Data + Women US West Coast 20 Oct 2026
D+W: US West Coast with Celia Fryar
53 min 9,560 words transcribed
Meeting: D+W: US West Coast with Celia Fryar
Summary
Celia Fryar, three-time Tableau Ambassador, co-lead of the Austin and San Francisco TUGs and training, enablement and AI lead at XeoMatrix, argues that great dashboards still go unopened, and that the fix is to push insights to where people already work. She walks through Tableau's push options from subscriptions and data-driven alerts to Pulse, Tableau Next in Slack and MCP, demos Claude drafting a Tableau Pulse insight brief and posting it to Slack only after her approval, and shows how a well-documented semantic model and business preferences keep AI answers trustworthy. She closes with how saved Claude skills, memories and a style guide now carry her technical voice, followed by a Q&A on trust in AI and where Tableau should end and an external LLM begin.
Key points
- Dashboards have a lifespan: logins and security hops are often enough to stop people opening them, while the KPIs they track keep accumulating regardless.
- Push analytics sends the insight into the inbox, Slack or Teams, and every option respects your governance down to row-level security.
- Subscriptions, data-driven alerts, Pulse and Tableau Next in Slack range in sophistication; be selective to avoid alert fatigue.
- AI agents need guardrails: document or hide every field, encode business preferences, and define vague words like 'high' or the agent will decide for you.
- Claude can query Tableau through MCP and post to Slack, ideally pausing for approval and labelling what it posted for an audit trail.
- Saved skills, memories and a style guide let Claude keep domain-specific Tableau language intact instead of smoothing it away.
- Let Tableau hold the governed semantic model as the source of truth, and use Claude for quick one-off questions.
Jump to a chapter (19)
- 0:00 Welcome to Data + Women US West Coast
- 1:23 Introducing Celia Fryar
- 3:36 Celia's crew, and why Data + Women matters
- 4:50 If you build it, who will come?
- 5:58 Push analytics: send it to them
- 9:02 Governance and row-level security
- 9:42 Subscriptions and data-driven alerts
- 12:12 Tableau Pulse and enhanced Q&A
- 13:57 Agents, semantic models and business preferences
- 17:21 Tableau Next, Claude and Slack
- 18:19 MCP and the REST API
- 19:08 Demo: Claude posts a Pulse brief to Slack
- 22:40 Demo: documenting the Tableau Next semantic model
- 28:21 How Celia started building Claude skills
- 32:05 Saved skills, memories and a style guide
- 38:21 What's coming: Tableau Studio and proactive intelligence
- 41:35 Q&A: trusting AI in the workflow
- 47:18 Q&A: Tableau as the governed layer, Claude for one-offs
- 50:15 Swag giveaway and close
Transcript
0:00 Tiania Chan: Welcome to Data + Women: US West Coast.
0:04 Um, thank you so much for everybody who participated in the survey to determine what our topic was for today, and to anybody new, thank you for joining us.
0:16 So if you haven't yet, the Vizibly link is right there, and you can always look at it later.
0:23 We came up with this in last year, I think now, so it's a year old-ish, right?
0:33 Confirmed? Yes. We are a year old.
0:37 Um, uh, and so today's topic is with Celia Fryar, If You Won't Build It, They Won't Come. Um, and I'll introduce her really quick in a bit. And then our next event is gonna be with our very own Michelle, um, who's gonna be talking about Vizibly, I believe. So that'll be on December 8th.
1:02 Um, we want to promote women and speakers, data dashboards, and challenges, and we wanted to make it at a time that was convenient for us on the West Coast 'cause we had seen other Data + Women, and we were like, "Okay, well, this is, this own- does not always work for us." So 1:00 PM Tuesdays.
1:18 Um, and of course allies are welcome.
1:21 Um, okay.
1:23 So have a bit of a blurb here for Celia. She's been in tech for 25 years, 12 years as a Tableau user. That's a long time.
1:33 Uh, 10 years as a educator, um, four years as gen AI user and educator. Cool.
1:41 Uh, three times Tableau Ambassador, uh, and she co-leads for Austin and SF TUG, so maybe we can all collaborate at some point.
1:51 No pressure, of course. Um, she spoke at the conferences last year and this year and Dreamin' Data in-
2:01 Celia Fryar: It's a Salesforce Chicago thing.
2:03 Tiania Chan: Oh, okay. Sorry. I was like dreaming, Dreaming in Data. Got it.
2:07 Celia Fryar: Yeah. It's Dreamin' in Data. It's, it… I think it's a new series that they've started putting up, but it's kind of a crossover between Tableau and Salesforce to some degree. It's been interesting to see.
2:18 Tiania Chan: Oh, that's super cool.
2:19 Celia Fryar: Yeah. Yeah.
2:19 Tiania Chan: Okay.
2:19 Celia Fryar: It's a good, good place to meet new people is what I was observing.
2:23 Tiania Chan: Yeah. Thank you for telling us-… 'cause I wasn't aware. I was like, "Oh, okay." Um, uh, adjunct professor at U- University of San Francisco, um, training enableant- enablement and AI lead at XeoMatrix, and then she's a mom to three humans, one of us, or one of them is here, I believe. A precocious mini golden Cavalier named Bella, and she's also a scuba diver, traveler, and gardener.
2:55 And then just really quick what
3:01 what we are talking about, um, If You Build It, They Won't Come: Push Analytics that people… that Meet People Where They Are.
3:10 Um, so we were seeing that great dashboards still go unopened, and this session moves from push analytics to proactive intelligence featuring Tableau and Claude, um, to help bring insights to people where they work, and it'll be with a live demo featuring Claude. So Celia, I will let you take it away.
3:37 Celia Fryar: Thank you very much.
3:38 Okay, so I do have a picture of my, of my crew and precocious, um, puppy 'cause, you know, she may actually make an appearance just at the whatever sh- time period she decides to make an appearance, so, um, but we are all divers.
3:54 We have two dive masters in the family, and that has really helped us with the safety profile of what we do.
3:59 But, um, yeah, these guys are, are… Man, they make life a lot of fun, so.
