get rolled out into our products, and it's gonna be pretty neat.

4:59 And what this encompasses are what you see below. If you are interested in understanding what this is, I mean, I can give you obviously a PDF of these, uh, PowerPoints that we put together.

5:13 Uh, but you can take a look at some of what's included in all of these, uh, the three that I mentioned here. So you talk about Tableau mult- you know, Tableau data sources on public, bring your own connector to the cloud, REST APIs. You have accessible viz, uh, authoring. I mean, you have, uh, MCP, which we're about to talk about, Tableau Agent, Tableau Next.

5:37 Um, uh, we've talked about Tableau Next before, if you ever want to know more about it. Um, pretty much I do that pretty much every day now of, of my, uh, professional life. So, um, please feel free to, you know, drop in a line what you might want to talk about, and we can, we can make it happen within there. What I'm gonna try and do is try to in- increase the cadence.

6:01 It's been hard because I've been traveling, so my, my apologies for the lateness of this one. But, um, I'm gonna try to get back into the monthly cadence of doing at least something with us.

6:13 So some of the feature ideas that were, were, were talked about were Tableau Solve, Radial Layout, and Project Bongo, which is interesting. So let me talk a little bit more. I'll, I'll… As we go through there, and then we had Iron Viz. Ann Pregler won the whole Iron Viz. What about– It was great. It was probably one of the best Iron Viz statements I've, um, contests I've ever seen, and I've been going to Tableau conferences probably for the last six years.

6:43 And, you know, all of… Kevin Wead did a great one, Brian Moore did a great one. And so you can actually watch the playback for free if you wanted to see it. But these are the actual, you know, vizs. I think some of them may be actually up online too. But you can get, download these and take a look at them and understand what the premise of each of them were.

7:07 And they actually did some pretty neat, uh, techniques that, um, it's pretty fun to be able to look at and say, "Oh, that's how they did this." It could give you some ideas. I know we're kind of like stuck in our dashboards and what people, you know, our clients want us to do, but sometimes when you can have the flexibility to have a little bit of more creativity, these are the kinds of fun things that you can come up with with Tableau.

7:39 So within the community, we still are celebrating fifty thousand Tableau group, user group members. This is fantastic.

7:45 Uh, I encourage you also to take a look at other Tableau user groups, not just cities, but everything. They have certain verticals. Um, there are about three, uh, three verticals that I deal with. I, I also, I deal with the Portland, obviously, Tableau user group. I deal with Black In Data, and I'm also one of the user group mentors for some of the other ones.

8:07 Uh, and there's some new ones coming down the pike. So I would encourage you to go ahead and log on to usergroups.Tableau.com to see if there's something else that you wanna be, uh, a particip-participate in, and maybe get some more inc-information from there too.

8:22 Uh, Slack is really important. I'm actually trying to get us more visibility as user groups on Slack, and maybe be able to communicate on Slack rather than sending just emails. If that's something that you're interested in, uh, I'm really trying hard to have us be available on Slack. Uh, doesn't mean you're gonna get bombarded with messages all day, but it may be a situation where if you have something, uh, we can…

8:48 that you wanna talk about, or you wanted to potentially have it as a new topic for a user group, being able to communicate on Slack would be great. So go ahead and scan this. If you're not on Slack, you're missing out a little bit. Go ahead and join. Scan this UR code that, QR code that you see over here on the lower left-hand corner.

9:08 So we did some product updates and resources. One of the big things that they're just doing now, and they're almost done with this exam, is getting certified in Tableau Next. Tableau Next is the Tableau instance that lives on Data Cloud within Salesforce. Uh, as I mentioned before, I do a lot of this. If you're curious about what that means and what that looks like, uh, I would either start looking at it on your Trailblazer, like your, your Trailheads and everything.

9:36 Try to figure out, you know, what you wanna do there, and try to see if it actually fits in with what you wanna do as a professional and what your, your, your, you know, supervisors are asking you to do as far as delivering dashboards in the Salesforce environment.

9:53 Uh, so lives on Data 360, Data Cloud. You build a semantic model, then that semantic model can actually be reused on your Tableau Cloud side with the, the semantic connector.

10:06 So, um, if you wanna, if you're interested, this is going to be… I, I believe it's offered now. If it's not, it's gonna be this week. Uh, but if you wanna learn more and understand what it takes, uh, go ahead and scan this QR code over here on the right-hand side, and we can carry on, and you can get more information.

10:25 AI is taking over the world it seems like, but you also have a lot of different trails for AI. So building a strong AI foundation is really important. You wanna be able to, uh, use, have AI as part of something that's in your toolbox. So this will help you do that in a safe environment, meaning that, you know, you can learn at your own pace. It talks about data literacy, data modeling, and reports, and, and how that fits in with your reports and dashboards.

