Build Your Dashboard on Trusted Data

Steven Hloros

0:00 Hello, everybody. Get some summer music going here for all of us. Hope you're having a wonderful week. Hope your Thursday's going well so far. I see Tyge, you're in here. Hey, Tyge. Hey, Megan. Hey, Gary. I see a lot of names. I, I love when people— y'all spend so much time with me on these LaunchPads and the Accelerator, and yet you still show up to see and listen to me. I'm so glad you're not sick of me yet. That's so kind of you. But it's great to see you all. Hopefully, as I said, you're having a wonderful Thursday, a wonderful week so far.

0:41 Hopefully, the summer's going well, and hopefully, these sessions are helping you figure ways to actually take the time to have some piña coladas. Not cheat on your wife, but have some piña coladas and get some time back. We're gonna start promptly in one minute because we do have a bit of a packed agenda here today. So give that just a moment. Hello, hello. Let a few more trickle in. Start in just a moment here. Hello, hello. Always feel free to use the chat. Say hello, say who you are, say where you're coming from, whatever you would like. Obviously, this is a community event.

1:30 We're just getting such big responses nowadays for community events. We're actually doing them as webinars. We never had to used to have to do this before. So this is so cool to see so many people now showing up. I think we had a hundred, over a hundred sign-ups for this one specifically so. Well, hopefully you get enough time back, Jamie, that you can enjoy your piña colada. That's the goal here. That's the goal, is you can use the MCP and save yourself some time. All right, everyone, we do have a little bit of a jam-packed agenda. We actually have Shubham here. I'll introduce him in, in a few moments.

2:03 He's gonna be showing off how we actually do things with the Atlan MCP here in Atlan. But welcome, welcome, welcome everyone to our second session of the Summer of MCP. This one is going to be focused on the business analyst persona. Now, if you missed our first session, that was on the data engineering persona, and you can always get access to that. It's on our website, and in the follow-up, you'll get links to that video, this video, and everything MCP. But here today, we are focused on the business analyst. So our goal for today is to introduce what we'd normally see an analyst life being like, and that's focused on building new dashboards.

2:42 You get that request. Then we're gonna look at a ready-made demo that I have prepared, kind of showing off how the Atlan MCP can help with that request. Can take it from ticket to exactly where you need to build in just a few steps. Then I'm gonna pass it over to Shubham, who's here to actually show the real use case that he used the Atlan MCP for inside of Atlan. We do dog food here, so he actually has been working on some different things with our Atlan instance on our own data, just like all of you, and got a ton of value from the MCP, and it relates directly to that analyst persona.

3:22 And finally, we'll just go over where to go next, what to do. This is the Summer of MCP. We have a lot of things going on, so just wanna introduce that. Now, before we start, I do wanna do a quick poll. Since we are focused on the business analyst persona, I wanted to ask everybody here, when you think about that type of request, "Hey, I want a new analysis," or, "I want a new dashboard," what actually slows you down the most to getting what you're looking for? Is it finding the right trusted data to build on? Is it pinning down the official metric definition?

3:56 Someone wants revenue, but you're not sure how to actually calculate that revenue because it's for finance instead of your sales team. Is it knowing the how the data's already used, how the business might already be using it, so you don't have to rethink of joins and filters and how to find it? Is it rebuilding something that already exists? I used to build some dashboards back in my day before I worked here at Atlan. I used to be a customer. That was part of my job, and that was always the biggest pain for me.

4:24 I felt like every time I built something new, I'd end up finding out something already existed, and I just wasted a ton of time. Or are you actually streamlined? Which would be amazing, and I need to talk to you so I can tell others how you've been able to streamline all of this. And responses so far, we have 13, mainly knowing how the data's already used. Interesting. Joins, filters, and common questions, and that's actually something we're gonna be showing off here today. When you use what we call SQL intelligence using Context Agent Studios, you can actually get that information right inside of Atlan automatically and can use that with the Atlan MCP.

5:00 And I also see finding the right trusted data to build on, absolutely. Pinning down the official metric definitions. And it seems like you all have have covered rebuilding something that probably already exists pretty well. Only two, two respondents there that see that as the biggest— Three respondents as the biggest problem. That's great to know. It's good when you're not redoing work that's already in place. And again, the Atlan MCP can help keep from that happening. So thank you all for responding. I'm gonna end this because I wanna share this internally because we're always looking for different ways to accelerate your processes.