4:04 Um, so thank you for letting me jump in with you guys today. I, I gotta tell you, it means so much to me that there is Data + Women user groups, uh, and just having started in the, uh, I don't even wanna tell you what year I started in tech, but, um, a while back, let's go with that. And, um, you know, just, I'm just thrilled at all the advancements that have come and that women have been able to step into in leadership roles and the way that women are represented broadly in the data world.
4:37 I love that. It's incredible. And, um, yes, so, um, I could talk about that for a long time, but I won't 'cause I came to talk about Claude and Tableau, so why don't we do that instead. Um, this has, uh, definitely been on the, the, the top of a lot of my clients', um, and colleagues', uh, conversation list, uh, because it seems like a lot of the things we have built…
5:02 First of all, before Claude became such a, and OpenAI made such a splash, we really were already starting to see that, you know, dashboards kind of had a lifespan. And, and, um, this is supposed to be kind of a play on words on that movie that I don't think I ever saw, but it definitely became a meme, meme in terms of like if you build it, they will come, and the reality of it is who will come and you're not sure because, uh, you wanna be sure that…
5:25 And, and a lot of folks that I work with were putting counters on dashboards to, to keep them fresh and to see what's happening with this. And, uh, a group I was working with this last, um, week was talking about how that their, their Tableau Cloud instance, which is relatively new and pretty robust, that they're, like, working hard to get really good data sources stood up on, that their users just wanna download their data into Excel and make Claude dashboards.
5:52 So, like, there's a new wave of, like, um, things to navigate. So, um, push analytics, I think is in line with what Tableau announced at Dreamforce in terms of proactive intelligence and, and it's just that theme song of, that actually is present to some degree all the way through the platform of, like- We don't necessarily need people to come, we can send it to them.
6:15 And so I wanted to kind of highlight the, the thread of opportunity there because a lot of us have built spectacular dashboards. Man, if you haven't looked at Ann's profile, and I'm sorry I don't know your, the other two of you's profiles. But Ann does some artful, spectacular Tableau dashboards. And I have really enjoyed, um, seeing what she publishes and, you know.
6:37 But for people to sometimes, the, the obstacles involved in logging in and, you know, whatever levels of security they have to navigate to get to the information, sometimes that's enough of a blocker to keep them from being as proactive with that, and it's kind of a, a shame and a waste and a risk all at the same time. Because a lot of us around, you know, are being also measured by that data.
7:00 So to ignore the KPIs doesn't even make them… doesn't affect them at all. They don't care, they continue to accumulate, right? So to send those insights to, um, to folks and put it in their path, in their flow of work, it has become a resounding theme song in a lot of places. So, you know, it's, it's, uh, it's everything from, uh, maybe I'll just run through a series of things inside of Tableau, but the, the ultimate would be to put it in their inbox, in their, in their workflow inside of, uh…
7:31 If they're in, if they live and breathe in Salesforce, integrate it there. If they live and breathe in Slack. I'm amazed at how my, my work environment, man, I, I could… I, I don't use my phone as much except for to check Slack, 'cause we d- very much are integrated with Slack, and then there's other environments where Teams is serving that role.
7:49 So to send it to folks, to make it in the flow of work, that's one of the things I think that can help smooth the path on people taking advantage of the things that we stand up for and might curate for them. So, um, we have a, a list of things that range in the level of sophistication, of course, through the Tableau platforms, but some of them are, are kind of underappreciated in my book.
8:10 So we have subscriptions. On Tableau Cloud, we have subscriptions and alerts and, you know, those can be set up with a level of sophistication in terms of, like, um, intervals and such. We have data-driven alerts that come off of, um, an assortment of tools. And then Pulse, speaking of p- Pulse metrics, and then actually metrics in Tableau Next. Um, then, then Tableau Next with the way that it can communicate very easily with Slack, where we can ship those, um, dashboards and Pulse metrics, which are just called metrics there, um, easily out to Slack, and I have several examples to share and show.
8:48 And then, um, the ultimate of course is when we have the next level of MCP enabled, and I almost got it, got nailed today, but I don't have quite the right admin. But the goal is to make it interactive in those next places. Now, fortunately, before anybody, like, it… I mean, the concern that usually comes up right about now is, "Well, what about, um, the governance for security, and, and, and what if that's out loose?" Well, if it's out loose in an email, then yes, if your email's not protected and underneath the right kind of protections.
9:19 But it's only the data that you have access to. So in all of these cases, everything is being respecting the governance down to row level security of what you have access to. So just wanna take that concern off the table that none of this is gonna, uh, create wreckage in that regard. But, um, you know, with that caveat, we, f- put it in the flow of work for folk, folks really can be a, uh, a win.
9:42 So subscriptions is gonna be that thing inside of Tableau Cloud. It's actually just on the frame of the Tableau Cloud, not even inside of any dashboard, and its ability to do the alerts, where we have a threshold. The caveat there is you have to have a number line that you can click on, and then you can set that threshold and have it go be shipped to a user, either with an image or a PDF.
10:05 Um, and then same thing for, um, actual subscriptions, is more like what that's called right there, and it's, uh, gonna be the, the image from the dashboard. And one of my first clients when I started working at Xeo was a big retailer, s- um, kind, uh, coast to coast retailer, and they had store locations where they, for them to see their KPIs, they had to log on, and they also used that same computer as a cash register.
10:30 So, you know, imagine the contention for resource that was happening there. So we did set up a lot of subscriptions, and the various levels of, um, security applied well. And so folks were being measured by l- their sales or their quotas, their whatever it was. Um, the right level to the right person was being applied and being respected, and was really very, um, useful and practical in the way that we were being able to roll out that implementation.
10:57 So this is not at all hard to do and very easy. So, um, these alerts is gonna, when it hits that threshold. Now, one of the things that's coming that this also instantly makes me think of, and I have a different slide for, is we have had all of this in here where we click on a, on a, a threshold and then it sets an alert, and it's kind of a little bit of a manual or hard-coded in my mind, um, process.
11:22 Nevertheless, what's coming now is, uh, a Slack inspector that's actually an app inside of Tableau Next, and it's gonna be using that MCP permission set that's set underneath those two products to be able to, um, watch kind of and be proactive intelligence. And watch for thresholds or concerns or outliers, whatever you condition it to be watching for, and be a little bit more in- intelligent in the way that it would alert us on things.