10:51 Uh, Tableau Next uses a lot of AI as far as, um, concierges and being able to un– uh, yeah, integrate Agentforce into the semantic model.

10:59 So if you're interested in that, go ahead and take a look at this. Um, sorry for the type down here. I'll make sure I get that fixed next time around.

11:08 And then more trails, uh, for Tableau. This is more Tableau classic, where it talks about Tableau Prep and understanding how you can explore insights and monitor KPIs, key performance indicators, that way.

11:23 Data storytelling and visualization is critical because we can have all this data, but if we don't tell the story to the right group of people, no one's going to use it.

11:33 This will help give you some tactics to be able to go ahead and tell your story, in addition to whatever you already know.

11:40 So, uh, go ahead and take a look at that and see what that takes. And then all, you all, when you actually go into the trails, it'll tell you how long estimate it'll take to go execute these trails, and you'll get badges and all that stuff. So it's just another way to, to do some upskilling.

11:59 And without further ado, I wanna introduce one of my, uh, former colleagues and good friends. Uh, we did a presentation like this at, uh, past Tableau Conference this year. Um, I'm excited to have her present here for you all today.

12:15 Uh, her name is Kanwal Kour, and, uh, she is a solutions architect for Concentrix. And, um- And yeah, like I said, I've worked with her for almost, uh, 11 or 12 years. So I'm really honored to have her here today. So I'm gonna stop with this, and I'm gonna share…

12:32 allow Kanwal to pick it up and share her screen. So without further ado, Kanwal, the floor is yours.

12:40 Kanwal Kour: Thanks, Chris, and thank you f- everybody for joining us back.

12:45 So we are talking about from prompt to platform. So I don't know if any one of you has used AI agents as yet.

12:54 Um, but we are talking about that, um, uh, specifically Claude connecting to, um, Tableau MCP.

13:04 So this would be our agenda today. Um, just, uh, so we have actually the data, synthetic data and a scenario which you, um, you all will be able to, uh, connect to and then understand, um, what, what does this bring, um, to the organizations.

13:27 So the Monday morning question, um, under the hood, which we'll talk about how it all works.

13:34 Uh, the demo, we'll, we'll dig deep into, um, Claude and we can… how we can actually ask the questions, the prompts and all that and get the answers.

13:45 And then we'll go to Q&A.

13:49 So the Monday morning question which, um, I think all the VP of Sales usually get, and then the question is like about, you know, how the sales is doing and, um, tho- those kind of questions. So a scenario which we have built is like of, say, Sarah is the VP, and she's been asked by the CEO, um, why the enterprise annuals slipped in Q4 2025. Now we are going to Q4 2025 because that was the full year.

14:16 We are still in August in 2026 too. But specifically we'll focus on Q4 and then go to Q1 and Q2 for this year.

14:25 So the board call starts in an hour, say, and then Sarah has to get the answers to her CEO within this 73 minutes.

14:34 She goes to the dashboard. The dashboard is there.

14:37 Um, she sees the dip, but then, then there's no way she can know why.

14:44 But what does she do?

14:46 She Slacks her analytical team, and the team actually goes to the Tableau server or cloud, whatever they're using. They open the dashboards, um, they get the, uh, screenshots, send them to Sarah. Sarah has to compile them, and then she enters the meeting, and then she's already late.

15:05 So, um, she's not there in time. She doesn't…

15:09 And the new way with the, with AI agents and the Tableau MCP connectivity, um, it takes only 90 seconds to two minutes. And Sarah, what Sarah can do now is like Sarah doesn't need to go to anyone. She can just, um, ask the questions in her AI agent, um, to the agent, AI agent, like Claude, and in a plain, um, language, and she'll get the answers right from the AI agent who is in turn connecting to Tableau MCP.

15:40 And Tableau MCP is actually able to fetch the data from everything which is published on Tableau server or cloud. Okay. So,

15:53 let's see how it all works, right? We'll go under the hood and see the exciting part of how it all works.

16:02 So there are three pillars of Tableau MCP, right? So you have MCP server, which is actually hosted on Tableau. Your Tableau instance, um, we'll talk about cloud because we are using cloud in our organization, so I'll just say cloud.

16:16 Um, you can have server as well, Tableau Server. So let's just stick to cloud. So the MCP server is on Tableau Cloud. You have your personal access tokens, which is PAT.

16:27 Um, they are the scoped authentication. They work like anything like you give privileges or, or, um, user access to your data sources or workbooks, whatever you have actually configured. That's what it is. So users who have, um, what access they have on your Tableau Cloud instance, they'll have the same access through PAT using the AI agents and the semantic layer.

16:49 The semantic layer is anything which you have published on the Tableau Cloud, the data sources, um, the certified metrics, any calculated fields, the definitions, everything. Cons- They, they all constitute semantic layer.

