Skip to content

Summary

The Analytics Tableau User Group's August 2026 meeting, Beyond Dashboards: Designing Analytics People Love, recorded by the group on its own YouTube channel and republished here with credit. Host Prasann Prem, who leads the group with Annabelle Rincon and Chimdi Nwosu, introduces three speakers. Igor Garlowski, founder of the BI consultancy Data Dune, shows how to build your own Tableau AI assistant: analysts lose a large share of their time to ad hoc questions, data democratisation has not fixed findability, trust or buried business logic, and AI can help with the first two but not with data literacy or ownership. His architecture sends a user's question through a RAG agent over the documentation, then in parallel to a database agent and the Tableau MCP, and combines the answers with citations and conversation memory, all orchestrated in n8n; he demonstrates it and explains what it takes to move from a Docker demo to production. Ilija Stojić, a senior business analytics architect at Toyota Motor North America, compares six ways to visualise performance against an objective, using live chat polls to show how the same data answers 'which day missed worst?' differently depending on the chart, and weighing each method on pass/fail status, the size of the gap and volatility, with win-loss charts and IBCS-style notation among the options. Angela Henderson closes with streaming-style UX: turning a cluttered fleet dashboard into a clean, zero-scroll Tableau app built around a three-step loop of problem, cause and action, using dynamic zone visibility, 'ghost highlighting' so labels and marks react together, a collapsible drive-time map, and Tableau Pulse with Slack for the action step.

Key points

  • Igor Garlowski: analysts spend a large share of their time on one-off questions; an AI layer over your data can make answers easier to find and understand, but it cannot fix data literacy, ownership or access control.
  • Much business logic is trapped in visualisation tools, READMEs and people's heads; an assistant needs a single queryable source of truth for definitions such as what churn means.
  • His assistant architecture: a RAG agent enriches the question from documentation, then a database agent and the Tableau MCP run in parallel, and a combiner reconciles their answers with citations; running the two in parallel avoids passing a half-formed query from one step to the next and saves time.
  • Disagreeing answers from two sources are a signal: they mean the answer is not yet trustworthy, and the assistant should show both.
  • Ilija Stojić: the same actual-versus-objective data can paint six different pictures; pick the chart by what your audience needs to see: pass/fail status, the size of the gap, or its volatility over time.
  • Colouring whole bars by variance shows the split between hits and misses at once but makes readers scan twice to judge the gaps; ask your stakeholders whether a miss means under target only or any distance from it.
  • Discrete date pills can silently drop a day with no data, making a gap look like a steady trend.
  • Angela Henderson: stakeholders want dashboards that feel like apps; design them around a simple loop of problem, cause and action rather than a wall of charts.
  • Ghost highlighting replaces Tableau's grey-out on hover with your own true/false hover calculation, so labels, bars and heat map marks light up together.
  • Keep secondary views such as a map collapsible behind a toggle, and hand the action step to Tableau Pulse and Slack so the insight reaches the person who acts on it.

Jump to a chapter (22)

  1. 0:00 Welcome and the Analytics TUG team
  2. 2:30 Introducing Igor Garlowski
  3. 4:20 Igor Garlowski: build your own Tableau AI assistant
  4. 5:36 Why analysts need it
  5. 7:08 What data democratisation didn't solve
  6. 11:41 The architecture: RAG, database agent and MCP
  7. 14:09 Building it and the demo
  8. 32:24 From demo to production
  9. 42:56 Questions for Igor
  10. 43:40 Introducing Ilija Stojić
  11. 45:45 Ilija Stojić: visualising variance
  12. 53:04 Six ways to show performance against an objective
  13. 1:01:30 Win-loss charts
  14. 1:06:52 Standardised notation and IBCS
  15. 1:13:15 A win-loss calendar in Tableau
  16. 1:17:20 Introducing Angela Henderson
  17. 1:18:09 Angela Henderson: streaming-style UX in Tableau
  18. 1:20:59 Open it like Netflix: problem, cause, action
  19. 1:23:44 Ghost highlighting
  20. 1:25:18 A collapsible drive-time map
  21. 1:26:36 The action step with Tableau Pulse and Slack
  22. 1:34:46 Q&A and close

Transcript

0:00 Prasann Prem: Hi, everyone. Welcome to Analytics TUG.

0:03 And what makes this TUG a bit different, it's… Yes, it's about analytics in Tableau, and recently we also got to know there are so many new things coming to Tableau that I can't talk about. But yes, uh, there is a lot coming, and Analytics TUG has been always there to share more stories, what is going on, what people are trying.

0:23 And before I go on to the speakers, this is our team. We have Annabelle, it's me, and then Chimdi. We all three are leading, have been leading, uh, for quite a while, while Annabelle started this TUG.

0:37 She is a Visionary and Ambassador, and currently leading the self-service adoption for a reinsurance company.

0:46 And she discovered eleven years ago Tableau, and then she felt in love immediately. She has six VOTD, enjoy baking, watching birds and nature. You can also go to her YouTube channel, you can scan the code.

1:01 And then it's me. I love Tableau, I love the community, and that's why I'm here too.

1:07 And it all started back in two thousand and twenty. Currently working as Manager, Data Visualization at ENBD. And, uh, I just love data art, and cats, and food, and anime.

1:19 And then we have Chimdi, who is also a Visionary and a Tableau Ambassador.

1:24 He does freelance contract data visualization for different clients in retail media and other domains via Westside Analytics.

1:33 Same, in two thousand and twenty, he also started using Tableau. Sixteen VOTDs, and just loves building.

1:44 Yeah.

1:44 It's, uh, his work is great. And he also enjoys documentaries, chasing sunsets and working out.

1:52 Okay. So it's me, and, uh, I can keep on talking, but let's move on to the ground rules, as there's none, just enjoy. It's being recorded. It will be there on the YouTube channel, so you can always check it out.

2:09 So today we have a great agenda where we are going from AI Assistant to also talking about performance, and then also UX and UI. So it's a complete day where we have different kinds of topics. And first we have Igor or Igor Garlowski, and he's a data visualization and BI leader who helps organizations solve complex data challenges. And he's a Tableau Ambassador, speaker, and founder of Data Dune.

2:41 He has worked with various organizations including McKinsey, Siemens, and, uh, Pfizer and the NHS.

2:48 And he, he's a strong advocate for practical BI and believes the best analytical solutions are easy to maintain, user-focused, actionable.

2:58 Ultimately, it should work for the people and businesses who rely on it. And today's presentation is going to be on build your own Tableau AI Assistant with MCP and n8n. So handing over to you, Igor.

3:15 Igor Garlowski: Thank you so much, Prasann, for having me.

3:17 And as I mentioned a second ago, unfortunately, whole YouTube doesn't work for me when I'm on VPN, and LLM is based on VPN, so good I have a recording.

3:27 Right, let me share my screen.

3:32 Mm, mm, mm.

3:39 Okay. Can you see the right screen or the presenter screen?

3:57 Prasann Prem: I can see just a big bar showing, uh-

4:01 Igor Garlowski: You're sharing your screen.

4:03 So it's a third screen maybe.

4:06 Is it?

4:08 Huh.

4:11 What should… Share entire screen. How about that?

4:14 Prasann Prem: I can see your presentation.

4:16 Igor Garlowski: Yay. Okay, technical.

4:19 All right. So let's get to it.

4:24 Uh, so thank you everybody for joining, first of all, and spending with us your two hours.

4:30 I much appreciating it. This session will be about AI and creating your own AI layer.

4:37 Uh, so the outcome of this forty minutes will be that you should be able to, on a basic at least level, within two, three hours, put everything I'll be presenting together and have on your own data and MCP.

4:52 And this page should work. Hey, it's working.

4:55 My name is Igor Garlowski. I'm a business intelligence consultant, and I'm leading a BI consultancy called Data Dune.

5:04 Several tools that we are predominantly working with, uh, Tableau is usually in the center as our visualization tool, and several clients that we had a pleasure of working with over the years.

5:18 So the agenda. We'll be discussing why are we here, how does it work. I will show you the demo, and then we'll go into the nitty-gritty details of what can you expect on the way to be a challenge and how to resolve it.

5:31 And finally, we'll be discussing how to get to production of all of this.

5:35 All right. So why are we here?

5:39 The existing problem is that we are, as analysts, wasting a lot of time on pulling numbers on an ad hoc basis and answering questions that are often one-off. So what are the quarter numbers? Where is the dashboard, et cetera.

5:54 And I would like you to raise your hand if you're receiving questions like this.

6:00 So where was this dashboard? Can I pull this data?

6:04 All of the things that can take sometimes hours, but realistically speaking, we have right now tools that could replace this boring work and enable us to do the interesting stuff.

6:17 I don't actually see any hands, I'll be honest with you. Otherwise, I'll have to disconnect all of this.

6:22 But thank you for trying.

6:23 Prasann Prem: I, I don't think, uh, anyone can raise their hand actually-

6:27 Igor Garlowski: Really?

6:27 Prasann Prem: … because most of them can just chat.

6:31 So-

6:32 Igor Garlowski: I will, I will say next time-… only go in the chat.

6:35 Brilliant. Trust.

6:38 Uh, so-

6:38 Prasann Prem: But Lisa and Angela had, uh, raised their hands, so you got some.

6:43 Igor Garlowski: Brilliant. Love it.

6:45 So depending on the article and statistic you'll be reading, is that between 20 and 40% of time that our analysts are wasting on questions like this.

6:55 And again, this time could be spent better doing things like automation or better reports or looking at the data and how we can actually use it to bring more value.

7:06 And we try to resolve it over time.

7:08 So all those questions supposed to be resolved by something called data democratization, which for those of you who don't know what it is, it's basically trying to give access to all of this to our users directly and us as creating the system.

7:26 And it became a nightmare, because we wanted to make it easy to find and understandable and trustworthy, and all those things that we would like the system to be, but it's not, because in reality it is warehouse as a data catalog.