11:50 I think, I think the risk with some of this is, is there's a balance between sending people a lot of things and, and not enough. But, but we all, like, tend to end on the side of, uh, alert fatigue or notification fatigue. So being selective about what's sent and shared that way is also gonna matter in, as we implement these things.
12:12 So Pulse. I kind of became, uh, on a campaign to evangelize Pulse metrics after I got home from TC this year because- That team behind it has been so responsive and so proactive about BI-weekly releases, and so, uh, and trying to enrich the feature set for the, the Pulse metrics that lives in Tableau Cloud.
12:36 It's remarkable what all is possible there, and, um, I do have a couple of examples pulled up in the background I'll show you in a second. But there's… If we quickly send those guys up, they are… Sure they show the KPI, but they don't have nearly the richness and value that they can. If you just spend another couple minutes and you, um, let them go ahead and show, um, get, get a quick link to go back to where you came from, and what other dimensions would it need to be sliced by?
13:04 Noticing that the Tableau dashboards with any kind of filter, if you put that dimension as a, an optional n- uh, slice and dicer in the, in the metric setup, then it's responsive to your, uh, dashboard filter. So even before we get into all that's capable, possible with our agents, which I've been spending, like, 90% of my time talking about nowadays is the agentic side of Tableau Cloud and Next.
13:30 But Pulse is phenomenal, and in the summer release, they have implemented a enhanced Q&A, is the language that goes around it. It's actually ChatGPT 5.2 in the background, and so it's much more robust in terms of what it's able to handle in terms of questions. So it's… If you've not looked at Pulse metrics in a while, it's probably worth a look because it actually has really, really improved.
13:54 So, um, then we get into what's possible with the agent. And so I have, um, like I said, 90% of my time right now goes into the agentic side of the house and what's possible, how we conditioned not only the, the context inside of our dashboard, um, configuration to help that agent give good answers, but also then what do we do on the data modeling side.
14:20 So you gotta have a good semantic model and/or hide things that are confusing, reduce ambiguity or eliminate it if at all possible, and use that business preference file to connect the dots on the, the last mile is what I like to call it, even though it's way more essential than that.
14:36 And for those of you that haven't had a chance to use Tableau Cloud's version of this is we can put a business preference file in Next with each dashboard. So, a- and it's a little bit smaller than what it is in Next, but it's still eight thousand characters, up to 50, uh, instructions, lines of instructions. And it's that way to connect the dots between jargon and thresholds that may be, like, policy violations or policy standards.
15:02 It's actually really, really, uh, come a long way. And I think that the way that they're bringing a- the agent presence on board with Tableau Cloud may breathe new life into it in a way that, um, may, may be unexpected. But I think it's really gonna be super helpful. So, um, then over on the metrics side in Tableau, um, sorry, in, in, uh, Tableau Next, we have just a really robust situation there.
15:28 I'm gonna… I was mentioning d- business preferences over in Cloud. I'm gonna just reiterate that over in, um, Tableau Next.
15:36 It's increased its capabilities up to being, um, unlimited numbers of individual lines, but the limit of 30,000 characters, which is way more than what you probably need. And, and it, instead of addressing one dashboard at a time, it speaks to the data model. So the semantic model is associated with one of those larger files of business preferences, and, you know, so you get a chance to address it kind of like more like on a one and done sort of setup.
16:02 But it's also, uh, really important to, like, I, I use Claude to validate or Codex to validate those rules slots to make sure that in the, as the list grows, we don't have things that are countermanding each other because then the agent, you know, struggles. And, and then it covers different points of view. Like, we always start with, like, the guidelines of bounding boxes, of what do you want it to do when there's something, you know, if it doesn't know, or, like, is it okay for it to make assumptions?
16:30 I was ask- I was using just a demo data set in, um,
16:35 Tableau Cloud and trying to demo this the other day, talking about discounting, and I asked it about high discounts. Well, I hadn't given it a threshold what high meant, and so it just decided that anything over 20% was high. I mean, like, who, what? So, you know, we need to, like, spell it out and not use vague language in a way that's, um, just gonna invite it to either make up stuff or…
16:59 And also, you know, it's not like just, uh, pulling it out of thin air. There'll be some extrapolation or some math it'll do in the background. It's more like a predictive algorithm, but it's still gonna feel to your stakeholder like, uh, what everybody likes to call hallucinations. So, um, you know, they are trained to give us answers, so it's really trying, and we just need to give it the guardrails to help it give good answers.
17:20 So one of the things I very much wanted to show you guys was how the Tableau Next and Tableau Cloud and Claude can communicate to get things to Slack. And so, uh, we're aim- we're eventually gonna land where we have, um, Tableau Next producing live assets over in Slack, and it's interacting. I've seen that embedded in various people's machines. Um, I just don't have it to show it to you guys today, but just know that that's possible.
17:49 Inspector's coming. I mentioned this. Inspector's that proactive intelligence that's gonna watch over thresholds and more of that. Like, we're gonna give it some guidance, and it's gonna be able to pr- reach out and say, "Hey, this is a problem." And more than that, uh, Salesforce has got it, uh, configured to where it then can take action. If you set up what's called a flow, so you, there's actions you can have it trigger as it alerts you.
18:13 It can go ahead and take care of that should that fit the, the situation. So, um, so MCP. So MCP is, as we all know, is an API, right? And it's one of a few. I have, I had some really good success with also a, another, um- REST API that, uh, Salesforce, uh, uses u- made available as a plugin of VS Code, where I can load my data models and u-upload all descriptions and tables and such directly into Tableau Next without spending the time that it requires otherwise.
18:48 MCP is another way for us to get really protective and safe read-only kind of answers out of, uh, Tableau Next, uh, any of the Tableau data stores. So on Cloud and/or Next, that works really well. And then, so to ship insights or knowledge back out is entirely doable these days. So I'd like to show you, um, an example with my Claude. See if I can make sure that we don't have that in the background, but we have it in the foreground.
19:16 There we go. Okay. So we have a Slack channel here. And this is the kind of things that I've already been asking it to do.