17:05 So now how it connects. So you have your Tableau, the cloud desktop, which is your, um, the client and, um, y- you actually have to give three values, um, or configure three values in your cloud desktop, which would be your server, um, uh, y- your, your site name, and your PAT. PAT is for the user. Like you, everybody will have their own PAT and which is revoci- revocable, and then you have to configure it on the, um, on your cloud instance.

17:37 Then it talks to the MCP, Tableau MCP, which is on the Tableau Cloud, and Tableau… The MCP in turns talk to the governed data, which is published on the Tableau, um, cloud.

17:52 And then I have also mentioned the steps which you need to do, um, to configure this all. And we'll– I'll also show you, like, um, we have a repository where you can go and get more information on it.

18:07 So this is the architecture. So you, uh, so on the left there is business user.

18:12 So you actually talk to the agent, um, I, I'm, I'm sure you all would have used at least ChatGPT. So you just talk in natural language, the LLM, and then, um, that talks, that is your, uh, AI agent. So in, in our, this demo we are using Claude, like I said earlier. So, um, so Claude, these all AI agents have LLM reasoning, and then they do planning, evaluation.

18:36 They talk. They actually are More like an an-analytic, uh, uh, an analyst, like any analyst will do.

18:45 Um, they, they, they will actually explore data, analyze it, and give you the results. Same way the AI agent will work for you.

18:54 And then you have Tableau MCP, um, which, which actually can… It's a bridge which connects to the certified data sources, your semantic layer, uh, your views, workbooks, um, and it's governed and audited because it's all whatever access you have given on your cloud instance, it actually uses those access for each user, so low le- low-level security.

19:16 And that, uh, that in turn talks to your other systems, um, uh, in your organization. I've listed a few here, but you can have a number of them. Uh, now the key guard-guardrails here is like PAT auth, which I already told you about. Like, it's a low-level security. It's like the access, whatever you have on the Tableau c-cloud, you will be given the same access through PAT.

19:39 And there's no direct SQL to any of the databases. It's all what you have published on the, on your Tableau Cloud. It will be just be able to access only da- those, uh, published data sources.

19:53 So that's semantic layer lock, and then there's a full audit trail, full audit trail, and there is no sh- there are no shadow pipelines. Like, these are the guardrails and the key and the moat, um, to this whole, uh, configuration.

20:07 Now, like, like I already, um, s- um, briefly said, data sources are, are the semantic layer. So because you have actually published something which is, um, certified, right? You have the calculations and everything done, and then they're published there. So the agent actually gets all the definitions, everything from the published data sources. It's not fabricating anything.

20:29 It's just giving you whatever is, um, there certified and on your cloud.

20:35 So, um, there is no hallucination, right? And, um, that's the key which I think everybody looks for. You don't need any, um, you know, anything which is v– the truth. You need the truth, whatever is there, and then that it, it's not fabricating anything.

20:55 So now let's go to the demo, which is the,

21:00 the exciting part.

21:02 So I'll open my Cloud, Cloud instance.

21:07 So I'm gonna start the fresh chat window. I have run through the demo earlier, but I would like to start fresh and then show you, um, what exactly it does. So I have my prompts ready.

21:20 Uh, I have underst– I have, um, from the runs I have done, one thing which is very key is your prompts. Your prompts should specify what exactly you're looking for. If you keep them, um, um, open-ended, and then you don't have specifics, it will, it goes, um, it gives you something which you might not require. I mean, it gives you more information than what you need.

21:42 So it's, it's nothing bad, but then you might not need the information which it gives you.

21:47 So be very specific what you need. So I'm asking, um, the a-agent, like, um: Our enterprise renewals slipped in Q4 twenty twenty-five, so just focus on twenty twenty-five data. Because if I don't give it to– give this prompt to the agent, it goes till twenty twenty-six, and then it starts comparing and all that. So I'm just specifically, I want the agent to focus on twenty twenty-five data, and then give me the root causes, um, through these three data sources.

22:17 I'm listing the data sources, and then tell me the journey because I think it gives me the ideas, but I'm more interested in looking at the names.

22:27 So I ask the agent and see what it does. Actually, it's, it's… Analyst will, will do is it does the same thing. It's, it's trying to fetch the, um, trying to understand what's the, what we are asking for. It's fetching the… It's looking for three data sources, like the way we have listed renewals, customer health, and support tickets. It's got all three of them, and then it's gonna read the data and then analyze it, and then give you the, the results or the answers you're looking for.

22:58 So if you see here, it's actually doing, like any human or analyst will do, um, getting the metadata throw through. And if you see there are three, right? So it's three data sources we have listed, so it will take three. It will– It's parallelly doing the three, um, um, calls to these three data sources.

23:19 And these data sources are on our cloud instance. Um, sorry, the Tableau Cloud instance. So it's, um… And we are connected to it. Like I showed you, uh, you have to do the configuration.