7:42 Trustworthiness, it's segmented. There are no doc- there's no documentation. Nobody's taking ownership of those things.

7:49 Your data literacy is a PDF and a corporate training that you're just quickly trying to click through because your manager is bugging you on Teams.

7:57 It's not really working.

7:59 And it's not working for several reasons that are very hard to resolve.

8:03 And they tell us that AI will resolve it all, but in reality it won't. Let's not be lying here.

8:10 But it can resolve couple of things easily.

8:13 So it can allow us to make it easier to find.

8:16 You just type in, and it's telling you where this piece of data is.

8:20 It can make it more understandable because it is normal, natural language. So, "Hey, explain it to me like an eight-year-old." It can bring some trustworthiness depending on how the system will be created, and I'll be showing you what in my opinion, my experience can actually work for a trustworthy layer.

8:40 And then the tools that the business can use. This is crucial because a lot of data democratization ends up being in a warehouse or with tools that demand still some kind of querying language. Let's be honest. C-level that doesn't know anything about this stuff is not going to sit and query it. They want to ask a question, go for a coffee, come back after five minutes and get the answer.

9:04 What AI is absolutely not resolving is data literacy.

9:08 If your stakeholders are absolutely data illiterate and they have no critical thinking, this is not resolving it at all.

9:16 And the clear ownership and accountability is also not there, because this is just a layer we are discussing today. This is on top of your data, on top of your tables, on top of your data sources.

9:26 If those are having no ownership and are bad, then nothing is really changing.

9:33 And the matched access is also not being resolved because again, we're just talking about the layer, layer on top of your tables. You have to manage the access.

9:43 Otherwise, you will give AI this either limited access or too much of it.

9:50 And continuation of the problem is the business logic. So a lot of business logic is being entrapped in visualization tools such as Tableau, and that's a problem.

10:01 It's in people's head, it's scattered across different documentation. So the SLA definition lives in a README.

10:08 That is the number. Is the number bad? Context lives in someone's head.

10:12 There's no single point of truth what it really means, and what is the history of this metric.

10:18 And then the one person that actually has a lot of knowledge because been there for a day, he quits.

10:25 Now, we need a single queryable source of truth for that.

10:31 So what is the vision of this?

10:34 What if anyone could ask question in plain English? It's as simple as that.

10:38 Get the accurate answer. The question is, what is, what is the churn rate this quarter?

10:43 And the answer is based on our data.

10:47 That's what we would like. Th-th-this is the perfect world, right?

10:55 So how does it work?

10:57 What's inside of this?

10:59 We will have three sources. You can have more of them, you can have less.

11:04 We will have it always cited, so the answer is actually going to tell us what is the source of this specific information, which is important because you have to figure out as the stakeholder, as the requester, is this answer correct or not?

11:19 And then it remembers the context. That's also important because you don't want to start every single line from the very beginning giving context on your own.

11:27 And finally, you can learn on the usage. So there is a small system implemented in it that allows us, the developers of the system, understand what's the input, what's the output, improve the whole model basically.

11:40 Very simple.

11:42 What is the architecture of it?

11:44 You start with user question.

11:48 This goes to a RAG agent. The RAG is all your documentation, all your context. This is the enrichment of the actual query, which later is being split in parallel to the database agent and to an MCP for Tableau

12:07 And finally, answers from all three of those are going into combiner, again using AI, trying to figure out, does this answer make sense? Let's say that we're receiving two different answers from different sources.

12:20 I would like to know all of that because this means the trustworthiness of the answer is almost zero to me at this point.

12:28 And you're receiving the output.

12:29 That's all there really is to it.

12:34 So why are we doing it sequential?

12:38 Because…

12:39 Uh, sorry, why, why did we find it out and not do it sequential?

12:42 Uh, originally I tried to do a sequential analysis, so we're having RAG goes to DB goes to MCP. However, it is being processed on the fly.

12:51 So now when the information is being passed from DB to MCP, the original query might be not formed exactly the same way, and the MCP or whatever is your third step would get already the answer the previous one that was coming in. So that didn't make any sense.

13:07 Additionally, we are cutting out the time because now DB and MCP work at the same time and then we're just going to combine it. So instead of having four steps, we're only going to have three steps.

13:20 Um, starting with couple of small pieces that you should know if you try to do it on your own. You should try to use Claude or any other, the agent, LLM rather than ChatGPT simply because the schema is very strict when it comes to ChatGPT and MCPs often, not all of them, but some of them won't work. Tableau MCP will simply not work if you try to do a ChatGPT.

13:45 Took me a couple of trials to figure it out.

13:48 Okay, demo time.

13:51 And again, unfortunately, I'll be just running a video for you. I tried to run it just before our call.

13:58 Everything is even open but, um, simply you wouldn't even see my presentation.

14:06 All right, so you can just write a script. It will be starting everything that you need. All of this is working on Docker currently, but I will be telling you in the end how to make it to production. You don't have to use Docker.

14:17 Uh, it is starting couple of pieces. First of all, n8n which we're using for orchestration of this whole idea.

14:23 Then it's starting the MCP and MCP bridge that we'll be discussing later why do we end up bridging it.

14:29 And that's about it because the connection to database, it's on its own.

14:35 So I'll open right now n8n and we'll start from the left side.

14:41 This is the RAG.

14:42 So how this works, we'll be triggering the…

14:46 This is normally going, let's say on a daily basis because your README files will be changing. And we'll be loading inside all the README files and all the DDLs, so all the SQL based on which your database has been created.

14:59 This will give all the necessary context to our system so it can actually look directly for it instead of exploring every single time.

15:08 The whole thing is running very quickly. Currently it's being set for local. You can very easily swap those two for GitHub and just pull from there directly. It is looping over those files, changing the format and loading them into something called vector database. If you don't know what's a vector database, long story short, it is basically a database that is easier for AI to read and query through for the right purpose.

15:33 And when this is loaded, I can ask a question that goes directly here and you can obviously expose it in Teams and Slack wherever you want to.

15:43 This is the initial node.

15:45 It starts from querying the RAG system.

15:49 So and read the query itself.

15:53 Then-

15:53 Prasann Prem: Um, Igor, uh, just one minute. Uh, are you sharing a video? Because we can just see your screen.

16:01 Igor Garlowski: No way.

16:03 You didn't see all of this?

16:04 Prasann Prem: No.

16:06 Igor Garlowski: Brilliant. Can you see it now?

16:07 Prasann Prem: Yes.

16:09 Igor Garlowski: Brilliant.

16:11 Let me quickly run it through. This is how the starter looks like.

16:15 You can write it however you want using AI.

16:18 So you're having the MCP bridge and Docker n8n.

16:24 Then inside of n8n it looks like this.

16:28 The left bottom, this is how the RAG is being started.

16:32 So trigger, README and DDL looping over, changing the format, loading them into a vector database.

16:42 That's all there is to it. The load of vector database is happening through AI, so it goes through all those files in a specific format. It's adding it to the DB.

16:53 This DB currently is working within, uh, n8n which means it's not persistent but you can replace with any other.

17:02 You're adding the question, so this is where I said you can expose it to Teams or Slack, and then it's going left to right. If you work with Alteryx or Tableau Prep, it's basically the same but here we have more possibilities.

17:14 So left side we are going through adding the context from RAG.

17:20 Then top, this is where we are querying database. This is just a node of how to query database using AI and here are credentials of the database I'm using which is a Postgres one.

17:31 And then in parallel it's going at the bottom, very simple setup. You have AI agent that is using AI and querying MCP from Tableau with the credentials.

17:42 That's about it.

17:43 All three answers here with those three nodes are being renamed so I know which one is which one and then end up being merged and again processed by AI using specific prompt that I want TLDR, I want the answer, I want the bottom the sources being quoted.

18:01 And then at the bottom left we're receiving an answer. The question was about the churn and what is the churn rate.

18:08 The churn rate plus clients is at fifteen percent, six out of forty clients. Then you have the details of it.

18:15 And at the bottom, you'll be having different sources that hopefully agree.

18:20 And you can also investigate it because it will tell you, all right, so I've been creating CRM schema and clients.

18:27 Docs was this and that. In Tableau with a link, I can select it, I can go directly to Tableau, and I can see the workbook, the data source. I can verify these things if I want to.

18:39 This is basically how it works. I will show you later if we have time, a couple of more questions, but one after another, everything is working.

18:49 Now, to confirm screen with the blue background.

18:55 Prasann Prem: Yes, we can see it.

18:56 Igor Garlowski: Awesome. Brilliant.

19:02 So let's speak about the knowledge layer. How does the RAG work? Because many people don't know what a RAG is. It's just something left and right that people are doing it. It's not actually that complicated. You have documents, they have to be curated.

19:16 Standard phrase, garbage in, garbage out.

19:18 They're being transformed into something called vectors, which is being used in similarity search. Think about it as AI

19:29 looking through data.

19:31 It's creating context to your question. The context often will be something in lines of the user is asking about churn. I can see in DDLs that we have a whole schema around sales and revenue.

19:44 I will be looking through tables inside. Within the tables, have a look at one, two, three, four, five.

19:50 And then when all of this context is being passed to the next stage, then the next stage knows exactly where to look and doesn't have to explore through all of your schemas and all of your tables. And the same would happen for Tableau context. So it would go into and say, "Hey, there are fifteen projects. One of them is called sales department." All right, within the sales department, I can see a lot of non-relevant ones, and there are three with high probability of success.

20:16 Look into them. Because you have to remember that MCP and all AI will not be looking into it indefinitely. It is not like giving it a question, it will be grinding through it for an hour. At some point, after a couple of tries, even if it doesn't find it, it will often try to guess, doing hallucination, or will just tell you, "I can't find it." So in a scenario where you don't have RAG and you have a hundred misnamed workbooks, make sure that you'll be receiving ninety percent no proper answer.

