Full Platform Demo: Building Context for Your AI Agents

Daniel Bergquist

0:04 don't get to them, I'll try to save some time at the end for the rest of those questions. All right, so when we think about context and we think about AI agents, one of the key components is that AI agents are incredibly intelligent. They're super smart, but without context or think about it as without direction, they start to make up some of their own thoughts. So if you're asking a question like, "Hey, how is our quarterly revenue targets looking?" Well, the AI agent is going to do its best. It might give you the best lie humanly possible, but it also might just give you the worst one. And so you may take that data from revenue that your agent thought was right, and you might go to meetings and look a little silly as you have the wrong information. So the goal is with Atlan is we're providing that context, and we're taking what your business actually means and being able to feed that into your AI agents, making your talk to data agents, your analytical agents, and even your operational agents, more effective and more trustworthy.

1:04 So the way this works is when we have Atlan, what we do first is we want to connect to your data estate. So we're going to connect to your data warehouses, like Snowflake, Databricks, GCP, BigQuery. We want to know how that data was transformed in all the pipelines that load in. We want to understand that ETL. And then we also want to connect to your BI tools, your dashboards, and even your CRM systems like Salesforce. And so that way we get the whole picture of how data flows through your business and how you're actually using it. Because part of getting information from those BI tools and dashboards and from Snowflake or Databricks, we're going to get the query information. So we'll be able to know what questions people are asking and how that data is being used, what measures and calculations you're using, so we can help then build meaning for all of that within Atlan.

1:56 Help get in glossary information and metrics so that way you have certified metrics and definitions. So like our last slide, when someone asks the model what revenue is, well, that's okay because it can go and look up the actual certified definition. We do support Fabric. Thank you for asking that question. Sorry, it's been supported for a while. I just always think GCP, Databricks, and Snowflake. But yes, we do support Fabric. And then we also will then be able to develop the lineage. So we'll showcase that lineage from those source systems all the way step by step with processes, even the SQL transformations, all the way from that origin, all the way downstream to your BI, and even provide you impact analysis. So even help with operational things like if I change this column or I drop this table, what reports or what downstream information might be impacted.

2:46 And then we can also add in policies. So we can help add definitions and policies around classifications and tagging. That way that data, maybe you want to mark certain data as AI-ready, or maybe you want to make sure that certain data is never going to be allowed AI information or to be used by AI tools. We can tag that so you can ensure that the right information is being used by the right people and agents. And then where that context shows up, we're going to enable search and asset profiles within our tool. We'll also enable information and asset searching within the BI tools themselves through our Chrome extension. We want to make sure that data teams have workflows to automate classification, and then we want your agents to be smarter. So we want to activate those agents and that knowledge through MCP. Today, what we're going to walk through, we're going to walk through five steps, from connecting, understanding the data, enriching the context.

3:37 So Atlan's going to help you actually enrich and build information. I know, I'm sure everybody on this call is just, documentation is your favorite thing. Atlan's going to help you avoid some of that work. That way, after Atlan helps you enrich, your humans can then verify what's been written, but not necessarily have to start from scratch. We'll also ensure you can govern, so enable tags, policies, verification, and certification information, and make sure that data is trusted. And then we'll activate that context by, we're going to actually going to show using Atlan's context within an MCP connection with Claude.

Daniel Bergquist

4:14 All right. So step number one, first thing we got to do is connect to that data. So we're going to do that in three steps. We're going to go look at a workflow, scope and schedule it, and then run and check out an asset. So with that said, let's dive into Atlan. So when you're first into Atlan, the first thing you have to do is actually make connections. So within Atlan, we have our connector marketplace, which is going to have over 100 connectors into data technologies across all sorts of platforms, from Snowflake, Databricks, and I'll make sure to prove it. Here's our Fabric connection. So we've got Fabric as well. But what we want to do is be able to connect to those systems. We have where we'll first thing, like with Snowflake, for example, we'll first connect to the assets. And just so everyone on the call is aware, when I talk about an asset in the world of Atlan, I am referring to a tangible object, like...