23:31 So we are connected to it through our AI agent, which is acting as a client, um, to the Tableau Cloud.

23:39 So let's see how much time it takes. So it's querying now.

23:43 So it, it got the metadata. Now it's querying the data sources.

23:49 Right. So let's see how much more time it will take.

23:54 So we actually, so we have asked it to sh-show only Q4 twenty twenty-five data and show us the results.

24:02 And, uh, while it's f- actually bringing it up, I can show you what exactly are we looking at, um, through our workbook. So we do have a workbook.

24:14 It's actually trying to fetch similar information. So we have a dashboard for twenty twenty-five, specific for Q4 twenty twenty-five, and now it's actually getting the information from the data source itself, not from the dashboard. We'll go to the dashboard as well, and I'll show you how it fetches from the dashboard. Currently, we are just asking for To fetch from data source.

24:37 So, still it's still fetching it. And if I would have used like already run, um, um, I have run the demo already, and if I have used this one, it would be faster because it has cache and all that, and sometimes it will just even ask you why you're asking the same question, which is funny.

24:56 But, um, because it's a f-fresh, um, chat, so it's gonna do everything from scratch.

25:03 And maybe I can go to the dashboard again while we're waiting. So you see the enterprise, there is a dip in Q4. So that's what we are looking for. We're looking for the answers why our enterprise, um, has a dip in Q4. From the dashboard, all we see is there is a dip, right? And then we can analyze, we can look at it. We can just analyze and come up with our analysis, but then, uh, that won't be, um, uh, that won't suffice for the, um, the board meeting, right?

25:36 So if, if, if you will look at the answers that we get from the agent, you will see the, uh, what it brings to the table.

25:48 Okay.

25:49 Hopefully it comes back faster. So still fetching the information.

25:58 Okay.

26:00 Bear with me. Yeah, when you start the new chat, you don't know.

26:03 Um, sometimes it takes more time, and sometimes it does give you the charts too, which is like very interesting. It takes more time actually building the charts.

26:14 Hopefully it's not doing that this time. Okay, so it says completed task, completed task. So three data sources it has completed the task of, uh, getting the information, the analysis done, and now the results.

26:28 So it's just telling you what actually happened, right? So it's like if you see it is actually giving you three root causes, um, highlighted them for you to focus on them. It's giving you the churned accounts, um, the churned ones in Q4 2025.

26:47 And if… It also gives you what you can do next. Um, I hope it gives you here also, but I've seen it doing it.

26:55 Uh, so if you see here what actually happened, right? It says enterprise renewals fell to five of eight.

27:02 This is what…

27:04 This is how it looks, and this is the dollar amount. The… There's a loss in ACV, um, and tells you analytics add-on is the, is the key you should look at. That's the bottleneck, the, the focus area, which actually caused the dip or losing the accounts.

27:23 Um, and then it talks about, uh, uh, support tickets, like you have a lot of P1s in, in, in this, in this, uh, um, the same, um, domain, like analytics add-on.

27:35 And then it talks about, um, only in the renewal quarter. So it says it's only actually focused on this quarter more. And, um, if you read through it, it tells you that you had actually, um, more… These many accounts were, um, up for renewal. And then because there were a lot of technical, um, accounts, the technology, uh, companies, so they were actually dependent on analytic accounts.

28:01 And then because we had a lot of P1s and all that, a lot of things going on in analytical space, so we lost those accounts.

28:10 Um, and then it will talk about, uh, concentration with one competitor, competitor loss renewals. So it talks about everything, what happened, why you lost it, lost these accounts, and then it tells you the three churned accounts and then the churn reason. This is product issues from competitor one and product issues.

28:32 And then you have the, the CSMs for this account, and then the P1s and the NPS values, right? So, um, so it tells you it's… There were four open tickets or… So all that an-analysis which, um, you would have to do looking at the dashboards, and the dashboard won't give you the, um, um… Can't dig deep only looking at the dashboards. You will have to go to the, um, databases and then look for the answers, dig through it, and then give you the root cause, and that might take couple of days, if not weeks, right?

29:06 Um, so it all can be done just with one question rightly asked.

29:11 So let's now go to the next one. So we now know these are the churned accounts which were churned in 2025 Q4. Now we want to, uh, make sure that we don't have, um, we don't lose more accounts, right? So we, we wanna know which accounts are at risk.

29:29 So now we are asking it, uh, based on the data, what accounts? Now I'm saying were because it's already run the Q4. What, what were the accounts which were, um, at risk?

29:42 So now, now since it has already kind of run the query and then hit the data sources, it knows now these are the data sources, um, it has to get the information from. It will be little faster. It doesn't have to, um, redo the whole thing. It's gonna build on the already, um, um, existing query, which it, it already, um, finished. So it's, uh, it's like, yeah, logically what, um, any system would do or an, an analyst would do.