19:23 And so we are in the mode here of…
19:29 Get my questions back up. Okay, this was gonna be to ask it to find these order trends and, um, you know, let's see if I can copy and paste that. That would be spectacular, right?
19:40 And then we'll, um, tell it over in Claude. So Claude has gotten, um…
19:47 You know, you have to have it connected to your cloud environment, but, um, it's going to always do a couple things. So it's gonna require you to be connected and then need authorization.
20:01 And-
20:02 Ann Stolzman: We're not seeing your Claude. We just see your Slack.
20:04 Celia Fryar: Okay. Yeah, I'm trying to get…
20:07 There we go.
20:09 Another F11 is what I needed to do there. So, okay, so we, um…
20:15 So I've given it this request and, um…
20:19 Oh, and I didn't, not get the whole sentence in here, so that's a problem. Let me go back to this other, so I have the whole question.
20:27 Yeah, there we go. All right. So we're, we're gonna have a question for it. And in this case, I wanted just to do this, um, summary information.
20:36 So, um, um, in one second.
20:41 There we go.
20:45 Right. And Claude is gonna be, um… If you can override some of the permissions and stuff, but you probably want it to, to ask you a-and as you for permission, right? So generally speaking, when I ask it a question, now it's reaching out to my Tableau Cloud, which is also, um, logged in here in the background. Um, but wouldn't necessarily need to be logged in, but I definitely would have had to give it authorization to get started.
21:12 And then, um, you know, I've given it some instructions already, as you can see, on what kind of style of output I wanted to have. Um, and then while it's working on it, let me also just scroll up and show you. These are things that came, um, out of Tableau Next. So I, I got a, an opportunity pipeline that's a metric. This is one of my dashboards, another dashboard.
21:31 So these things I can share to Slack and it– because of the MCP connection, it works very well. Now, notice it's given me, it's drawn it up, and it's asked me if I'm ready to post it based on the, the, um, the draft it's given to me here.
21:47 And based on our agreements about how we wanted this to be formatted, it's, um, I wanted it to also not ch– It's posting as me, right? Because it's my account logged in this way. And wherever I tell it to send these messages, it will do that. But I also like the fact that it's saying, "Posted by Claude." I, I don't know. I just like the accountability and the, and the audit trail.
22:10 So that's one of the things that is, um, happening here. And, um, I feel like that's, that is an appropriate way to implement all this. So, um, another example of this would be, um, you know, I mean, there's a lot of ways we could do this. So let me show you from the, the Tableau Next environment where I think that also is gonna be a good place to go.
22:36 And then we'll come back and talk about Claude a little bit more for just a second.
22:41 All right. So let me go back to
22:44 this screen, and we're gonna look at a couple things here. So, um, so if I've got a, uh, this is an… This actual screen is, this is my demo environment for Tableau Next. I have, um, a really nice data model here. We talked about the, the way that we can, um, address these, give context. I'm just gonna spend just a moment because it's just essential that we do this, that we have, um, things well documented.
23:13 This, it rises and falls on us having a, a solid semantic model.
23:18 Uh, if I c-come on here, you can see that I've got anything that is business related. I think I left a few. So we, we either need to document them or we need to hide them. And when you bring them in through, um, Tableau's, um, Data Cloud 360, there are a few fields that get added that is just kind of the normal Salesforce kind of way of doing it. And these, these tables I did pull in that way instead of direct injection.
23:42 And you'll see that it added some things that were meant to be like, um, unique identifiers, 'cause it was not necessarily understanding what, upfront didn't necessarily catch that. So it's, it's preparing to make sure that you have some way to uniquely identify each row.
23:55 But I don't want my agent to be tangled up in that. So I'm gonna hide those fields because that's a hard stop on how it, what it is gonna see. I mean, clearly I can see them, they're visible here. But for the agent, it's, it's not in the way. It's, it's transparent. So back to the descriptions. We wanna have every object documented. Anything that can be algorithmically expr-expressed needs to be.
24:17 You'll see I've got twenty-seven metrics defined as calculated fields here. And the dominant feature in terms of priority order is it's gonna go to our metrics first, assuming that those are our governed KPIs, and then to calculated fields, and then the individual things inside of tables. So that sort of pecking order or priority in terms of how the agent p-processes information is important to know.
24:38 Now you'll see that I have a, a robust, um, definition in here. We can have up to two hundred and fifty five characters, and that's, um, just a, you know, a huge win. Now, side note, if you log into Tableau Next and you download the semant- Tableau semantics when you– from the into desktop, all that comes with it, okay? All the descriptions, all that good labeling.
25:02 The s- the only catch is the metrics do not, because this is essentially pull what Pulse metrics does over in Tableau Cloud, right? So you do have to do some recreation on those bit, but the, the time-consuming part of this really is getting everything labeled up in a way that's very robust. Now, back over here on our descriptions, here's my business preferences, and I'm not in edit mode, so you can see it's grayed out for me right now.
25:26 But you can see that I have, um, only used, you know, not even… just over seventeen thousand characters. And this is a tin table, um, data model with a lot of different things in here, and I've had plenty of space to give it the, the all the connection on jargon and, and things that were labeled in a way that was kind of vague and would have been a problem for my agent.
25:47 And this would take a measure of time for me to understand what was a problem for the agent. So make no mistake, this was really a, um, a work that took a measure of time, but it was really worth it because now I can ask it a lot of questions. Because I had plenty of space to do what I was doing here, I even was able to give it some, a little bit of guidance on the outputs when I ask it a, an interactive question.
26:09 But you always wanna start with those sort of, um, you know, core guardrails on what to do when it's not sure, right? Give it some guardrails and, and take that ambiguity off the table. You'll, you'll also see right here about, uh, me, uh, giving it instructions regarding causation, so, um, and giving when it's gonna make credit in that direction.
26:31 So, um… And, and then also in the Tab– in both of these, I've seen and now would implement in a new batch that if you have a que- ask for what you need, you know. 'Cause we need to ask it for what we need, and it needs to ask us for what it needs to get the job done. 'Cause it, it's very much more capable. Been testing Tableau Next since May of twenty twenty-five.