20:48 So when we speak about knowledge base, what do we actually do?

20:51 Well, add things like README, add things like DDL, add the architecture of the system, add the business logic, add the why's, add the, um, all the fixes that you could did over time.

21:05 Everything that you would consider that the new hire– Let's say you're hiring a new senior analyst.

21:12 They are not as good as the senior analyst that's been with you for five years.

21:16 The difference between them, it's what should be documented.

21:20 Then instead of having a big, very intelligent baby, you will actually have a senior AI layer.

21:31 Naming is everything.

21:33 If you struggle to understand sheet twenty-two, you don't know what the workbook means, and you don't know what the project name means, AI will do exactly the same. As simple as that.

21:43 A lot of this is actually about preparation for AI rather than the AI creation, because you can create this layer honestly in two hours. At the end, I will give you the, uh, the t-t-the free, uh, schema that you can use on your own. But if you don't have proper data behind, it will not work.

22:02 So AI readiness, you can Google it. There is a lot about governance. There's a lot about accessibility, ownership, and data quality. All of this has gotta be in place in order for all of this to work.

22:15 Now, don't raise your hand, write on a chat.

22:19 Who knows what are comments on fields in Tableau?

22:24 I don't mean writing a comment calculation. I mean, you already have a field.

22:28 Do you know that you can comment on it?

22:31 I'll give you five seconds and drink some coffee.

22:36 Actually, people have no idea.

22:38 All right. So default properties, you have a magical thing called comment.

22:44 If you comment on it, this is metadata.

22:48 Now, first of all, you in the future can read it. You can understand what was it. You can understand what was the purpose of it and add plenty of different data inside if you want to, things that do not change. Like for example, our decade goal is three percent.

23:06 Then, thanks to this, AI will be able to read it.

23:12 You're adding context using comments. That's very important because you can add a lot of why's at the very spot.

23:19 For example, you can use it also for your work. If you have, let's say, three different dates named almost the same, even naming them differently sometimes doesn't make sense. You can write a comment.

23:29 This specifically comes from this source, and the purpose of it is XYZ.

23:34 Both you, other devs, and AI will know later what to do.

23:38 I'll ask you a question later if somebody writes something.

23:41 Uh, so our RAG. Okay. So the idea here is that you should explain to the system what the system actually is. Otherwise, the system has absolutely no idea, and that feels a bit like working blindfolded.

23:56 If you know that after your system, there are several other systems downstream, then it helps you understand how the current work makes sense.

24:07 So explaining to RAG that downstream in this, for example, system itself, there is DB and Tableau, will allow it to give it a proper prompt and pass it forward because it will know that it's writing for Tableau and writing for DB.

24:22 Explaining the architecture of itself really helps it.

24:29 Business logic and data.

24:31 Okay, so we're going to MCP.

24:34 The, I would say, difficult part.

24:38 What is MCP? Model Context Protocol. Quite a new thing. Think about it as the translation layer between your services and the AI itself. That's all there really is to it.

24:52 Before you could do those things using, let's say, REST API, but, well, you don't want to write all the REST API. What you would like to do is AI just go and query it on their own. That's why you have MCP.

25:05 You don't have to know more about this.

25:08 What is the MCP within Tableau, and what are the parts that we have to remember? That took quite some time to figure it out, and it's still on– everything is updated.

25:19 When you install MCP for Tableau, workbooks with direct links will work, publish data sources, a list of it, and parameters.

25:29 This is by default what MCP will see. The rest of it, it won't.

25:36 Fields and data structure.

25:37 I would like to list all the fields that we have around a churn, and you only have a data source. It will not happen.

25:45 You need something that I call bridge workbook. So if you just show data source on Tableau, it is not being discoverable in any way, and it shows you all the time errors. You have to use it within a workbook and then publish this workbook.

26:02 Only then actually it will be able to query those fields.

26:08 Calculated field formulas. If you would like to say, what is the churn calculation? Not only the bridge workbook, but also the field itself has to be used somewhere.

26:18 So now again, we're trying to go lazy. We only shown data source, then we connected it, we created sheet one, it's all empty, we just published it. This will not happen. It will try to read it, it will get an error. You have to actually use somewhere this calculation for it to be available for the MCP and for the AI.

26:38 The same with those field comments that we just discussed.

26:43 Then you have the live numbers from dashboards. So I would like to say, what is the actual churn number for this specific month?

26:50 Then the return sheets aggregation will only work. You actually have to show it. So it will go and it will look through the sheets in your dashboards, but just because on one sheet you have something like, let's say, you have annual churn on one of the sheets, and you are asking AI what is the monthly churn, because it's nowhere there, it will not be available.

27:14 This is not a tool that makes an analysis. This is not something that's replacing the analyst. This is something that is a very intelligent search tool.

27:26 Then you have visualization data. Again, you have sheets and rows and aggregates. Whatever is there from Tableau will come, but it will not start running through data itself.

27:36 And lastly, visual analysis.

27:39 Show me a bar graph or whatever that doesn't exist. That, that will not happen. This is not going to Tableau and not replacing your analyst.

27:47 Again, it's a search tool.

27:50 So all of this, organization keep ninety percent of business logic in workbooks.

27:57 This is why this is important, because you could say, "Well, there are so many limitations, and the database is working well. I don't care for it.

28:04 And, and MCP is burning a ton of tokens." Well, yes, but business logic lives there.

28:11 Usually, it does, at least some part of it, and honestly, too much.

28:15 So we will always need those tools.

28:22 Now, the answer to this could be, to all of those issues I just explained to you, those caveats, to create a specific set of dashboards for AI when you're actually enabling a lot of this.

28:34 Because on a dashboard, you could say, "Oh, I will have all those nice parameters that are moving the data trunk between month and quarter and year and whatnot." Great. AI can't read this stuff. Whatever you publish, there's a default, that's what it can read.

28:47 So now, in some cases, it is worth considering building a simplistic dashboards that are only available to the AI.

28:59 Permission gotchas. Okay, so how to enable all of this. First off, you have to enable API on your server, which is very simple. It's just you go to Settings, General, Enable API.

29:10 And then you can create a PAT token out of this.

29:13 Now, the problem with the PAT token is that you are creating it from within your account. Which account will you use?

29:20 Will you ask your users to use their PAT accounts, their, their accounts and create PAT and insert in the system? Probably not. I don't see a C-level doing that.

29:30 On the other hand, will you create it as an administrator with access to everything in Tableau, so then anybody can go and query HR data? That's also not the solution.

29:41 So you have to figure out the right amount of access with the right amount of sites with the right amount of accounts to reduce the friction without sacrificing the security.

29:52 Site admin access works very well, by the way.

29:55 The problem with this is as follows: imagine yourself that you had never been on this client before in your life, and I'm giving you five minutes to bring me an answer, which is equivalent of AI scrolling through this for thirty seconds. You would assume that you can do everything you can normally do on every other client, but then you go in here and you start hitting walls, the invisible walls of permissions.

30:19 This is one of the big problems. If you give it the site admin access, it just goes through, queries all the workbooks, and is able to access everything without hitting the walls.

30:28 But again, will you create a site per department?

30:33 All of those are problems that you're gonna resolve based on your scenario.

30:38 And finally, data sources connected to workbooks.

30:41 You have to figure out the permissions for both the data source and the workbook itself within the same account that is being used for your PAT token.

30:52 Governance and tech.

30:53 So when you are having a hundred forty unrelated data sources, it will be a problem.

31:01 We're going back to the same topic of you gotta clean it. There's gotta be a system that you're actually using, you're not just throwing garbage out there. If your analysts struggle, AI will struggle as well.

31:17 Quickly about MCP and Tableau. So when I told you at the beginning that you have something called a bridge between, it has to do with language translation.

31:27 So MCP is using Studio.

31:30 Think about it as being in the same room and screaming across the planet.

31:37 Studio is local on your machine, HTTP is across the network.

31:43 So when you're doing this, Studio is the CLI tool.

31:47 HTTP is the API in browser. You need both, so you need a translation layer called Super Gateway. It's on GitHub, it's open source. It has to run in the background. It is translating between n8n and MCP. That's all there really is.

32:05 So it goes next to the MCP agent on the right side, and it goes…

32:11 It's just running in the background translating it.

32:15 Very simple to set up. Ten minutes.

32:18 Demo time. All right, I will leave it to the end. I put it here because usually it's loading to three minutes every question.

32:25 And let's move to from demo to production.

32:28 So how do you make all of this happen and why?

32:31 First of all, if you try to make it happen with your client or with your employer, they would ask about the ROI. What are the costs of this?

32:39 Uh, I was checking those costs in May based on Sonnet four point six. Now we're running Sonnet five majority of time, so cost will be different a bit. Run it on your own. One query few months ago was one thirty, so a thousand people, five hund- uh, fifty queries per month, it's about seventy grand.

33:00 Now, you can reduce this cost. You can use prompt caching, you can pre-cache Tableau metadata and result cache for twenty-four hours, which will significantly reduce the cost down to sixty cents per query.

33:14 So a thousand people, fifty queries a month, you're looking at only twenty-nine thousand per month. Sounds like a lot of money. It's a thousand people that's accessing almost real-time answers.

33:26 So things like, "How much time would it take us?

33:30 Where is the dashboard X?" That's five minutes of work for me. "How is the churn calculated?" That's at least ten minutes, let's be honest. I have to verify this information before I give it to you, to my client. Deep dive, explain pipeline, ninety minutes at least, assuming I have good documentation.

33:46 All those four questions would take me about two, two and a half hours at a hundred dollars if I'm– what was the cost? At seventy-five thousand salary a year.

33:58 Now, comparison.

34:00 You have unoptimized one thirty-five, optimized about sixty cents.