5:35 So before we dive into the analyst persona, just to give a quick overview, I assume most of you already know what the Atlan MCP is. The MCP is just the way that your AI can actually use the tools or connect to another system, right? Claude, Cursor, Copilot, they use MCPs to connect to a different system, and it allows you to talk in plain language and execute on those systems.

Steven Hloros

5:58 And the Atlan MCP does that for you.

6:01 And while today is focused on the business analyst, I want to point out that the Atlan MCP and its tools and its abilities aren't restricted to just a business analyst role. As I said, we went over data engineering yester— last week, but overall, you can do things like search and discover. You can find and govern your definitions. You can find your SQL intelligence, which we'll show off today. You can also traverse lineage and find trust signals, and you can make sure that every answer that you get is governed by your single context layer of Atlan.

6:35 But today, we are focused on the business analyst, and the business analyst, let's face it, one of the biggest requests that they always get is, "Hey, we have an urgent need for analysis on our orders and our profit or our MRR, and we need a dashboard. We need a new tile. We need something for our meeting on Friday to be able to present to a board or a customer. I need you to build it." And when that happens, it's a process. That analyst has to go through and find what are the right sources of the data. They have to understand the actual metrics that the business wants.

7:12 And a lot of the times, even if that request comes in, there's a lot of back and forth to understand maybe how the source data was already used, that filter ability, that join ability. Where can I find that data? How do you actually think about it? What is that definition? And that back and forth can take up a lot of time. In using the Atlan UI, obviously, that context all lives in there for you. You can go to the glossary and find the metrics. You can go and use the asset search to find your certified tables. You can do all of that in the Atlan UI.

7:44 But the key here today is showing off how, as a business analyst, maybe I'm working in Claude Code, and inside of Claude Code, I usually do my ticket triage to see what's actually coming up, and I actually do my SQL building in there, and I connect to other MCPs, and I kind of live in on there. And I might not want to switch context windows every time I want to find a new column or investigate or understand from Atlan. And the Atlan MCP is the solution for that business analyst to stay in the tool they already are in. So, as I said, think about this from that business analyst perspective.

8:24 The first question is: Which data do, do I actually trust? How do I know where to start for a request? The second is: What are the real metrics and definitions, especially considering the source audience that's requesting it? Is it sales? Is it finance? Maybe they have different definitions of revenue. And finally, understanding how the business might already use that data. That can lead to breadcrumbs on building things out much quicker than having to start from scratch on a table that just has regular columns. Here, we're gonna switch over to the actual demo on this.

9:01 And in this case, we're going to have a request that just landed from our sales ops team, and they're asking to build a product sales dashboard with revenue, orders, and units by product and color trending over time ahead of their next business review.

Steven Hloros

9:15 And as I said, all of this can be absolutely done inside of the Atlan UI. Apologies. Inside of the Atlan UI. But here, we're going to actually use the entire flow right from Claude Code. So with that, I'm gonna switch over to my coding agent. And here in Claude Code is where I'm going to start. Now, as I said, this flow is going to start from the actual, you know, starting point that most people start their day with in these type of roles. I want to understand what's actually on my plate. I just signed in for the day.

9:53 I'm using Jira, and I actually have the Atlassian MCP connected, so I wanna pull my open request inside of my project. So I've already built a skill that knows exactly, hey, this is the project that I work on. This is my daily task list. Go check it. Make sure you have the right assets and everything of that nature, the right tickets. And so what it'll do is go ahead and call Atlassian, and what we see here inside of Jira is that exact ticket. Again, didn't have to log into Jira, didn't have to do anything, but I can see for my day, I have that new dashboard request. Sales leadership wants it.

10:26 They want certain parts like revenue, order count, units sold by product and color, and they also say that they want to make sure that the revenue calculation is quantity by unit price before tasks. They need to make sure that's actually there. All of this could be done in the Atlan UI, but we're gonna do all of this right inside of Claude. So what I'm gonna go ahead and do is first go ahead and ask it. Obviously, it knows me a little bit, so it's already asking me if it wants to look at what to build on, and I say, "Yeah. What is that trusted data source that I should start with?"

11:02 And here's where the Atlan MCP kicks in. Because I'm actually asking about the certified table, Claude knows, hey, that's a tool call that the Atlan MCP has, and it's gonna go ahead and call Atlan to search for a certified table that matches with the information inside of the ticket. It's combining the context from both, right? I'm not reiterating what the actual ticket was. It already has all of that context. So I've already skipped a step at having to say, "Here's what it needs to have," because the ticket was well-defined. And so what it's happening here, it's calling Atlan. It's actually doing that search.