26:52 It is so– it's grown up so much, and it's really, uh, so much better. So, okay, back to the main story here.
27:00 Um, if I wanna have one of these, um, uh, visualizations shared over in, uh, Tableau, um, all I need to do is hit my, um, uh… not Tableau, into Slack. I need to send it like via this, and then I'll have something, um, en-enabled. And this one is my demo environment that is not connected, so let me find the right one. But, um, we also then can share images and messages out of, um, uh, Tableau Cloud.
27:31 That's what my, uh, Claude… This, this is the cloud site that Tableau Cloud, that Claude was using to make those connections and send those instructions.
27:40 And, um, so, um, this is, this is the fruit of it all. Of being able to get these guys to come straight into Slack and, uh, at the, at the, uh, as we do shares from the different environments, or we ask questions of Claude, those are all, uh, very possible. So let me see. In the interest of time, let me go back to, um, this, uh, little bit of presentation, and then we'll have some time.
28:11 I'd be happy to answer questions about… I could so easily talk about the agentic side of all that, and I wanna finish this first so… Okay. So, um,
28:21 so the way this really started for me in a practical way with building more, um, skills and going more deeply into, uh, Claude, quite frankly, was accelerated by two things. One was a project this summer where I needed to package up Tableau Next dashboards to, uh, be part of the agent exchange, um, share, like as an accelerator. And but before that, I really started using it for presentations and for an, uh, like a big expanded Google.
28:53 And let, let me just back up for a second. Also mention, I did start using it back in twenty twenty-two because I was teaching people and getting them market ready. And to be fair, people need to be nimble with gen AI to be, um, really, uh, leaning in and doing well in a, in market. And then also, um, we need to be wise as educators on ways that our students are gonna be efficient or cheap.
29:22 I mean, you know, one language or the other, right? So I needed to give them challenges that would still help them develop critical thinking skills and domain expertise, but not like have them just bypass the process of learning. Okay. So I spent a lot of time thinking that through and, and bringing that to the classroom. And then back to my own workplace, I started seeing how, wow, the presentations are so much…
29:47 This is so… I can do so much better with case studies and presentations and training materials. And then I got re- became responsible for editing the blogs that our company was putting up. Well, um, turns out that the contract marketing team was doing a little bit of ChatGPT to give us, uh, on a non-subscribed basis, not on a subscriber account, but just, uh, to do the w- the language there and to smooth it out and to publish, you know.
30:15 So here's the problem with that, though.
30:17 As all of you know from Tableau's point of view, we have specific words that mean specific things. So last January, when I did a dashboard session where I was showing all the dashboard actions and I was talking about parameters and sets and set actions, to smooth out those words and talk about it being more interactive or efficient was a swing and a miss on absolutely not the right words, right?
30:41 And talking about dynamic zone visibility, this sentence on the left is what I had started with on a draft back then and ended up seeing this come to me to review as ready to go to publish. Um, later in the year, I… 'cause Tableau Conference, I spoke about the semantic model A lot of my work and a lot of my webinars have been revolving around semantic everything, right?
31:03 And trying to get it so that we were more comfortable with Tableau Next, understood how to make it work, how to, how to make it work well for our stakeholders. And same kind of problem as the specific domain-specific words were being used in my blogs and such, then they were coming back to me in, on blogs to be reviewed and just published where, um, the words had been lost.
31:27 And, and then I ended up spending more time on this because we, um…
31:33 I ended up, like, collecting up what I considered to be editing tasks. And this is a little bit fancier than it needed to be, but these are all the ways that the different, um, if, if it's not skilled and trained toward the technical part of my job, these are the ways that the things were being lost. So like where it summarizes to and interpreted wrong, you basically have to rebuild the thesis of what actually happened in that session.
31:58 So, um, it became just like a pretty big problem. And so what, what we've done now instead is I have, um, saved a lot of skills. And because I use this more and more for my technical presentations, and I, I a- invite it to go research and validate the claims that I'm about to make, I bring the technical expertise of being able to, to say, "Well, I tested this.
32:26 This is what I found to be true for this moment in time." You know, but in the midst of this, in the last six months, I've got ninety-two saved, saved memories and, um, a bunch of rules that I've made, and then also one GitHub repository. Because I also travel, as I mentioned in the beginning, with my work. So I needed to not have my tools that were a part of my laptop be distinct and different.
32:51 And boy, the reason I know that is because I made that mistake in the beginning, where I had, you know, had even had Fable active on my home system, and then there's a lot of, like, richness that came with that level of, um, assistance and research capabilities and, you know, it also went through the credits really quickly. But, um, in any case, um, we, um, have– I've learned this system.
33:15 So the, the payoff for me in the end has been that like last night, I finally was able to have a– there was a presentation I gave for a webinar. Here comes the blog, um, summarized from the transcript, and then it's lost a lot of its meaning and then, and then the actual, uh, spine of it. But then I had curated this skill that then Claude was able to do a first pass on it and identify the kinds of things that would, that I usually was having to pull for.
33:44 And then the review and fix process was significantly faster. B- but it's back to this, um, cooperation between me and the saving of the skills, and then it being well-informed to be able to be a good, um, helper to me. So there's, you know, everything from the fact that in the beginning, because I had spent time around these gen AI engines, I didn't want anything in my s- my printed anything that had double dashes.
34:11 I've gone from like that being one of my normals to being something I'm allergic to. So, you know, I've given it style guide. I've given it rules of how we do and don't communicate. You know, I don't care how bad something is that comes to me from a vendor, I'm gonna find the silver lining, and I'm also gonna not overstate though what the reality of it is.
34:31 So, you know, th-those sort of like the attitude of how you write and how you want to be, um, how you wanna communicate with folks outside of your, your, um, smaller environments, those sort of things need to be written down and communicated to anything, any tool that is representing you. So 'cause we've moved way past the kind of Grammarly support, um, where we're s- checking for spelling errors to where something can really give us more, uh, time-saving support, but it has to be trained.
35:00 It has to be in alignment with your values and your, um, your domain of what it is you need it to, to do. So, um, we've got, you know, clarity on who my audience is. We've got, um, h-how we wanna make decisions, what decisions I'm okay for it to make, what decisions it needs to escalate to me. And, um, then no way are we doing anything with passwords and things that need permission.