34:04 We're looking at analysts spending half a million a month to do those thousands of queries, while only twenty-nine thousand on AI. And this is not even the interesting job. We're literally saving analysts from doing the most boring work ever, playing a Google inside of their own system that's poorly documented, instead of documenting the system, thinking about stakeholders, thinking about UI, UX, thinking about all the interesting things that are actually bringing a proper impact instead of, "Can you pull out data from quarter three?" It's boring.

34:37 We are giving users the actual tool that they need. And what's also very important is how much it takes.

34:44 If you ask me to bring you some kind of numbers, even if it takes fifteen minutes of work, you're receiving it in fifteen minutes at the best.

34:52 At the worst, you need it at the hours that I'm actually busy with something else, so you're gonna receive it tomorrow or in a week.

34:59 With AI as a stakeholder, you'll receive the answer at the very moment. So instead of context switching, requesting those things from many people, and then in the future coming back to the same task, you're actually receiving your basic answer you need right now.

35:15 And how to make it real. The web chat is built in in n8n. Then Slack bot, MS Teams is just a swap.

35:22 You can very easily make it in n8n.

35:25 The true parallelism, what I was showing you on the demo, I will show you in a second again, is actually going one after another. I put it that way so we have everything on one screen. But you're wrapping them into a single pill and they start working in parallel.

35:39 Few hours with testing.

35:42 Progress.

35:43 How do you make this demo inside of your company? Well, you can make it on a Docker like I'm doing it here, except if your LLM is not behind VPN.

35:51 And then you have vectors in memory.

35:54 Like I showed you, the vector database is inside of n8n.

35:57 Two hundred bucks for LLM for testing everything max. You can present it to your C-level. If you want to productionize all of this, and you don't care to host everything internally, hosting, uh, last time I checked was like fifteen bucks a month. This is where you're hosting your n8n because n8n doesn't actually do any job. It's an orchestrator. It just goes around and tells everybody what to do.

36:18 Then you have the super base, pgvector. You have to enable pgvector, just so you remember.

36:23 Super cheap, honestly.

36:25 And LLM, again, a thousand queries, six hundred bucks.

36:29 Pretty cheap for a system like that, honestly.

36:32 Couple of problems that you might have. If you're not using n8n, use an alternative. Doesn't really matter. There are plenty of orchestrator systems.

36:40 Uh, if Tableau upgrades, MCP upgrades as well. Currently, Tableau is also hosting an MCP that doesn't work in this specific system. It will work if you connect it to ChatGPT or Claude or whatever. In a system like this, we still have to host on our own.

36:55 Then you have the, if you haven't upgraded your Tableau Server, then just do it. This is not a very good reason to do it. There are plenty of better reasons like security to upgrade your Tableau Server.

37:05 Don't run it on fi- five years ago.

37:09 Like, your DB is not Superbase. Use any other. Superbase is basically Postgres with this extra small thing on steroids. You can use anything else. What about security? If you have PII, that's self-hosted. The same if you don't trust OpenAI and Claude. Just self-host this stuff you have to. You can do it in AWS or in your Claude provider.

37:31 And then if you have six thousand tables, then, then you have garbage of a governance and you just gotta figure it out first.

37:40 Um, so what have we solved if you're going to do this? Again, we created easy-to-find system, which is understandable, trustworthy based on the sources, and it's a tool that your business is actually going to be using.

37:57 So, what you can do right now, audit your naming, put everything in Tableau and in your database in order, do the proper documentation, start owning it, add Tableau comments, explain everything. Imagine yourself that you have– your last job is to put everything in order and then just quit.

38:18 How would other people that you will be not handing over things to at all, only via this documentation, what would they benefit from?

38:26 That's it.

38:28 That's, that's about it. AI is just this kind of person that has to read it from documentation. It cannot have a call with you.

38:36 Um, do we still a couple of minutes, Prasann?

38:39 Or am I already after the time?

38:47 Prasann Prem: Y-you can take a couple. That's fine.

38:50 Igor Garlowski: All right. Let me then quickly show you, uh, a video.

38:54 All right. So we're asking another question.

38:57 I hope you can see the video.

38:58 Prasann Prem: Yes.

38:58 Igor Garlowski: "The delivery times on dashboards haven't updated in weeks. What's going on?" So here is a question of actually coming from a developer. I would like to have a system that can tell me that, because I'm receiving an email in the morning, and I really don't want to investigate this stuff. That sounds like digging in logs and whatnot.

39:16 But you can connect those logs. You can connect live logs. You can scrape them from AWS, put it into the DB, or connect it directly into the system.

39:25 You can have this end-to-end. You could take out technically right now Tableau, MCP, and DB out of this and connect only your logs, and it will be searching only through your logs, and you would ask exactly this kind of question, "Hey, something is off. Can you look through the logs," which I absolutely hate, "and tell me what's going on?" The delivery times, again, this is connected to the, uh, to both dashboards and to the DB.

39:50 So it's looking through it. It hasn't updated since fourteenth.

39:54 Something is off. ETL pipeline silently broke on August fifteenth.

39:58 But this is great.

39:59 This is great information. The fix is thirty minutes field mapping.

40:04 I mean, this whole query ran for three minutes. Don't tell me you can figure it out in three minutes on your own. It's not possible.

40:11 And then also, let's say that the person responsible for it is not there.

40:16 You have to wait for those, or it will take you significantly more time. Finally, you might be a manager or a director that noticed this, and you are not sure. Maybe you would like to check it on your own in three minutes instead of immediately delegating it.

40:30 Tools like this will really be the future, in my opinion.

40:34 It is a tool that helps us do our job better.

40:38 Which open deals should I focus on next? That's for salespeople.

40:42 They don't want to dig through your one dashboard that you have created, and nobody wants to dig through a database. And let's be honest, many CRM systems are difficult to use because they're also not cleaned.

40:53 All right.

40:54 Focus on the specific deal, and then you have it why, and you have it with numbers. You can also request automatically in prompting to give you links directly to Salesforce or to any other system.

41:05 And again, you will have details on them. You will have a detail on the second best, what's going on with this deal.

41:11 What if you would like to screen as a sales manager? You don't want to go through ten people that you are managing. You would like to ask a question around, okay, is everything up to date? Did they do any mistake in the data?

41:24 Did they miss any reminders, any important deadlines? Are any deals stale?

41:30 They have to go to an analyst and create an entire dashboard for this, and maybe that's not something they want right now.

41:36 Also, think about this system as a perfect POC.

41:40 This is why we are saving all the inputs and outputs. Thanks to this, I can later take this entire database of what people are querying and ask, is there something being queried weekly?

41:52 Ooh, by whom?

41:54 Great. Let me have a chat with this person and ask do they want to have a dashboard. Because it seems that they keep coming with the same and the same question.

42:01 Has anybody been showing frustration based on the data, or based on which queries? And you keep upgrading the system all the time.

42:10 So that's about it. If you want to do this one on your own, then you can scan this QR code. There is a whole article, and you'll be able to download this end-to-end workflow.

42:25 Uh, it will be stripped of, well, obviously credentials and prompting. You just add your own prompt based your own use case. You connect the credentials, you connect the LLM, and it just runs. And I know I've been presenting this at Tableau Conference, and a couple of people wrote to me afterwards that they've been doing it in their company, and it actually worked pretty well for them.

42:46 Uh, so grab this one, and if you would like to connect with me, you can scan this QR code on LinkedIn, Igor Belosky.

42:53 Uh, that's about it. Thank you so much for having me. And are there any questions?

43:02 Prasann Prem: Thank you so much, Igor, for the whole presentation. Even if, let's say, you're, uh, you're not able to run it, but I think there, there is a lot of content.

43:14 I don't think we have questions, but I can see comments around that people would like to go back and then check what it is about. We'll, we'll have the video as well, and maybe they can reach you out if they need any specific help.

43:27 Igor Garlowski: Absolutely.

43:28 Prasann Prem: Right.

43:28 Igor Garlowski: And if you need any help, you have any questions, just ping me on LinkedIn. We can have a quick call. I can help you. Like, happy to be there for you.

43:37 Prasann Prem: Great. And thank you so much again for taking this time and sharing all this information.

43:43 Igor Garlowski: Thank you for having me.

43:44 Prasann Prem: Sorry?

43:45 Igor Garlowski: Thank you for having me.

43:47 Prasann Prem: Great. Okay, so now we'll move on to our next speaker.

43:53 Let me share my screen, and first you will see the infinite screen because right now I'm out.

43:59 I'm not at my home, basically. So here we go. Zoop.

44:06 And

44:08 okay.

44:11 Now we have Ilija Stojić, and he is a senior business analytics architect for Toyota Motor North America near Dallas, Texas, while he's also pursuing a Master of Science in Analytics from Georgia Tech.

44:28 He has core focuses enablement and sta- stakeholder engagement, but also 25 plus years of experience for UX governance analytics, and he has been a great computer- contributor with the Tableau community, and also, okay, occasionally mistaken for other baldies who do data.

44:49 That, that's funny. And it's also a cert… He's also a certified Tableau cen- consultant and architect, and a TUG addict. And he has also been a mentor for DVS, that's Data Visualization Society, for four years, and he's excited to restart the, and co-lead the Automotive Industry Sa- TUG.

45:10 Congratulations, Vijay, for this one. And, uh, I think it's more of a point to make, which he can further go on as much like Arthur Dent, he's more deferred from than differing. Okay, so that I'll leave it up to you. And today he's going to talk about visualizing variance, comparing ways to show performance against an objective.

45:36 So the stage is yours now. I'll stop sharing my screen.

45:45 Ilija Stojić: Okay.

45:47 Let's see if I can share my screen. Hopefully everyone can hear me.

45:53 Thumbs up?

45:56 Prasann Prem: Yes.

45:57 Ilija Stojić: Awesome.

45:59 Uh,

46:02 and can you see my screen?

46:05 Prasann Prem: Yes, we can.