30:11 You have the information, you have the data, and you have already analyzed, and on top of it you're analyzing another, um, another question or, um, a pressing, um, question you have for the business need.

30:26 So, um, so it's just saying added task, right? So it's added task. It's not the, um, from the scratch like it did for the first one.

30:36 So it's getting the data sources all three parallelly.

30:41 And let's see. This should be faster than before.

30:47 Okay, I need to close this. I don't know how to do it. Okay. Anyways, this is showing you the progress and all.

30:54 This is cool. I never noticed this. It's actually pro- showing you all the progress, what exactly it's doing.

31:02 Interesting.

31:03 Okay, so it's completed that. So see how fast it's this time? Because it's building on whatever it has done already.

31:12 So it's like kind of a cache. So, so it says that we are asking only for twenty twenty-six. It knows there's data beyond twenty twenty-six, but because we are specifically asking for twenty twenty-six, it's gonna give us what we are asking for. So prompts are really very important.

31:30 That's why we want text engineering too.

31:32 Okay, so, um, now it's giving you, um,

31:37 the accounts which are, which are at risk, right? So NPS, NPS collapse, product users collapse in the same quarter with P1 spike underneath, so ranked risk going into twenty twenty-six. So going into twenty twenty-six, these are the accounts we should be focusing on. They're at risk of, um, churning, and we're at risk of losing them.

31:59 How cool is that, right? So we know we should focus on these, um, accounts. So top five carry this, this much, this much of, um, you know, ACV.

32:09 So, and then why the top four? So, um, these are the top four, and why these just gives you all the, this thing, why you should focus on them. So listed all five of them, and it, it, it also tells you, like, if there is something it sees in the data, okay, this seems like, you know, I see this in data, and it, this doesn't look good. So, which is really amazing.

32:31 So this is synthetic data, so I knew we can't, um…

32:35 You know, it's not, uh, the production data, which would be, um… We do, we do see caveats in production data too, so I can't even say that.

32:43 This one, um, so have between zero and four in all events in this. So, so now I'm going to, um, Q1 and Q2 of twenty twenty-six. So we are asking that, uh-

32:56 Chris Williams: Hey, Kanwal

32:56 Kanwal Kour: … I'm going, yeah. Should I go faster?

32:59 Chris Williams: No, no, no. We lost your screen for a minute. So-

33:04 Kanwal Kour: Oh.

33:04 Chris Williams: So maybe recap maybe the last twenty to thirty seconds you did before.

33:09 Kanwal Kour: Okay. So where exactly… We, we, we got this one via the top four, right?

33:15 Chris Williams: Yes, yes.

33:17 Kanwal Kour: Okay. So yeah, so the question I asked was, like, just let us know the accounts which are at risk for the next year, and then we don't want to lose it. So it gave us the top four accounts which are at risk, and then it gave us some caveats of, from our data. Like it… I don't know if you got me, uh, heard me saying that because it's synthetic data, so there are some caveats in the data.

33:37 But yeah, it's really cool that it's, it's actually telling me, um, or telling, um, in this case, Sarah is asking these questions that this, this data has this many, these caveats, and you might want to look into it, right?

33:49 So we got all the answers, uh, which we were looking for. Now, the next question I asked was, how did the enterprise segment perform in Q1 and Q2 of '20? So going to go into Q1 and Q2 of twenty twenty-six, we want to see how we perform. So we have data until Q2 twenty twenty-six because, um, we are still in Q3, right? So we are looking for only first two quarters of this year.

34:11 So now it's looking for tw- twenty twenty-six, which is also kind of fresh because we restricted it to, um, Q4 twenty twenty-five earlier, and now we are telling it to go beyond that. So it's kind of new query again.

34:24 Um, and then so again, fetching the data from three data sources which we listed, and then it's trying to get us the answers. And then, uh, while it's doing this, I can show you how our twenty twenty-six looks.

34:37 So we do have, um, on our, for, in our workbook, which tells us the twenty twenty-six, uh, trend and, um, um, data. First, specifically for enterprise renewals, but because we were focusing on enterprise renewals, right? So we had a dip in Q- Q4 twenty twenty-five. We see it's getting better. It was mu- It's much better now.

34:58 But let's see what our agent does, right? So when we look at this, we're, oh, it's improving and all that, right? We see the improve in the renewal rate, and we see the P1 tickets are lower now and all that. But now you will be amazed to see what your agent tells you.

35:17 So it's telling you…

35:22 Okay, so the trends in Q2, there are no Q3, okay, that we know. And it tells you, okay, this is what it is. So we have improved a lot. The renewal rate has climbed back up. But it's, it also actually compares with if you see Q1 and Q2 of last year, it says, okay, we ha- we are still lower than, uh, last year. Last year, Q2 twenty twenty-five was one hundred percent.