35:26 So like I showed you of it asking me to, if I'm ready for it to post to Slack, you know, I could set up systems to make it be more automatic. I probably wouldn't for a while until I was sure that that was being well-received. But that's, that's, this is the process of the some of the things that I would recommend that you consider. And these files that are saved are, um, they're just text files or they're saved as Markdown files.
35:50 So they're very efficient.
35:52 And then that also gets saved in my GitHub repository because then it's available on both sides, um, you know, whether I'm at my home office or where I'm, uh, wherever I'm am on the road. So, so the– my t- my teaching voice, my, um, webinar voice gets preserved. And so then, um, sim- very similar to our, what we ha- do with business preferences. That's part of the reason that I kinda diverted and am showing you those business preferences is because that's where we are able to connect the dots for our stakeholders as well, for it them to have good agentic experiences with the things that Tableau now has, um, has implemented.
36:33 And make no mistake, Claude's gonna need the same information. So if you've got more going on there, that's a huge part.
36:40 So, you know, uh, one, one private GitHub, and, and I can't tell you that– it's been decades since I was a developer, so I haven't needed a GitHub repository in a while, except for specialty projects where people were asking me to store things there.
36:55 Um, but now I definitely do, I do, uh, pushes and commits and pushes every day. And I wanna make sure that that- That repository of knowledge that I'm building, both in, in, and because it helps me with my presentations. There's a level of technical expertise, and then I ask it from time to time to go review, um, the post and the, the, uh, announcements and things just to make sure that my stuff is as accurate as possible.
37:22 So, um, yeah, this is some of the recent stats on, you know, how many commits and project folders and whatnot, and it's been really interesting and very, very helpful. Um, and, and part of what makes it trustworthy is the fact that it is in, it's con- You know, as diligent as I can be, it will be, uh, up to speed. So it's really on me to inform it and to keep it up to date, and it's just such an interesting, um, partnership.
37:51 So, um, it… I would definitely agree with this part about it. It amplifies whatever you build, and then, um, it's, it, but it can keep you out of the ditch a lot of times also if you give it, um, good guidance and point it at good resources. And then if you can't figure out how to make things happen, like just connecting the dots with various parts and pieces in the Tableau community of resources, you just tell A- let it tell you what it needs.
38:15 And you'd be surprised how well it's able to walk you through the connecting and the, and this establishment of those things.
38:21 So we definitely have a, a very exciting thing around the corner here of Tableau Studio, where we're gonna have Desktop and Claude sitting in the same space. I think that there's some, uh, trainings happening for that coming up this next week or two with a lot of the ambassadors and then coming to market hopefully soon right after that. Proactive intelligence was announced at Dreamforce, and it's that idea of it's gonna come to us less because you have a threshold you've set and more because it's watching over a little broader, uh, point of view and being able to recommend an action as opposed to just saying, "Alert, alert, there's an issue here." And then, um, the Slackbot and the MCP path to be interactive in, in Slack.
39:06 It does require some permissions, a paid subscription, some of those sort of things, but that, I have seen that work, and I observe that as being a powerful addition to the, the, the path here.
39:19 So when you're back to the headliner story here, we're gonna push analytics out. We're gonna send it out as proactively as possible. It's kind of like, uh, you need to evaluate, though, what level of urgency or immediacy is needed. Is it they need to see it once a week? Do they need to see it when a certain threshold is, is, uh, hit or achieved?
39:42 Um, they need to know why it's changed. That's a different question, different, different tooling on the background that's gonna need that. Um, and then do you need to connect and con- continue the conversation in Slack, which Claude can do. You just need to give it those parameters and tell it what time period, and just be aware of how that works, um, over there as well.
40:05 Um, and then, then we have the new workflows that are coming in, and you can establish and then have it just do these sort of follow-ups for you. I think we all experience some of that sort of thing and, and with maybe a little bit of a negative connotation around it whenever you make a phone call these days for help in any way. Um, but we have opportunity to do something way better here with more intelligence wrapped around it.
40:28 So there we go. Yeah, and Christine's commenting in there about having your Claude brain backed up and get… Yeah, that's a, it's a powerful place to go.
40:37 So the, if we close the gap and we bring the, the analytics and the analysis and the important information bubbled up to the person, um, it will, um, surely help us, especially in this moment where people are very much in tune with wanting to get things done quicker than normal and the patience is lower and the intensity is up. So push analytics, I think, has potential for real being the win and helping people not overlook the context that really needs to be expressed with those, uh, little bit more context and details is what's…
41:14 And that's the role that the dashboards have served up till now anyhow. So and with that, Ann, I am up for questions, if there's other things people wanna see. I realize I have a whole lot of things that I had queued up to, um, sort of share and show, but that would take way too much time. But so if there's anything that anybody wants to hear more about, I'm happy to talk about that.
41:36 Ann Stolzman: Well, we don't seem to have any questions in the chat.
41:40 Celia Fryar: Then it's time.
41:41 Ann Stolzman: Yep. I think we can, uh, move on to our next parts, and we'll keep an eye on the chat in case anything pops up.
41:54 Celia Fryar: Well-
41:54 Michelle Maraj: Well, thank you, Celia, so much for the presentation. It was really cool, especially, well, for me, I'm one of those people that has not really played around with Claude too much yet. Just, um, yeah, I haven't had the opportunity to. And so it's really cool just to really think about, like, at what level you want to, I guess, like, rope it in to your workflows, right?
42:12 Celia Fryar: Right.
42:13 Michelle Maraj: Um-
42:13 Celia Fryar: Right.
42:14 Michelle Maraj: Yeah. Do you… I was actually kind of wondering, so do you see, I guess, in some of your clients or people that have been implementing it, is there hesitation to-
42:25 Celia Fryar: Oh, yeah
42:25 Michelle Maraj: … like, trust-
42:26 Celia Fryar: Understood
42:26 Michelle Maraj: … the AI or, or is it more like-
42:28 Celia Fryar: Mm-hmm.