11:38 And in just a few moments, we're gonna go ahead and see what it actually finds inside of my Atlan tenant. So give it a second here. And what we will see Always likes to do this on that one. We have a result, and what it's actually telling me is that there is a specific table that is verified that's even owned by sales of orders denormalized. And because it has good descriptions and understanding, it tells me it's the flattened pre-joined order mart. Order lines already stitched together with the product, color, quantity, unit price, tax, salesperson, and dates, purpose-built for reporting in BI. It's exactly what I need.

12:25 And this kept me from going into Atlan searching for something like orders and maybe seeing the raw layer or looking at other assets. Because it knows exactly what I'm looking for, it's able to match that to the assets inside of Atlan and point me to one that should fit exactly what I need. Additionally, it gives me an output, so I can go ahead and hyperlink to go look at that asset itself inside of Atlan if I wanted to verify. But let's go ahead and say, "Yeah, I know this asset." Now, sounds great, but the biggest thing from that ticket was they want to make sure that the revenue is calculated properly.

13:01 And so I say, "Hey, the ticket says they calculate revenue as quantity times unit price before tax. Does that match a governed definition on this?" Because you pointed out that it has revenue. And so again, it's going to go back to Atlan, and it's gonna check because that asset itself had a term of revenue that is verified from the sales team attached to that table. And it tells us, because inside of that term, it has order revenue equals the sum of line quantity times unit price before tax.

13:31 It even has the SQL that does this, and it's telling me exactly, "Hey, not only do you have the right table that has all the information, but the the actual calculation of revenue that is tagged to that asset fits exactly what they're looking for." So now I have the asset. I know that the actual revenue calculation is right. But the last thing I want to do, and this is where SQL intelligence comes in, I want to know how it actually is used. Now, Context Agent Studio, if you haven't seen it, is our way for us to use AI agents to enrich your assets.

14:06 That can be through description, that can be through README, or that can be through SQL intelligence. SQL intelligence actually reverse engineers the SQL from assets that are being queried, like Snowflake, Databricks. It reverse engineers it to find common filters, joins, and even the business questions that those assets answer.

14:25 And so here, this asset, orders denormalize, actually had SQL intelligence run on it, the agent, and it actually came back with the common filters of the SQL that's run on those assets, telling me that there's an order date and color name, telling me this is exactly the slicers that people are using, how they actually utilize it, and it actually has business questions associated with that table, such as headline KPIs, monthly trends, order line details. All of this is produced and kept on that asset, but the MCP can get all that context and bring it right to me, giving me exactly what I need. Does, does that table actually fit everything that I'm looking for?

15:08 And here's the coolest thing. It's giving me, "Hey, existing query panel is leaning heavily on black. Does the work do the same? Well, the same shape generalizes. Let the color be slicing rather than hard coding it." So it's even giving me how to build out that dashboard based on the query usage and the patterns that it's seeing by the business. And so here, my work is essentially done for the pre-work. I know now where to start at the source. I've confirmed that the revenue is calculated correctly, and it even gives me a suggestion on how to actually build this out in my BI tool. And it's as simple as that.

15:46 I never once switched to Atlan. I never once switched context. It's all here for me. And now I can go and start actually doing the work. So in a recap, essentially, we just saw everything that you can do inside of Atlan all inside of our tool. We find the certified mart, not a lookalike. There might've been raw tables. We're not hitting on those. We match the calculating, calculation to the governed one that's already on that asset. And because of SQL intelligence, we saw the filters and the questions that people actually run on it, saving us so much time.

16:24 Now, that is a great example of how it can be used, but Shubham here is here to tell us about how we actually use this in Atlan, as he has been tasked and was tasked with building out a use case for a talk to data analyst, and he actually found a ton of value using the Atlan MCP to understand where to start in the context that he needed to actually build this out. So I'm gonna turn it over to Shubham, but first, he's gonna go ahead and talk about why his use case aligns exactly as a build-out of a dashboard that a BA, a BI analyst would do. So Shubham, if you would.

Shubham Bhargav

17:02 Thanks, Steven. Hey, folks I'm Shubham, and as Steven mentioned, I was tasked with building a customer Health 360 agent of not, not long ago, a few months ago. And one thing I realized is while building the agent, I actually went through the same series of steps that as an analyst I went through back in my days.

Shubham Bhargav

17:25 I had to find the right asset. I actually had to understand all the definitions, the measures, the dimensions, the filters. I had to assemble them together and verify if they make sense. And imagine, I'm working with the complexity of business users, terminologies, and the dataset of the with the complexity of multiple years of transformation. So with that said, now I believe the notion of dashboard and agent is quite similar. It's just that the ship it, the last step, is different. So we are gonna get till that point, and Just assume that after that point, it's just a skill to ship the context to an agent or to a dashboard as Steven showed.