35:43 We are not still there. So it mentions that too here.

35:47 And the real story, the segment lost half its book, right? Still tells like we still have some issues. It's not like all, all happy and like all good and happy-go-lucky. No, there are still issues, so you should still focus on these two.

36:00 So these are churned. These two got churned, and then it also tells you the healthy-looking renewal rate is partly arithmetic. You're renewing a high percentage of a much smaller denominator. So it's still showing you that there's still issue. Like on the surface it looks good, but there's still issues, right?

36:18 So root cause never got fixed. We still have product issues. We still have P1s, so look into it.

36:24 Uh, which is amazing, right?

36:27 So this is churn, this is churn, and then renewed twice. This has been renewed twice. It is also mentioning that, right? And then this is at risk, so look at it. Try to hold them, try to not lose them.

36:42 Uh, and this is the second one. So amazing it tells you everything beyond the dashboard.

36:48 Now let's go to our…

36:51 Go beyond this, and we are asking, um- Now we actually have found, like we caught, uh, accounts which were, which were at risk, right? And then we never asked, like, did they… They were at risk, but you listed them, but did we lose them or no? So we're asking, like, okay, we got that these are at risk, and did we work- we did our homework, we, um, how, how could or how well we did to retain them and not lose them.

37:22 So that's the question we are asking. So it's saying two of the ten who bought two products, so we actually lost two of them.

37:28 But, um, but the rest eight, we could still hold them. But yeah, we lost two of them, and they are both big accounts, I guess, that's what it says.

37:38 Uh, only two enterprise churns recorded in twenty twenty-six. Two other I ranked this and this disappear from the renewals and health data afterwards with no churn record at all. So they are not churned.

37:48 So now it's actually telling about the data, right?

37:51 So they're not churned, but I don't see them. Um, and then, um, this could be a data gap. So it's, it's calling out the issues in data too.

38:00 Um, are still… The other six are still active, right? And then it lists the number of, uh, the names of the accounts. It's amazing actually, the all it can do for you.

38:11 And now, I… So, uh, like, like I showed you, there's a dashboard, right? We have a workbook. Here is the workbook we have, and we have two tabs, uh, uh, in the workbook.

38:22 So it's, it, it not only, uh, can fetch the information from the data sources, it can fetch the information from workbook. And it actually… And anything which you have, uh, published in, on your Tableau Cloud instance, it can fetch the information from them and give them to you.

38:38 Um, um, you can ask for even what admins will ask, um, um, like go to the cloud and see like the site, uh, stats and all that. You can ask all those sort of questions because it's connected to your instance, so, and you ha- if you have access, right? If you're the site admin, then you… If y-you're not site admin, you won't be able to get it. So that is your PAT authentication, which works with your, mm, at the level of authentication given to you on the cloud instance, and that's governance.

39:10 Um, so you can go beyond this if you have, um, your v- your, uh, site admin or, or, or for your organization or for your sites on the cloud instance, and if… Or your server admin, if you have Tableau Server.

39:26 So now I'm asking, will the agent ready summary twenty twenty-five. I'm specific twenty twenty-five I'm mentioning because I tried it earlier, and if I just see, ga- give it a gen- ready summary, it just kind of… It's gonna give you both.

39:38 It is gonna fetch that information from both of them if you're not very specific. And then if you have a copy of… Like, I had a copy of this workbook, um, similar names, but you know, an old copy which we ha- which we presented in our, um, this twenty-six, uh, presentation. It was actually fetching the information from that also. So and told me like, "Because since you have two, I'm getting information from both of them." So I had to remove it.

40:07 Um, so that's interesting. It's more like it works like a wild card thing. If you give it a wild card, it is gonna fetch the information from everything matching to that, um, uh, word or at the prompt you're giving.

40:22 So be very specific. I can emphasize more on that, but yeah.

40:26 So now, um, it's asking us a question, what format do you want to the board summary in? See, amazing. I didn't tell it that it's a demo, so it's taking it very seriously that's the board, we have to present to the board, and how do you want to see it? So you can actually, if you are the VP sales or y- anyone who has to sh- give the presentation or, um, you know, in the, in the board meeting or give it to your, your manager, you will have to…

40:55 You can choose anything. But let me just go with summary here in chat for our demo purpose.

41:01 Then how much scope would, should the summary cover? Now see, it's asking for, should I go for H1? H1 is like, you know, the Q1 and Q2 both. So it is, it is actually divided into two parts, H1 and H2. So since we have Q1, Q2 already, so it's saying, "Okay, shall I go beyond this?" So I am saying, no, let's, let me, just give me twenty twenty-five.

41:20 So

41:23 now it's gi- it's giving me the information the way I want it.

41:27 So agent ready summary twenty twenty-five.

41:29 So enterprise annual rate is, it… Do you see the dips?