42:28 Michelle Maraj: I feel like there's two ranges of the scale, right? It's like the hesitation to trust it-
42:32 Celia Fryar: Mm-hmm
42:32 Michelle Maraj: … versus people that are like, "I'm going to implement it everywhere and have it do everything." Um, so I guess-… like, where on the scale kind of
42:40 Celia Fryar: I have seen… Unfortunately, for better or for worse, I've seen fewer of those, um, where it's, "Let's do everything." Um, and I think that people need to have a few wins before they can more relax into it. And the agentics question and answer inside of Tableau's platform is kind of a good place for people to get their feet wet, like the enhanced Q&A part of Pulse metrics, right?
43:03 Those sort of features or ways where to, like, um- You know, within bounding boxes, give us some security that it's gonna do well. Now, um, you know, I definitely have just a whole separate side of my life that is, uh, where I'm needing information, I'm researching, I'm trying to find out what's going with products, what works together, what doesn't work together.
43:24 Um, I do everything from troubleshoot logic on calculations to trying to find ways to execute faster. So like that part of things I have really valued, um, being able to use it. It's like, um, when I first in '22, '23, I was teaching a lot of Python classes and, um, you know, my goal was to teach it for analytics, not to teach it for web development.
43:49 And so there was a lot of like syntax, um, support that it could give me. And then also like all the little things you run across like versioning for libraries that, and compatibilities and, and stuff that just would take an, a nightmare amount of time to figure out in the past were just like, oh, and there we are. Oh, now we're gonna get to move on.
44:09 So yeah, it was, that was, um, that was probably how I got won over. That, plus I was able to really put together better business case studies for my students at USF, and I feel like people learn better when there's applied logic, logic and knowledge, right? So I didn't wanna just bore them with, um, information. I wanted them to, to do stuff together so they're team working and they're developing domain expertise.
44:34 But, you know, to come up with the things on the back end of that so there's enough variety of it, takes a measure of time. That is tough. So yeah.
44:43 Michelle Maraj: Mm-hmm.
44:43 Celia Fryar: But I don't know if I answered your question, but more, more cautious than not, but more like people are under the gun to apply, and then to just show-
44:51 Michelle Maraj: Mm-hmm
44:51 Celia Fryar: … how they're improving their workflows.
44:53 Michelle Maraj: Yeah.
44:53 Celia Fryar: Which is-
44:53 Michelle Maraj: 'Cause I mean, I feel like yeah, there's a huge push of like, "Oh, we need to be more efficient. Oh, we want…" I mean, just the other day I was making a slide deck by hand and I was thinking like, in this day and age, like, why am I doing this by hand, right?
45:05 Celia Fryar: Yeah, yeah, yeah.
45:05 Michelle Maraj: Like, what am I doing?
45:06 Celia Fryar: Yeah.
45:06 Michelle Maraj: And so I feel like, yeah, there is that big push to do it.
45:09 Celia Fryar: Mm-hmm. Mm-hmm.
45:09 Michelle Maraj: But then yeah, it always comes back to-
45:11 Celia Fryar: Mm-hmm
45:11 Michelle Maraj: … yeah, can you trust? But I feel like especially with the semantic layer and that ability for it to see the breakdown of like-
45:17 Celia Fryar: Mm-hmm
45:17 Michelle Maraj: … okay, it's giving you this number, how exactly did it get there?
45:20 Celia Fryar: Yeah. Yeah.
45:21 Michelle Maraj: Um, really adds a lot of comfort to like people using these types of implementing.
45:24 Celia Fryar: Yes. The audit trails that it provides in both sides of the house, Cloud and Next, is-
45:29 Michelle Maraj: Mm-hmm
45:29 Celia Fryar: …
45:30 is essential, right? And-
45:32 Michelle Maraj: For sure.
45:32 Celia Fryar: Yeah. And if you ever are doing something freestyle inside of Claude or, um, um, Codex or one of the OpenAI platform engines, this is like, you can always ask it how it got there and it will. Just, you know…
45:48 Those things are… They tend to be more verbose than we might even want them to be sometimes, right? So-
45:54 Michelle Maraj: Mm-hmm
45:54 Celia Fryar: … um, it's just like they are eager to tell us all the things.
45:58 Michelle Maraj: Right.
45:58 Celia Fryar: And, uh, yeah.
45:59 Michelle Maraj: No, it's good.
45:59 Celia Fryar: But I, I used to spend so much time on presentations and on materials for, you know, everything from talks to classroom.
46:08 You know, in a classroom, especially with learners these days, you have their attention for a nanosecond. And it has to be engaging and you need to like pull them back in and have something that's kind of on the move. And, um, you know, that, this has significantly changed. Been, I've been able to upgrade the graphics and then also I have to stay in the trenches with it though to get it to reduce the words on the screen lots of times.
46:32 Michelle Maraj: Mm-hmm.
46:32 Celia Fryar: And to change the words on screen and to audit everything it does because, um, no matter how quickly it generates it, it, it, um… Well, the, the person of our company recently said something about how the, the onus for, um, accuracy is really shifting to the reader, you know?
46:51 Michelle Maraj: Mm-hmm.
46:51 Celia Fryar: 'Cause it is giving you what it understands to be true, but how, how big is its scope? And does it actually have the whole, whole picture? And then there's a lot of stuff that's published out there that is what people wish to be true, but it's not true.
47:06 Michelle Maraj: Mm-hmm.
47:06 Celia Fryar: So, you know, especially with product stuff, I, I wanna test it, and then, uh, there's some, you know… And then Claude and I argue it's been like this long. No, that doesn't work like that yet. So like…
47:19 Michelle Maraj: So I know that we've got our surprise coming up, but we did have one question come in.
47:23 Celia Fryar: Okay. Yeah.
47:23 Michelle Maraj: Um, and so I'll… It's a little bit long, so I will read it, um-
47:28 Celia Fryar: Okay
47:28 Michelle Maraj: … because I think the recording doesn't always show the chat. But basically, um, since Tableau is rolling out its own agentic AI capabilities, how do you view the ideal division of layer, labor when layering an external LLM over Tableau Cloud via MCP? Specifically, should we treat Tableau as the governed semantic layer and the source of truth, um, or, you know, other, leveraging Claude as the cross-functional tool?
47:52 What do you think?