46:06 We're good.

46:08 Ilija Stojić: It's a presentation, I hope, and not heaven knows what.

46:12 Um, so today, like Prasann mentioned, I'm gonna be talking about visualizing variance.

46:18 Uh, you'll get a little nap, I guess, between two other amazing speakers as far as topics go, and it's an issue that everyone, I assume, at some point has faced in their data visualization career, right?

46:35 How do you show performance against an objective or a prior period?

46:41 Whatever the case may be, you want to compare A to B, and probably over time. So there's a lot of different ways of doing this.

46:53 I'm going to ask a question now. So if we could leave an emoji in the chat.

46:59 Which day did the worst?

47:06 This type of chart where the full bar is colored by variance is kind of where everyone starts off with in Tableau, right? Very easy to implement.

47:18 Um, and so far from what we're seeing in the chat, different answers.

47:26 Um, still very different answers.

47:31 So this is where folks might start, and what I meant by worst in this case was the biggest miss from the objective, the objective being these horizontal lines.

47:42 We'll go on to a different way to show this.

47:48 Same question, which day had the biggest miss between its objective and its actual, in this case, sales?

48:02 It's…

48:04 It wasn't an intentional trick. This is just what happens, Yusuf, when we're doing visualization. So far more consistent answers now, um, although interesting.

48:19 Uh, there is an interesting point here where, uh, most individuals said day nine had the biggest gap.

48:30 But day 13 did if you're looking at absolute values, right? Like, how far off were you, regardless of whether too much or too little. Extremely valid point.

48:44 So- I'll leave that one for a bit, but definitely a valid point for us to consider on what we're trying to communicate or what our stakeholders are expecting from us. There are times when too much of something is a bad thing. There are plenty of times when too much of something is a bad thing. But in this particular case, yes, I was looking for day nine, but for the lesson in what your audience is seeing versus what you're seeing.

49:20 On a related note,

49:23 I like asking AI agents to evaluate my charts just to kind of get a precursor to what my audience might think of it or be seeing.

49:37 I gave, I think it was Claude in this case, the chart on the top, and it spit out some analysis that just did not make sense. So I went back to the drawing board, and after a few rounds, it's like, "Well, can you just show me what you're analyzing?" And it gave me the chart on the bottom.

50:02 There are, uh, some very obvious changes here, and there is a super subtle change.

50:13 Um, two subtle changes. I'm wondering if anyone could mention, you always write and, uh, is there a subtle change here? So whatever happened with the colors and whether it met objectives or not.

50:32 Wow, that was so much faster than I picked up on it. So someone has already mentioned 18 disappeared, and we have an extra 31 for some reason.

50:43 Extra 31 is unlikely to happen, uh, when you're doing visualizations, but the missing 18 could very easily happen if you're using discrete pills, for example, in Tableau, and the 18th didn't have any data.

51:00 It looks like it's a steady trend, and in actuality, there should be like a little gap wedged in there.

51:09 This was just an aside.

51:12 By coincidence, that wasn't a Confederate, happened to be related to the, well, what is the biggest miss? Is it o- over, under, either?

51:23 Um, it just relates to

51:27 being in sync with what your stakeholders are looking for.

51:31 We're gonna narrow it in a little bit for what can you actually show? Like, what kinds of technical criteria can you actually show in a chart?

51:42 Um, I chose these three in particular.

51:49 There are others, but these seem like the least overlapping ones to me. So you have things like pass/fail status.

51:58 Was it over or under?

52:02 Um, you have the size of the gap.

52:07 So we saw that with the first two questions, right? In terms of which day had the biggest gap, and visualizing it one way resulted in a lot more variety of responses than visualizing it a different way.

52:22 And the last is the volatility. So what does that look like over time? Are there any trends or patterns?

52:29 Is the variance constantly changing, or is it kind of consistently sized and not changing as much?

52:38 Uh, nothing I'm gonna show you today can handle all three exceptionally well.

52:44 Normally, there's a trade-off with at least one element that winds up being a little muted or reduced.

52:51 The catch is knowing what your audience is after and then working your way backward to which chart you want to use.

53:00 So now I'm curious to know, I have six different methods here for showing variance.

53:09 They're all based off the identical data, but they kind of paint six different pictures, is what we were seeing with some of the earlier questions, the first two questions I asked.

53:23 So of these six methods, which do you or have you used in your day-to-day work?

54:18 Thank you for letting me know I'm on mute. So hopefully it wasn't for too long.

54:25 Um, we, we saw a lot of As and Bs, so the two charts on the left where it's either only the variance or the full bar is the color of the sign of the variance.

54:38 Haven't seen any of E, which is interesting given that it's supposed to be a standard.

54:46 And some win-loss charts I'm curious about. I don't see those as much in my work.

54:53 Um- By far and away, A and B are the most common.

55:00 Uh, I only saw two Cs, which also curious about what kind of work they have.

55:09 We'll get into the reasons why in a minute.

55:12 Um, because the AI earlier had completely, and I hope I'm not still on mute, uh, missed the picture of what I was trying to analyze. I fed it the data directly, so this is where we could use some of Igor's tricks a moment ago with using an MCP.

55:34 And I just asked it, "Visualize this performance against an objective." The data was just three columns: date, sales, objective.

55:45 It came back with this. So this time it didn't hallucinate. We didn't have two 31s, a missing 18, lots of above and below objectives that obviously weren't.

55:57 But

55:59 bands aside, they have their place too.

56:04 It sort of mimicked one of the IBCS standards, but not quite. And so case in point, it still colored everything here to match the color of the variance up above. So it kind of combined A and…

56:22 Well, not quite E. A and F, let's say, in these instances.

56:31 But going back to those six methods, so we have one that it used and that was super common. You have the entire bar is your color.

56:41 So if you start off with a bar chart that's maybe just black, and you have your objective, and you're told it's boring, and can you add some more color to this, this is something you might result in.

56:53 Um, the benefit is you can pretty quickly see are there more green or red, or are they about even.

57:02 Um, there's zero ambiguity about that. You know the full volume or scale of what happened, so you know that super tall green bar on the 13th is much more impactful than pretty much anything else up until you get to around the end of the month.

57:24 The downside is your eyes or your stakeholder's eyes need to scan this twice.

57:32 They need to do it once for the colors and again to s- kind of zero in on those gaps, because the gaps themselves aren't visualized, it's just white space.

57:46 Um, that part can be a little bit difficult, as we saw with the variety and the fir- this was the first question I asked.

57:55 Moving on to another method that seemed popular, I think it was the second most popular one, is visualizing the miss or the gap, for better or worse, right? Whether you're over or under the objective.

58:15 In this instance, what you gain out of this chart, or what a stakeholder would gain out of this, is the gap size that's visualized is the variance. There's no trying to guess it.

58:31 Unless things are super similar, you'll be able to tell, "Oh, this was my worst day. This was my best day," assuming going over is good and g- being under is bad.

58:42 Or if it's just variance, period, and it's bad, you st- should still be able to tell.

58:50 Um, the downside with this is when that variance is small, your ability to interpret this or distinguish between them also suffers.

59:06 Um, accessibility aside, it can be f- like, as it gets smaller, especially on a dashboard when it's not taking up two-thirds of the screen in a presentation, it can be hard to distinguish between some of those. So when you visualize only the gap or the variance, it can be more difficult than these, but at least you know what those variances are right off the bat.

59:38 Um, we have another style. Uh, an aside, I encountered issues with the stakeholder being able to interpret the negative variances, so the ones that had a gap between the objective and sales.

59:53 And so I started visualizing things this way.

59:57 Uh, it's a sort of modified plunger plot. It's called a plunger plot because they look like those old TNT plungers people might have seen in cartoons.

1:00:08 But here we were trying to emphasize the, the gap between an objective and sales is actually the lack of something or something is missing.

1:00:21 When you exceeded your objective, it kept the full bar.

1:00:27 Um, because it's encoded several ways now, it's a little easier to distinguish between, well, was this a variance that was under? Was it a variance that was over?

1:00:41 And the downside is it's also more accessible in that sense in that you don't need to worry as much about what color you're using in your dashboard.

1:00:53 The downside is most individuals have never seen a chart like this before, or if they have, it's because they're p- A scientist, statistician, engineer, someone that works with error bars, which is what this would traditionally represent.

1:01:12 So it could be a little misleading for those kinds of audiences, so just take it with a grain of salt, let's say.

1:01:23 Um, another pattern, and I did see several of these, is win-loss.

1:01:33 This chart type, far and away, you'll be able to determine patterns very quickly.

1:01:41 Uh, the– so answering questions like,

1:01:46 how often or how many days out of the month did we hit this?

1:01:50 This might come up in things like attendance, days open, days since last accident, whatever the case may be. The individuals that said they use win-loss charts, I'd be curious in the chat if you could mention how you use them, if you're allowed to.

1:02:05 The downside is you completely miss out on magnitude.

1:02:11 So whether something just missed the objective over or under, or it would cause issues of scale because it missed it so much in your chart, you can't tell the difference between these two.

1:02:30 Um, so case in point, I think day thirteen was the one that went extremely over. Day nine was the one that went very under, and it's just

1:02:42 not discernible here.

1:02:45 There is a type of win-loss chart or a variation of a win-loss chart that can pull out patterns.

1:02:54 I'm curious whether anyone sees any patterns here.

1:02:59 So feel free in the chat if you see a pattern, drop it in there.

1:03:06 I'm just making sure no one's already called out the pattern.

1:03:15 Thoughts? And then I'll change the layout.

1:03:25 Does anyone see any patterns now?

1:03:28 This is the same win-loss information, just in a calendar layout. Calendars are a little bit more natural for people to look at.

1:03:40 You lose the momentum or the chronology of it, right? As it's going through the month, it's a little harder to read it in this format as a timeline, but you do gain something else.