41:33 The story in one line. So it's actually telling you, you can just take this and take it to the board, and then just take this few lines and liners and then just present it. So let's see what it says.

41:48 So one line, a P1 defect wave in the analytics add-on hit in Q4 enterprise sentiment collapsed with it and three large accounts left.

41:55 Amazing, right? If you have to be short, this is what it is.

42:00 And what the charts established, the slip is enterprise only. The cause is one product module, which we talked about, analytics add-ons. Customer said so directly.

42:09 Okay.

42:10 You see the product issues and com- competitor one and budget cut. Support capacity broke, adoption led sentiment down, the NPS is tracked down. The three enterprise accounts lost. These are the accounts which were lost, right? Fortune manufacturing alone is fifty percent of the loss. Two framing notes before this goes to a board.

42:29 Okay, so it tells you if you… There's any caveat or anything, do you wanna change it? Do you want to tweak it? How do you want to present it?

42:35 Um, all that information you can get.

42:38 Now, um, since I think we have, we have time, right, Chris?

42:42 Chris Williams: Uh, we have ten minutes.

42:44 Kanwal Kour: Okay. So I should be able to wrap it. So, um, now since we asked for twenty twenty-five, so now we want the how we did it in twenty twenty-six, right? And from the dashboard. We, we saw it in the data source, but now we are asking specifically look at our dashboard and let us know how the KPIs look for Q4 twenty five, twenty twenty-five, twenty twenty-six.

43:04 I'm just showing it to you because, um, just to tell you like what all it can do for you.

43:11 And you, if you s- if you pay attention, it says get view data. Earlier it was saying curated data source, right? Or metadata. So right now it tells you it's a view. It's a view which is in that workbook, um, we call the views, right? Uh, since we all are Tableau people here.

43:27 So we are, um, it's telling you now getting the information from the image, and then it- it's gonna give you the, uh, key KPIs for 2026, similar way as it did for 2025.

43:42 So let's see.

43:47 Since it has done it already, uh, for you, uh, here it will be fast, right? S- I, I suppose and I assume, and I think that's what I see usually. So curated data source, curated data source because it's actually getting the image, and then from the image it is going deep into it. It's not only dashboard. It's gonna tell you, like it told you, right, what dashboard is showing.

44:07 It's not only that. If you go beyond this, this is what it is.

44:12 So, um, so it's a clean recovery story. The underlying charts com- complicate three of the four.

44:20 And then what generally imp-

44:23 what gener- generally improved. What the value beyond the dash- dashboard's saying, it's not what you see. There's more to it, right? So the renewal rate rising on a shrinking base and part- parts left closed.

44:37 So if you read it, it will tell you actually they're still, they're still concerned. The underlying issues are not yet resolved. You still should see the P1s are really high number, and it also talks about somewhere that, um, the, the, the time it is taking to, for the support to, uh, close the tickets, it's like it has increased dimen- tremendously.

45:00 So you look at those.

45:01 Um, so you are not yet in the green zone. But in the, in the… If you see the dashboard, you'll feel like, wow, so we are doing great.

45:10 Uh, we have improved a lot. It's all green, green, but it's not. That's where it's telling you.

45:14 And that's why it is fetching the information from the data sources beyond the dashboards. Now, I have grayed this out because the thing is I wanted to show that, uh, it… You can, like m- my point earlier when I said you cannot hit the databases directly, um, but we don't have a database here. These are all, uh, synthetic data, um, the CSVs and Excel files.

45:37 So, um, but so I actually framed the question, that's why I'm saying the prompts are very important. So I'm, um, my, my question is, is this a data… If this is a database connection, can you hit the underlying database directly?

45:49 Right? So because if I say, "Can you hit the database directly?" It says, "Oh, there's no database." It's that smart. It will just say, "These are the CSV files. Come on, what are you talking about? There's no database." So I'm asking it, like, if in case this is a database connection, would you be able to hit the underlying database directly?

46:06 Yeah, um, please ignore my grammar. It should be a comma here but it's okay. It actually, if you have typos also, it just ignores that. It knows what you're talking about, um, which is great. You don't have to worry about gramm- grammar or, uh, the typos.

46:25 So now it's telling you,

46:29 can it connect to database?

46:35 Let's see what it says.

46:38 Chris Williams: Okay. We've got about seven minutes.

46:41 Kanwal Kour: Seven minutes, yeah.

46:42 Chris Williams: So take your time, take your time. Just giving you a time check.

46:45 Kanwal Kour: Yeah. This is the last prompt, and after that, I think we are done. Uh, we'll just wrap it up.

46:50 So, so it's talking about it's CSV, right? Okay. So every time it gives me different answers, which is amazing. So right now it's saying if it had been a live database… So it says, okay, these are the CSV files, so there's no database here. But if it had been a live database connection, I still couldn't reach it directly. The Tableau MCP, it will proxy queries through Tableau Cloud.