47:54 Celia Fryar: Yeah. It's a great question, and it's one that I really, um, am seeing people wrestle with a lot. And, um, there was a very good podcast I just wanna call out between Kirk Munroe and the, and Tim, who now goes by just Tim. It was Emergency Broadcast. And toward the end of it especially, they talk about the transition in market and how that, uh, we're likely to become curators and also, um, you know, um, analytics engineers.
48:25 And so back to your act- your question though, um, it will… I think that we'll have better outcomes. And, and I actually did a presentation about doing viz extensions in Denver recently, and this question came up. And so like I think we'll have better outcomes if we, if we will allow something like Tableau to, to contain the governed semantic model.
48:48 And then for those one-offs that are quick hit one, one-hit wonders, sure, do it in Claude. But then, um, the things that are more e- enterprise grade, going on for a while, trusted resources, I think that people are gonna need to have a little bit more, um, vetting and standing in the gap with that, and it be a little bit less of what- Um, Tim and, and Kirk were calling AI slop.
49:12 So it's just, it's too easy for things to be generated and then… You know, think about Pulse metrics, for example.
49:18 Uh, part of the reason what those get kind of like when they're brought into an organization, sometimes there's friction around them because you're, you're talking about the number, the big number, right? The BAN. And, um, the one that people get held accountable to, and how is it calculated, and is it the, is it the truth for everybody or just some?
49:36 I mean, like, since we're, there's a lot of factors involved there. So if we can land in some space where we do have a semantic model that's vetted and, um, like blessed, some gold layer or, you know, that's gonna be really important.
49:53 So your beautiful little child is on you. That's so sweet. I love that.
49:58 Michelle Maraj: Then I was like, "Can we do a face shot?" He's asleep, so.
50:02 Celia Fryar: Aw. Precious.
50:04 Michelle Maraj: But awesome. Yeah. Thank you so much, Celia. Um, really appreciate having you-
50:07 Celia Fryar: Thanks
50:07 Michelle Maraj: … having you here.
50:08 Celia Fryar: Thank you very much.
50:09 Thanks for having me. Means a lot.
50:12 Ann Stolzman: Thanks for coming. We appreciate it.
50:13 Celia Fryar: Yes, ma'am.
50:14 Ann Stolzman: A lot of information. So, folks, for those of you who stayed till the end, we have a surprise, like I promised. We have swag. We are giving away swag. We are pulling up a wheel of gains. So if you win, we're gonna do two drawings. If you win, please contact me, I dropped it into the chat, and we'll work out the details, and you will get your choice of a $50 Tableau gift card or a handy-dandy really cool-
50:45 Celia Fryar: That's a big gift certificate, girl. Wow
50:47 Ann Stolzman: … It is a big gift certificate.
50:48 Celia Fryar: Impressive
50:49 Ann Stolzman: Or an Owala free sip bottle. So you'll get a choice. Now, the free sip, we can ship you that. You would have to give us your address, and we will ship that to you, and you will, like, pay for shipping. The $50 gift certificate, it's big enough because shipping prices are a little expensive, and that will hit your card. So that's why we're giving you a choice.
51:13 So our first winner will be
51:24 Christine.
51:26 You get your choice.
51:31 So please email me at the email address, and we will pull you out of this.
51:37 And winner number two…
51:40 Tiania Chan: I feel like we need music or, like, a drum roll or something, you know?
51:45 Ann Stolzman: Right?
51:46 Celia Fryar: Next time. We'll do it next time.
51:47 Ann Stolzman: Is Amy.
51:49 So Amy and Christine, please email me at… Christine, you do not get to choose. This is for you because I'm gonna contact Celia about her own separate little giveaway because she was so kind to speak with us.
52:02 Tiania Chan: Thank you so much, everybody, for attending. Uh, as a reminder, our next event is going to be on December 8th with our very own Michelle.
52:12 Um, Celia, thank you so much for presenting, uh, Claude and showing how it could work with Tableau. That's amazing.
52:21 Um, and I think that's it for us, right? We'll be sending out, um, an email with the recording, um, as soon as it gets uploaded onto YouTube, so you guys will see it there.
52:34 Ann Stolzman: Okay.
52:35 Celia Fryar: Congrats on your little one. One month old tomorrow?
52:38 Michelle Maraj: Thank you. Yes.
52:39 Celia Fryar: Wow.
52:40 Ann Stolzman: Yeah. Time does fly.
52:41 Michelle Maraj: I know. Somebody's-
52:42 Celia Fryar: I think he-
52:42 Michelle Maraj: … hungry, but…
52:44 Ah. Oh, my hair. Sorry, little dude.
52:48 Just getting-
52:48 Ann Stolzman: Oh, buddy.
52:49 Michelle Maraj: Yeah, yeah.
52:51 Yeah. We got-
52:52 Celia Fryar: All right, friends
52:52 Tiania Chan: … seriously, we'll be in touch.
52:54 Ann Stolzman: All right.
52:54 Celia Fryar: Thank you very much.
52:55 Ann Stolzman: Take care.
52:56 Tiania Chan: Oh, yeah.
52:56 Celia Fryar: Bye.
52:56 Ann Stolzman: Buddy.
52:56 Celia Fryar: Thank you.
53:01 Tiania Chan: Bye, y'all.
53:02 Ann Stolzman: Bye.
53:03 Tiania Chan: Adorable.
Shorts for this TUG
- Dashboards had a lifespan before AI arrived 5:02
- Users just want Excel and Claude dashboards 5:33
- Ignoring the dashboard doesn't stop the KPIs 6:12
- Push analytics still respects row-level security 9:05
- The store PC was also the cash register 10:15
- Too many alerts and people stop reading them 11:51
- Haven't looked at Tableau Pulse lately? Look again 13:30
- The AI decided a 20% discount was high 16:31
- Claude asks before posting to Slack 21:38
- Generic AI rewrote my Tableau terms wrong 30:17
- Teaching Claude my style: no double dashes 34:03
- AI amplifies whatever you build 37:43
- People need a few AI wins before they trust it 42:40
- The onus for AI accuracy is shifting to you 46:45
- Govern in Tableau, do one-offs in Claude 48:39