1:03:54 And there we go. So we've had several folks call out

1:04:01 Tuesdays perform very well, or sorry, poorly.

1:04:06 I got that backwards. Fridays perform very well.

1:04:10 And we just kind of charted them out here. Technically, Tuesdays never hit, and Fridays always hit.

1:04:18 With the other v-variations I'm showing you today or that we're discussing today, none of them would show you this. Not for a human being, anyway. An AI agent might be able to make the call on, oh, I happen to notice every Tuesday, Friday, whatever the case may be.

1:04:38 But people in this format, if weekdays matter, this is the chart to use. This is where win-loss, even though you're losing the magnitude, win-loss can be very helpful.

1:04:58 Going from there, back to the ones where patterns make a little less sense, we have one of two IBCS standards. This one is a data series comparison.

1:05:12 It's a standard, so the good news is it's defined what it should look like, what colors you're using. There's no need to debate about can you make it a little more green, a little less green? Can you not use red and green?

1:05:28 The downside is if your audience isn't familiar with the standard, this can be a lot to look at.

1:05:34 Um, a second downside is for accessibility issues. There are certain types of visual impairments that can make this hard to read. You can either get a moiré effect going on.

1:05:48 You could have issues with the red and green, trying to distinguish the size of the gap again.

1:05:54 But as a standard, this is it. This is the official standard. Losing the magnitude in my head. Pop, pop. I'm–

1:06:06 I should not read comments while presenting.

1:06:09 That's a mer- or a catch-22, I guess.

1:06:14 A different standard and one that, again, there's trade-offs being made, is multi-tiered columns.

1:06:27 One downside with all of the other ones that did show the size of the gap or the magnitude is it's hard to compare that variance going across, right? It's hard to see how much of a difference, and we saw that with how there were a lot of variations in which days were doing the worst or the best.

1:06:51 With this, you still have standardized meaning because it is based off of one of the IBCS standards.

1:06:59 It's based off of our nice boring bar charts, which even if you're in third grade, you could probably read. And so-

1:07:10 There's little confusion or you don't have the issue of uninitiated or cold audiences that have never seen this before.

1:07:20 The other advantage is now you have two baselines, one here so you can compare total sales, and one here so you can compare the variance.

1:07:34 The added baselines are being shared across the mini metric, helps quite a bit.

1:07:41 Downside is now you're using up more real estate on your dashboard. You're splitting the attention of the audience because now their eyes are kind of scanning up here, it's scanning down here.

1:07:51 It might be jumping back and forth between the two.

1:07:56 Can introduce a little bit, but for some, this helps represent what's going on the best in theory.

1:08:08 So for anyone that's a fan of taking screenshots, we have this where we just kind of go back through and compare

1:08:18 what the different methods might have been, what, in my opinion, they're best at, what they're worst at or blurriest at, what they bring into focus or not, right?

1:08:31 There is one thing, and there was a chat actually earlier that alluded to it.

1:08:39 I want to say it was David.

1:08:45 And that is

1:08:48 sometimes a problem can be looming, right? The, the message here is a bit of a joke from a side view mirror. But when you're plotting daily variance, it might look negligible, right? Like, you might barely notice this.

1:09:06 When you plot it cumulatively, it turns into this hole that's being dug under a future objective, right?

1:09:16 A future objective is going to have to work that much harder just to break even and then if it…

1:09:26 Like, it would exceed by some amount just to break even, and then it would have to exceed by even more so it can exceed its own objective for any particular day.

1:09:41 So to address that, there's yet one more.

1:09:46 This is a bonus method.

1:09:49 Um, and that is using a horizontal waterfall or a kind of cumulative variance total.

1:09:57 So the good news here is you could see, yes, there was a sharp dip and a lot of red days, but that boost we got on day 13 really bumped us and kept– kind of sustained us staying positive or exceeding the objective.

1:10:18 So definitely helps you to know at the end of the day, doesn't matter, well, at the– in this case, at the end of the month.

1:10:28 Comparing the individual variances though, we've sacrificed that for having this horizontal waterfall. So to know where we're at cumulatively, we've sacrificed knowing or comparing individual variances.

1:10:51 Which kind of brings us to what can you live with or what can your audience or stakeholder live with?

1:11:00 Um, what blind spot are they gonna be least tolerant of?

1:11:06 The analysts can use any one of these charts and probably combinations of these charts.

1:11:14 Um, different roles. I just made up little happy roles here.

1:11:18 Uh, sorry for channeling my inner Bob Ross.

1:11:22 But you have the executive. They might care about the cumulative IBCS, right? They might not need to see that day-to-day granularity, but they very much might care about how is the year or quarter going and are we gonna hit our targets.

1:11:38 Ops manager might be doing scheduling or heaven knows what, and so they care about magnitude, and they care about possibly weekday patterns, right? If they're doing inventory, shift scheduling, whatever the case may be, they might need to know both or a combination of things.

1:11:59 Auditors might be looking for the outliers, right? They might already know what's typical in their head or what's expected in their head, and so they want to know what is off, not necessarily the cumulative effect of it. They just want to know, are we seeing something that we shouldn't?

1:12:22 And then do something about it.

1:12:25 And in a nutshell, that is that.

1:12:29 So thank you.

1:12:31 This is a Hitchhiker's Guide to the Galaxy fan, in case you can't tell from the title and my introduction.

1:12:38 Big fan.

1:12:39 I just wanted to give a shout-out to my co-leads that are restarting the Automotive TUG. We'll be meeting on September 15th at two PM Central, that's Chicago time.

1:12:51 And we would love to have you all, whether you work in the automotive industry or not.

1:12:56 This workbook was available on Tableau Public.

1:13:01 I just wanted to also give a shout-out to Steve Adams, Steve Wexler, Claus Schulte, Sam Parsons, Ivett Kovacs, Kovacs, sorry, that have all done blog posts or different kinds of charts I was showing today.

1:13:15 And if we still have time and you're still awake, I'll entertain any questions

1:13:23 Prasann Prem: Thank you so much, yeah, uh, for the amazing presentation. And, uh, yeah, I, I loved it. And we can see so much interaction over there. And plus some of the things, yes, like the IVCS one, that really hurt my head as well.

1:13:41 So, and funny comment. So, uh, that is great. I think we do have question.

1:13:48 There is one. We have, uh…

1:13:52 Okay. So, uh, it's question from, uh, Wendy.

1:13:56 I think she wants to know, can Tableau create a calendar and shows data? Would you like to take that?

1:14:02 Ilija Stojić: Um, sure. So i- if I misinterpret the question, just let me know again in the chat, but I think you're talking about a sort of trellis chart or small multiples where you have it in a calendar format, but you're still showing something else like magnitude or whatever the case may be.

1:14:27 So that is possible.

1:14:29 Um, you would want to be careful with it because you would lose that baseline, so it can make the comparison a bit more difficult.

1:14:39 What I would suggest, or what I've done in the past myself, is combining the two.

1:14:45 So I might use a variance only or a plunger plot to show it, and then next to it you can have the calendar, the win-loss calendar, or because we're in Tableau and we have lovely tooltips, I will put the calendar in a tooltip, and that way it will highlight which day in the calendar you're hovering over in the other chart, the primary chart.

1:15:12 But that's how I handle it traditionally. If I understood your question correctly.

1:15:17 Prasann Prem: I hope so. And yes, uh, that's a great sugges- suggestion how you can do it, or you can overcomplicate things, like sometimes we like to do.

1:15:28 Ilija Stojić: Yeah. Um, so I do see one more question, and then I don't want to take up too much of Angela's time, but-

1:15:34 Prasann Prem: Sure

1:15:34 Ilija Stojić: … uh, the question is, what do you think of adding a mark label to counter against the small variances being hard to see on the color on visual B?

1:15:42 Um, for any one of them, what I traditionally do is label the largest and the smallest.

1:15:52 It is a question of what your stakeholder is after, because do they want you to label the biggest sales and the least sales, or do they want you to label the most under the objective and most over the objective?

1:16:06 I just didn't do it today because it would kind of give away the game.

1:16:12 So it's a totally accurate call-out.

1:16:16 I just didn't wanna give it away because that would make a dry talk a bit more boring, and so.

1:16:22 Prasann Prem: Mm-hmm.

1:16:22 Ilija Stojić: But yes, I would normally add the mark labels.

1:16:26 Prasann Prem: Yeah. The plunger one was interesting too in some ways. If, if you want to, like, look at that in that way, a bit of information on the first go might be needed, but it was interesting way to see that. Thanks for sharing all the different examples.

1:16:40 Ilija Stojić: Yep. Thank you. Thanks everyone.

1:16:44 Prasann Prem: Okay. So now we are on to our next speaker.

1:16:49 Last one for the day.

1:16:52 Let me just quickly share my screen.

1:17:04 And now we have Angela Henderson. She is a global data visualization leader with experience across finance, digital market- marketing, and product strategy.

1:17:18 She terms herself as viz obsessive, who finds herself mapping everything from global private equity portfolios to Jeopardy!

1:17:30 stats. She recently launched strategic BI solutions to help clients translate ambiguous data into actionable roadmaps. And she is a mentor at heart too. She's dedicated to proving that when you find the human story in the data, you unlock the true power of performance. And, uh, that, that's a really beautiful way to look at things and how you are, uh, solving this.

1:17:56 And today she's going to talk about Streaming Style UX: Building Clean, 3-Step Business Apps in Tableau.

1:18:04 Thank you so much, Angela, for being with us here.

1:18:09 Angela Henderson: Thank you so much, Prasann. Um, and, um, I'm gonna share my screen here. And while I'm doing that, I just wanted to let everyone know I'm really excited to speak at my first Analytics TUG, um, first time ever. Thank you so much to you and Annabelle. Bonjour, Annabelle, in your absence. Merci, Chimdi, for having me out. It's really an honor to be here with this community, and thank you to Igor and, uh, Ilija, I'm pronouncing that wrong probably, for the great presentations and for s- sharing the stage with me.