18:08 So without further ado, let me just tell a story first around the situation we wanted to avoid when we were building this agent, because with agent, the hallucination also kicks in. The same thing happens with dashboards in a different fashion. Imagine a QPR. We are reviewing all accounts, all our customers, and looking at their health status. The health status is saying green, this customer is good. But the CSM, the customer success manager, silently is very, very anxious because they have not been able to talk to the champion or the, on the side of the customer, and they believe this customer account is actually at risk.

18:46 Now, that's not the situation we want to be in when we are reviewing things as a dashboard or as an output of the agent. That's the thing we wanted to avoid. So with that as an anchor, let's jump right into it and see how we can avoid that. I'm gonna take over the screen, and let's go through the steps. Just a sidetrack before we start. Just some housekeeping that I generally do with all my cloud sessions, similar to how Steven did with skills. I have a bunch of instructions on how I work, how I behave, and in this particular situation, what not to expose because we are also, like, working on our internal dataset.

19:26 So imagine if you look at it, the way I behave is more around terms first, dataset later, and that's what the agent has been instructed to do. I don't like Geo IDs, I like names, so the, the instruction is also fine-tuned to that. So just some housekeeping on how you can tweak your setup to behave in a certain fashion that you behave, your team expects, or your organization has principles on. So with that at the back of our head, let's start. So first, we need to find the right dataset. And imagine the spot I'm in.

20:02 Given that this dataset, whatever dataset I'm gonna find, is a series of transformations over the course of years, I'm gonna find raw, silver, gold, insane number of aggregate tables, insane number of terms for me to skim through and figure out what I even need to use. And I have two really bad options in the absence of Atlan MCP and Claude. One, I guess the worst one. The second, I actually hunt down the person who may know this, set a time with them, wait on their availability, get the understanding, and then start working. I don't like either of them, so I'm gonna take the third approach.

20:39 I don't like talking to humans to that extent, so I'm gonna just take an agentic approach, and let's see what it comes out with. It's very, very interesting. It's not just that what I should use. First of all, I found a really good anchor point. It's not a good, it's a best anchor point because I got a data product. That means somebody actually has done the work of collecting information, persisting it in a way that I can use. And at the same point, it's not just that what the data product is, it's the trust signals I need. Remember the instructions I talked about? This is what it gets.

21:10 Because I want to focus on some trust factor to know this is the right asset to use or the right table or the right data product to use. And interestingly enough, it actually started calling out some of the things I actually need. Not all of them, but some of them. Imagine going back to the earlier notion of health. We want to know what health is all about. So that's, that becomes our next step. Let's, let's go into that because as with anything, health has multiple meanings. If you talk to a product person, they will talk about engagement, the usage.

21:50 If you talk to a customer success persona, they will talk about the time it takes to get a reply from a sponsor. If you talk to support, they will talk about the number of tickets open. And I believe, I personally believe all of them should be accounted for. Let's see what we are accounting for in this case, whether it's a product, it's a CS, or it's a combination of the whole. As with anything, it takes a little bit of time. And I also added one more nugget because I wanted to check. In my personal belief, the support level, the access on the support to a customer should be an indicator of the health.

22:28 If it is not, I want to make sure that that's, that's that's being addressed. So it's still going to do the search. Let's give it a few more minutes. Okay. In the meantime, let me just show you what I had done before while it is doing the thing. So for me, when I asked it the definition of health, it came back and mentioned that, "Hey, you know what? We actually have a single health measure, and it accounts for everything under the hood." It is not giving me the formula, but it is giving me a trust signal.

23:06 At the same point, let's say for the sake of an argument, I can use it, I'm still worried about how to use open ticket count. And this is not a term problem, this is a data problem. I actually wanted to add a tile in a dashboard or a question in the agent to always add the number of open tickets at the end so that we can manually gauge it. And in this case, this filter was not right, which I initially added. The filter needed to be solved and closed.

23:37 So at this point, imagine we have gotten to a situation where we have found the dataset, we have found the definitions, we have found the measures, and still one thing is remaining for us, and that I discovered the hard way, which is default filters. So we have this term internally in the organization called active. An active customer is the only customer you want to talk about. Now my question was, is it active o-, is it active customer on product? Is it active customer on engagement? Are these accounts who are paying us? Are these trial accounts? Do we also consider churned accounts? All of that was valid questions.