47:10 That's what I wanted to, um, you know, focus on or emphasize on, that it's all y- the authentication is all through Tableau Cloud. Whatever access you have, you will have the same access. You cannot hit the database directly.

47:26 So, um, it's talking about there are other MCP connectors you can use. I have tried the Salesforce MCP. You can do that. That's another one. But it talks that this is a way you can do. And sometimes it will give you the SQL query to run. Um, if you have database, just go and run it, but I can't do it.

47:43 Um, okay. So we are done with the, the sand of demo, but let me go back to the PowerPoint presentation and then wrap it.

47:54 So, so what are the three data sources we actually used or, um, um, the agent, uh, got answers from? They were renewals, customer health, and support tickets, right? Um, you saw it all live, so don't have to go deep into them. And then how do you make the data sources or your data agent ready? Uh, like, like, it's like what you do, what you will do for any production-ready data source, right?

48:20 You will… You have to have human-readable field names and acronyms. It should be, um… They should be easily understandable. Um, or, and the definitions should be clean and clear, one metric, one definition, rich catalog descriptions, and all that. So it's the same thing which you will do, uh, uh, to make your data sources production-ready or your data dashboards production-ready.

48:43 So same thing applies here. So you need, um, um, sa- same kind of data for the agents or your dashboards for the agents to get the answers you're looking for, right? So this is the punchline. We stopped treating Tableau as a destination and started treating it as the source of truth the agent stands on, right? So it doesn't… Dashboards don't die with this, right?

49:06 Dashboards are, are still there. We're just promoting them. So it's beyond the dashboard. So we could see the answers, we are, um, the trend and everything we're looking for in the dashboards, but then we can go Beyond that and then get our answers, um, um, what we're looking for.

49:24 I think I'll conclude and, um, if there are any questions and then I do have… I want to show you this. I do have a GitHub link, um, which you can actually, um, uh, go to and get the, um, how you configure the information, how to configure it on Windows or Mac.

49:46 Um, if you have Mac, how to do it on Mac and how to do it on Windows. I did it on Windows just yesterday after our mishap with Mac. Like, um, I don't know why my Mac was acting crazy, so I didn't do it on Windows earlier, but this is, this was really, um, um, interesting and, um, you know, a learning experience for me to do it on Windows as well.

50:06 And this is my email ID. You can lickot- link… uh, reach out to me on, uh, my email and then this is my, um, LinkedIn, um, uh, link, so happy to connect.

50:19 Chris Williams: Well, thank you, Kanwal. We did have one question.

50:21 Kanwal Kour: Yeah.

50:22 Chris Williams: Um, uh, David asked, "What is your process for checking the AI output for accuracy and completeness related to the answer you are giving to your boss?"

50:31 Kanwal Kour: Yeah, so actually, um, that's a very good question, right? So for me, I have done… Like, when I was doing this, I did both. I did my own queries and my own check and then, um, um, did my check on what I got it from AI, uh, agent answers. And s- I do the similar thing. I do the checks, both the checks when I am actually giving the, uh, answers to, um, the, my manager or the executives are looking from me, from me.

51:01 I do both. I, I, I do multiple steps so that I don't have any wrong answers, um, given to them. But what I have seen so far, um, it, it has been always ac- mostly, I think always accurate. I have not seen it giving me wrong answers. Sometimes even it gives me the answers which I might have missed or the numbers or if I miss something, a human error or whatever.

51:27 But, um, it has so far given… I have not seen any issues with it, so I, I can say that I can rely on it, on it fully.

51:37 Um, but I still do all, uh, I'm getting more lenient and lenient when I check with, um, on, uh, from my end, but then I still do the checks. But maybe I'll give up on that, like the way it is, um, it has always confirmed that it is doing its job well.

51:55 Chris Williams: Okay.

51:56 Kanwal Kour: I hope I answered that que- um, the question, yeah.

51:58 Chris Williams: Definitely got to keep the human in the loop for sure.

52:01 Kanwal Kour: Yeah, I know. I know. We, we want our jobs too.

52:04 Yes.

52:05 Chris Williams: So, um, Kanwal, I want to thank you, uh, again for presenting today. This was fantastic. Um, uh, appreciate, uh, all the content always. And, uh, for everybody else, uh, our next official one will be October 1st. I'm probably going to try and shoot to squeeze one in on the 22nd of September, so stay tuned. I will work that out in the next 24 hours.

52:29 Um, if you missed, uh, if you wanna see this recording, I will make sure that this gets turned around pretty quickly and put up on YouTube, and I'll make sure that link is available to everybody.

52:40 Uh, so without further ado, thank you so much for taking the time today and, uh, have a wonderful rest of your day. Thank you.

52:47 Kanwal Kour: Thank you.

52:49 Have a good day, everybody.