1:18:39 Um, so with that, I'm going to share.

1:18:43 And let's see. Hopefully you guys can see my screen. Prasann, give me the… Let me know 'cause I can't see-

1:18:49 Prasann Prem: Yeah. We can.

1:18:50 Angela Henderson: Okay. Awesome. Okay. So I'm gonna get started with a quick quote, and the quote is, "Honestly, I don't know much about data or dashboards, but if my real dashboard looked like this, I'd drive straight into a ditch." And that comment came from a long-haul truck driver after seeing one of my early dashboard examples, and that's fair. If you look at this example, it's terrible.

1:19:14 The long-haul truck driver was actually my cousin. And every now and then, we catch up and talk about the routes, deliveries, and how crazy diesel fuel prices have been lately, not to mention regular fuel prices. But some time ago, he asked me what I actually did for a living, so I showed him one of my early career executive dashboards. Now, this is a fictional one, but my dashboard was almost just as bad.

1:19:37 It had multiple worksheet, but basically, um, a truck driver, who was my cousin, said, um, basically if this was his dashboard, he'd drive, drive his truck into a ditch.

1:19:47 This is something that I would have created early in my career. Um, it's dashboard soup. It's got all of these charts, and colors, and things, and it's not helpful at all. Um, dual axis charts, basically everything. And honestly, he's right. In his truck, he needs three simple things: Is there a problem? What's causing it? And what should he do right now?

1:20:09 And that made me think about the kind of dashboards I never want to build again. Like, as data developers, we, we are used to treating reporting like an open-ended puzzle. We put every single metric on the screen and tell the user to go hunt for their own insights. But our stakeholders don't want to do all that digging anymore. They want something that feels like an app.

1:20:29 So for this project, I built a fictional command fleet dashboard around a simple real world problem, and that's fuel waste is expensive, dispatchers are busy, and nobody has the time to dig through a wall of charts during a hectic work day. I wanted the dashboard to help someone spot their issues, understand what is causing it, and decide what to do next.

1:20:53 All right.

1:20:54 Next slide. Are you seeing the next slide?

1:20:58 Yeah. It says, "Open it like Netflix, not a spreadsheet."

1:21:03 Prasann Prem: Yes, we can see it, Angela.

1:21:04 Angela Henderson: Okay, great. Okay. So think about it. When you open Netflix or Spotify, you're not looking at a giant messy chart collage. You're looking at something simple that helps you decide what to do next.

1:21:16 That's the feeling I wanted this dashboard to have. Clean, easy to follow, and focused on what the user needs in the moment. So that's why I built this dashboard completely in native Tableau, Tableau Public actually, using basic design principles. I focused on the human side of the data, thinking about, um, what a real dispatcher doing a crazy work day, um, and, uh, would do to organize their screen to guide them through a simple three-step loop.

1:21:43 Um, problem, and that would be problem, cause, and action. When we start looking at fleet data, we want the dispatch team to understand the big picture right away before we get into the weeds. And across the top, you'll see we have high level executive KPIs.

1:21:59 These quickly show leadership our active financial risk and leakage totals for the month.

1:22:05 Under those numbers, you're going to see the Chicago skyline. It's not just a pretty picture I decided thr- to throw in there. I think of this kind of like the opening screen of an app, like Netflix or Spotify. You're not dropped straight into every setting, filter, and detail. You get a clean landing page that helps you understand where you are and what to do next.

1:22:26 For our dispatchers, that skyline does the same job. It anchors the dashboard in Chicago, which is their home, the actual market we're watching, and reminds the user that these numbers are tied to real routes, real streets, and real decisions before we move into the deeper charts and maps.

1:22:51 All right. Okay.

1:22:57 So, uh, now we are digging into what's actually causing the waste.

1:23:02 By filtering out the normal trips and focusing only on active fuel anomalies, the bar chart puts the highest risk drivers right at the top where we can miss them. So if you notice, um, now we're drilling down into cause. We were at problem by those buttons on the left. Now we're at cause.

1:23:21 Um, there's a little n- native Tableau limitation I want to call out, though.

1:23:26 Usually, when we hover over a bar, Tableau fades everything else to gray, or even that white look, that, that shadow look that you don't want to see. And even when the bars and numbers react, the labels usually don't. I call that the gray out trap because it breaks the clean dark streaming style look I was going for.

1:23:45 So to get around that, I use a design trick that the data fam affectionately calls ghost highlighting. So instead of letting Tableau decide what fades, I control the color behavior myself. I use a kind of a true false hover calculation for the bar labels, the numbers, and the map marks. Then I kind of do a max if wrapper on the heat map color shelf so the whole view reacts together.

1:24:09 So what you're seeing on loop here is me selecting the drivers by bar. Each bar res- represents a different driver ID.

1:24:18 It then highlights a little heat map down below, and tells you which route the driver is on and which days they drive.

1:24:25 In the screen clip, watch what happens if we hover over driver 61.

1:24:30 The bar, the driver ID, and the dollar values all light up in that bright warning orange at the same time. The heat map, other, the other heat map labels stay clean, but we can immediately see that the route covers lines 301 and 302, and they're, um, off on Mondays and Thursdays. They don't drive on Mondays and Thursdays. So those are the insights you can get at a glance right away, um, when…

1:24:55 and still you're looking at your KPIs, which are dynamic.

1:25:00 All right, so now we're going to move on to the next slide.

1:25:11 Okay, now we're going to expose our, our map.

1:25:16 Notice the map area on the right. We're using the drive time function, uh, native to Tableau, uh, very new to Tableau, actually. So instead of letting a big map take over the analysis space when we don't need it, I built a simple hide shell toggle into the container. If you look to the left of the bar chart, you'll see a little small globe, and that globe button, it kind of blends with our dark theme, so we can collapse the map when we want to focus just on the driver trends, or open it back up when we need the full map view.

1:25:49 Once we ho- hover over a driver, an orange truck icon appears on the map showing that this is a heavy truck, a delivery box, or a small van actively running in that zone. And the way we can tell that is I have a little cheat sheet up at the top there that's kind of telling me which driver we're headed, hovered over in the top right next to the header and the title, and it tells you, like, heavy truck, delivery box, small van.

1:26:16 Um, or you can actually see that the icon, I love to use iconography, so you can see the actual picture of the trucks that we're using there.

1:26:24 So now we've moved from a high level problem into a connected view of the cause, but we're not done quite yet.

1:26:33 All right.

1:26:36 Now that we know the cause, we have to take action and stop the fuel waste on real roads. This is not just a dashboard sitting there anymore. It starts acting more like a working business tool.

1:26:48 To make that happen, I use a Tableau layout trick. I nested DZVs, which is dynamic zone, um, visibility inside other DZVs.

1:26:58 And you'll first see this in the top right corner that I kind of mentioned, where the vehicle type only appears when I hover over a driver bar.

1:27:06 You'll see it again when we move into the next action step, which you'll see is highlighted now to the left on my button there. You see we're in the next steps. Um, the main layout opens right in the middle, and a terminal log sheet pops into the space between the bars in the map. Wow.

1:27:23 Inside that new view, there's another layer of hidden visibility rules turning on at the same time.

1:27:31 Um, my screen went blank. Did it go blank for you guys?

1:27:36 No. Okay. It's just… All right, so let's keep it going. So, but inside that new view, there's another layer of hidden visibility. The middle panel, it's going to go on loop, and you'll see it again. The middle panel reads the raw data and turns it into a plain English message. It explains the fuel waste triggers and the protocol we should follow.

1:27:57 At the bottom of the log, it also flags the exact vehicle type, showing that that driver, it's sixty-one or forty-two, is running a heavy truck that's consuming seven hundred and five gallons of fuel.

1:28:08 And at this point, you may notice a lot of fuel gauges. Sorry, I became a little obsessed with fuel gauges along the way.

1:28:16 Uh, one is just a static image in the title that I gave, that I drew. The other is a glorified horseshoe I figured out with trigonometry, which took me a few days to get right, and it still looks pretty basic and mid. And the last one is the live Data viz version, shout out to Tristan, which is polished, clean, and took me about thirty seconds to do.

1:28:37 I wish I had remembered that when I did all that trigonometry, and it took me several days. But you live life forward, and you learn, you know, you learn backwards. So, um, I left them all in to kind of show that it really depends on what you need and how much flexibility you want. There are pros and cons, and you can see the branding on the viz extension because I'm using the free version, and it's a quick licensing note.

1:29:00 A lot of viz extensions work free in Tableau Public or the desktop, but some need, um, a paid license before they will render in Tableau Cloud. Either way, the second your mouse touches the row, that needle moves forward and frames the fuel leak percentage of total gallons or, uh, percentage right on cue.

1:29:22 Now, by making the labels, the matrix blocks, the fuel gauges, and map marks all follow that same ghost highlighting logic, the whole dashboard feels connected.

1:29:32 The dispatcher's eyes know exactly where to go, and a regular dashboard starts to feel more like a responsive software, too.

1:29:39 What you're seeing last flashing across the screen is how we can, um, monitor those in the dispatch center, the fleet availability and routing li- limits in real time.

1:29:51 So here's how that map fits into our overall narrative of tracking driver fuel efficiency.

1:29:57 When an emergency comes up, a dispatcher can hover over a driver bar, and the system isolates that backup truck on the map, calculates its road distance from the downtown depot, and creates a tactical alert based on that vehicle speed profile. So if you see me there dragging the truck and, uh, lassoing it, you'll see it's creating that drive time view and all those little trucks around it.

1:30:23 It's saying where all of the backup trucks are, and it's telling you where the trucks are, how long it'll take to get to our truck that's having an issue, and where they are in the neighborhood.