24:19 And the interesting thing, and that's where I believe the power of MCP plus Atlan kicks in, is you know what? We do have a definition of active, but we don't have a standalone definition of the formula. So yes, we do talk about it a lot internally, but we have not cataloged it for people to use. And you know what I had to do? Now, this is not a problem anymore today, but back then I had to hunt the person. This is where humans are still important to get to that answer and then catalog it for the other users to find those answers in the future. And that's exactly what we did.

25:01 Now, taking a step back, what we did, we found the data set, we found the measure, we found the right filters. The only thing left is now to build a dashboard. I'm not gonna show you today because that still takes a lot of time, but after this, all I had to do was nudge it to create an agent. And just to drive the point home, in the pre-GenAI, pre-Atlan world, it would have taken me at least a week or two to find the right person to talk to them to get the context.

25:29 In pre-GenAI world, but with that said, it would have taken me still a few days, because imagine I need to click on a bunch of UI components to find all this information out.

Shubham Bhargav

25:38 Today, we did that in less than 15 minutes. Truth be told, few months ago when I was still naive, it took me a few hours, but that was still much, much, much better than pre-GenAI and pre-Atlan world. And that's about it about the use case, folks. I hope it was helpful. I hope you can leverage the same value. And over back to Steven.

Steven Hloros

26:01 Thank you so much, Shubham. Awesome. So those are two different situations. Again, one was more of a curated demo, one was exactly how we here in Atlan use the MCP to find the answers. And it's so funny that Shubham brought up active. I used to be a CSM about four years ago, and what an active customer was, was something that I even didn't know. I was, I was a CSM 'cause we never have it documented, and it was through his exploration and understanding with the Atlan MCP that we got to a human in the loop to now make those decisions, figure out exactly what an active customer should be, what that definition is.

26:43 So this is all about taking all of that work, all of that context, and putting it into a place where you can understand, don't have to switch context, and make sure you're getting the right answers accelerating you overall. So with that, we're gonna go ahead and close up here. I know we only have a few moments left, but a few next steps for you all. Please, please, please, if you haven't already, please set up the Atlan MCP. We have tried to make this as easy as possible nowadays.

Steven Hloros

27:17 If you haven't taken a look, if you tried it before, it has become so much easier. I've gone ahead and put that into the chat, and if you would, you know, take a look at it, it has all the different clients you can associate with, whether it's Claude Code, Codex, Cursor, it's all in there for you. And there are different plugins, everything of that nature. Second, use it on your next request. If you are an analyst, try it out, get connected. You do have to have an Atlan user. It uses OAuth. But we wanna make sure that you actually have access to it, so just go ahead and do that.

27:55 And then finally, send the recording to your team. We will have the recording out tomorrow. We do take it, we'll build our webpage for it, and you will get a direct link to this recording. We, we have to take it in and get it all there. I saw James, "Can you post the recording link before we end?" I will have that. You will get an email tomorrow with the recording link. I apologize on that. It does take us time to actually put it together. So I will have that for you tomorrow, James. Apologies.

28:22 Finally, what's still happening this summer, it's only mid-July, but next week we have session three, which is more focused on business users, that search and discovery aspect with the Atlan MCP. So if you have business users that haven't ever used Atlan, but they can get benefit from the context, or maybe they are using Atlan a little bit, but maybe they're... Spend their day nowadays in Claude or in ChatGPT, it's a great session for them to attend so they understand how they can get the most out of the context. Also, we're gonna have an Atlan MCP use raffle coming up, so be on the lookout for that.

29:01 If you've attended one of these, you will get notification when we're kicking that off. But it's gonna be a refer and earn, and it's extremely simple. It's just gonna be use the Atlan MCP. That's gonna be it, and you'll be entered into our raffle. And finally, in the coming week, we will have the Atlan MCP Cookbook. This is a cookbook that we're building on our end of different use cases with the different prompts you can use. Pretty simplistic, but will get you going with the Atlan MCP. Our goal, though, is for this to become a community cookbook.

29:33 We're planning a community cook-off by the end of September, where you can actually submit your own recipes for others to understand how you're getting value with the Atlan MCP. And with that, we are right at time. So I wanna thank everybody for joining. As I said, tomorrow you will get the email with the link to the actual recording. You can share it with anyone. And thank you all so much for joining. Hopefully, this is helpful. Hopefully, you'll get some pina coladas this weekend. Thanks, everyone, and thanks again, Shubham.

Shubham Bhargav

30:08 Thank you, folks.

Steven Hloros

30:11 Thanks, everyone.