1:30:35 All right. So why are we doing all this? We're lassoing these backup trucks across the city to answer a real dispatch question. How fast can a support driver swap loads with the faulty truck, bring that truck back to headquarters, and dr- and stop the waste?

1:30:51 This gives the command team the routing numbers they need without guessing.

1:30:56 And at the bottom left of the control panel, you'll see a new button that links directly to our live cloud alert doc in Tableau Pulse, connecting this desktop view to the executive web view. So if you see this recording on loop, I'm recording like everyone else because Tableau is kind of unpredictable live, but eventually it is gonna get to that Pulse link.

1:31:18 And when that Pulse link is gonna take you directly to my Tableau Pulse. And it will open up the Tableau Pulse in the cloud. You'll see it's clicking the Tableau Pulse. It's going to the cloud right now, and this is how the screen kind of looks after you've set up all the metrics in Tableau Pulse.

1:31:35 Um, and you can set it up and to ask it plain language, natural language questions. It'll give you, depending on the dimensions, uh, and metrics you put in there, there's kind of a, a lot of things you can do with Tableau Pulse. It's really cool.

1:31:50 So I'm gonna s- uh, pause there and see if anyone has any questions. Prasanth, can you help me out with that?

1:31:58 I'm kind of going through this really fast.

1:32:02 No. Okay.

1:32:03 Prasann Prem: I think we can move ahead.

1:32:04 Angela Henderson: Okay.

1:32:04 Prasann Prem: Yeah, just checking. Okay.

1:32:05 Angela Henderson: Okay. All right. Thank you. So, um, now to close the loop, we move this data back to executive leadership through the cloud.

1:32:14 Once a business problem is this clear, the old way of working starts to feel expensive, not just slow.

1:32:21 So we start by uploading the Tableau workbook into our Tableau Cloud Sandbox and saving that data source as a published data source, which is important for Pulse.

1:32:31 Then we bring that data source into Tableau Pulse and choose the dimension, metrics, and, and filters we want, as I just noted, and from there it fills the total fleet financial leak metric card with a clear historical trend line.

1:32:44 Even with a cl- a clean dashboard, we still do not expect busy executives to log into full desktop views during a hectic week, so the Tableau Pulse to Slack integration pushes those metric updates into leadership Slack channels on a regular cycle. This is a picture of my actual phone here. Uh, luckily the… it was at 100% battery. But, um, it gives us that natural language…

1:33:10 Let's see. Go back a second. It gives us that natural language, uh, questions from the Pulse view that we just saw, and we can click into our actual Pulse, um, our actual Slack, um, uh, interface there for more detail.

1:33:25 So for anyone wondering how to get started, I built and tested this layout using Tableau Public and the Tableau Cloud Sandbox.

1:33:34 For me, data is about building community and mentoring people.

1:33:38 I love sharing these resources because they are public, so anyone in this room can start practicing these design ideas tonight, and you can also check out my other public visits. I'm going to upload this one soon, like my nursing home proximity to hospitals map using that drive time feature, to see how the same spatial ideas can show up in healthcare.

1:33:58 Um, as, as someone who loves part, being part of the data fam, however, and showing what Tableau can really do, I also want to mention that the full enterprise version of Tableau Cloud has an amazing set of advanced features.

1:34:11 It gives organizations strong security controls, deep cloud data pipelines, and smart automation like Tableau Next to help scale these kinds of applications across global operations.

1:34:23 When you go back to your desk tomorrow, you want to stop building generic places to dump data. Start building ex- experiences that actually guide people. Give your users the same clarity my cousin needs on the highway, highway, excuse me. Lead with the problem, show with the cause, and then drive the action.

1:34:43 Um, thank you very much for your time today, and I'll open it up for questions.

1:34:54 Prasann Prem: Thank you, Angela, for such a great, uh, presentation. Also hitting on the point where you need to actually focus on. We do have one question. "How did you implement the responsive narrative? Is the text static or dynamic?" It's in the chat. Yes. "Based on selections in the dashboard."

1:35:14 Angela Henderson: That's a great question. So, um, uh, some of it is done with calculations. I started out with these lengthy string calculations, which everybody knows takes up so much bandwidth on Tableau. You're, it makes everything slow. So I had to rethink that. It was kind of like, if it's this, if you select driver 031, then make it say, "Go to this dispatch," or something like that.

1:35:36 If-then types of things. I had to change that around though because, like I said, it was taking up, it was making my dashboard slow. So I put it, I was using an Excel spreadsheet that I, that I made up. I got some data from, you know, AI, some fake data, and I said, "I need to make this faster." So I actually put answers to every possible combination and question in my data source, um, so it would work out, and I made a column in Excel for the, uh, if the, the log questions and, you know, it fits in with the region and so forth.

1:36:07 That's already in the data source. So when you click on the driver ID, and you have a sheet dedicated to that text, um, then all you need to do is pull them over into those dimensions onto your marks and pull that large calculation if you need it. If some, some of it's a little bit of a cal. If it's this, then show the truck or whatever. Just pull it onto your marks, and then when you click a bar on the driver ID, and since they're in your dimensions and you do a set action, not a set action, a filter action or parameter action filter, and you click on the bars, it'll automatically pull up that correct, uh, log for each one.

1:36:42 Also, a little bit of DZV, like I said, to keep it hidden.

1:36:46 Was that the question? Did I answer that okay?

1:36:50 Prasann Prem: Yeah.

1:36:50 Angela Henderson: Is that what they were asking, Prasanth, here, that you got?

1:36:52 Prasann Prem: Yes, yes.

1:36:53 Angela Henderson: Okay.

1:36:53 Prasann Prem: And, uh, thank, thank you so much for sharing that, the process, as to as how you're doing it.

1:37:00 And, um-

1:37:00 Angela Henderson: No, no worries. Let me turn my camera back on.

1:37:04 Prasann Prem: And thank you so much everyone in the audience, everyone having this whole interaction. It, we always love that at Analytics Tag, that all our participants and everyone are always here, like, communicating, having interaction. So it's interesting. Uh, Lisa has a question.

1:37:24 Angela Henderson: Sure.

1:37:28 Prasann Prem: He is back in attendee, so maybe I'll just put him in presenter mode and he can ask himself.

1:37:34 Angela Henderson: Yeah.

1:37:40 Prasann Prem: It's a, it's a joke, Soli- Joe, yeah.

1:37:50 Angela Henderson: Hi. Are you on mute? You're on mute.

1:37:53 Prasann Prem: No, we can't hear you.

1:37:58 Angela Henderson: Uh, he had a question. He's on, uh, uh, he's on mute.

1:38:04 Lisa, you're on… Lisa, you're on mute.

1:38:08 There you go.

1:38:13 I still can't hear you. Put it in the chat. We'll answer it. And while he's doing that, while he's doing that, I just wanna give a shameless plug that, um, I will be leading the Chicago TUG, um, going forward. We just started that. I'm, I'm really honored to be able to do that. So we are right now as we speak planning our first meeting. Um, we're gonna be doing that in the Salesforce Tower.

1:38:34 Hopefully they'll let us in there, and, uh, more to come on that. So anyone interested in that, please join us. Whether you're in Chicago or wherever you are in the world, we are gonna have some star-studded guests, uh, entertainment. No, with no entertainment, but, um, we'll… It'll be nice. Thank you.

1:38:55 Ilija Stojić: Hi.

1:38:55 Angela Henderson: Hi.

1:38:55 Prasann Prem: Now we can hear you.

1:38:56 Ilija Stojić: Yeah.

1:38:57 Prasann Prem: Yes.

1:38:57 Ilija Stojić: Yeah. I…

1:39:01 Irony, but Bevy muted me while I was saying I was also having technical problems. It went four for four today with each of us, and it just decided to cut out my audio as well.

1:39:15 Angela Henderson: Oh.

1:39:15 Ilija Stojić: Um, my, my not serious question, Angela, was how did you respond or what did you do to your cousin after they said they'd drive their truck into a ditch if it was your dashboard?

1:39:30 Angela Henderson: We joke with each other all the time, so we just laughed. I, I didn't do any harm. No cousins were harmed during this process.

1:39:38 It just, it always makes me think a lot, you know, when they, people ask me what they do, what I do, and then if they critique it not knowing anything about data, it makes you wonder, you know? It makes you really look at yourself and, uh, yeah, do a thorough in- overview of what you're doing, so.

1:39:57 Ilija Stojić: Thank you.

1:39:58 Angela Henderson: Thank you. That's funny.

1:40:01 Prasann Prem: Okay. So I think lots of good things, lots of presentation, but also lots of, uh, new TUGs coming up, so that's, that's helpful for everyone. And, uh, please join Chicago Automotive, and I also saw Pittsburgh.

1:40:18 These are the new TUGs that are coming on the space, so please look out for them and do join the events.

1:40:25 Again, we'll meet next month with new speakers.

1:40:30 And so thank you so much. Uh, if you are, it's late. For me, it's early mor- not early. It's almost now 2:00 PM.

1:40:38 Totally different from, uh, when I'm in Dubai, it's 8:00 PM for me-… or late basically. But today it's like morning hours. I went for a run in the morning because I'm on a solo trip right now and, uh, and then came back and then I was like, "Oh, it's at 12, so I have to reach somehow." Then I got back.

1:41:00 Well done. So thank you so much everyone. Have a good day, good night, wherever you are, so.

1:41:06 Angela Henderson: Thank you so much. Thank you all.

1:41:08 Ilija Stojić: Thank you.

1:41:08 Angela Henderson: Have a great day.

1:41:09 Prasann Prem: Thank you.

1:41:10 Angela Henderson: Thanks for tuning in.

1:41:11 Prasann Prem: Bye-bye.

1:41:12 Ilija Stojić: Bye, guys.

1:41:12 Angela Henderson: Bye-bye.

1:41:12 Ilija Stojić: Take care.

Leave a comment

Your email address will not be published. Required fields are marked *