5:10 Well, tangible, that might be the wrong word, but I'm referring to a structured object like a table, a column, a view, a semantic view, a skill, an AI skill. Those are the things that are considered assets. So everything within Atlan that gets an asset profile, even a glossary term, is considered an asset. So first we'll run and grab those assets, get all that information loaded into Atlan, and then from there, we're going to go and mine the information around it. So what we'll do is we'll go and look at the information schema. We'll look at the account usage and query history tables, pull that information in so that we can connect those dots from each transformation between maybe your bronze, your silver, and your gold layers. But also, we'll understand what questions are being asked.

5:57 We'll get popularity metrics by how many of these tables are queried. So even without doing any enrichment, we're already giving you a tangible way, right when you make those connections, to be able to search and understand who's using that data and how popular, how accessible this information is. So once we've then connected to our data warehouse, we then start to connect to the additional systems. So we're going to want to connect. So maybe you have Power BI. We'll connect to the Power BI, get those reports, those dashboards, those datasets, and then we'll go in and mine and get that activity information as well from Power BI. We'll also need to get insight into your CRM, so we can do Salesforce, crawl Salesforce, bring that information in, bring those objects.

6:43 That way we can see from those sources to the destinations. And then, of course, any of the ETL or tools that we can connect to, want to make sure we have that. So whether you're doing modeling with dbt or you're doing ETL with Fivetran, we've got all those connectors to make sure that you can tell that Atlan is aware of the full story of your data estate. So what does that look like in practice? Well, let's grab a Snowflake connector. We'll do a quick look at one. So when we look at a Snowflake configuration, what we can see here is, one, we're going to be able to actually choose a direct connection to Snowflake. So we'll go and make that connection. We just need those credentials, or we can use a key pair, we can use programmatic access token, or even do offline connections via other means.

7:30 We then will select our extraction method. So we'll pull from that information schema. You have the choice. So with Atlan, you have full control over what data assets that we are able to pull the metadata from. So based on the user or the service account you give us, that user is going to have the access for the information that you want us to have. We also, any of the policies that you have placed on your source systems, those take into effect. That user is the only input we have to get those metadata, the columns and tables and things like that. From here, you can even break down additionally where maybe you want to start with a smaller use case. You give access to that service account, but you want to start small. You can actually select what components of datasets and databases within Snowflake you actually want to add in.

8:22 So you can start small or go with the whole thing. It's up to you. You can include those by hierarchy. You also can include by regex or excludes. From there, we can have some other additional settings where we can import tags. We do offer with a number of our warehouses, we offer the ability to not just import the tags, like from Snowflake, but we also will have reverse sync. So we have two-way communication. So if you tag another Snowflake asset in Atlan, the next time this workflow runs, it's going to add that tag in Snowflake, and then if you tag something in Snowflake, the next time the workflow runs, it's going to add back in Atlan. So we do reverse sync there. And then, of course, once you've configured your workflow, you set that schedule. So you can then choose a schedule. This is going to be what works for you.

9:07 We want to make sure that we can support any of those time frames that you might want. Some folks need to run it every hour, some might need to only run it once a day or once a week. But it's up to you. Every single workflow has its own schedule that you can set and configure. That way, your assets consistently are updated, and they are getting the right information, the right columns are where, and even then, once loaded into Atlan, there's lots of ways that we can even apply consistent governance around those as well. From there, once you've done and connected to your systems, you start to actually get an asset. So anytime I refer to an asset, this is an asset profile. An asset profile is going to include all the information around this asset from owners, tags, connected terms, and we're going to dive a little bit more into this here in our next section for understand.

9:59 But once you've connected, these assets will then be compiled and built in Atlan. All right. So we just went through the basics of connecting, where we'll connect to our systems of record, our systems of data, our systems of knowledge, pull that information in, set those schedules, and then build out assets.

Daniel Bergquist

10:18 All right, step number two. From here, I want to show you how you can actually use Atlan to search and build data. So at this point in our Atlan journey, now assume that we've got the connected assets, we've added in our glossary terms, and now we have some information that we can actually go and search and start looking around, whether I'm a business user or an analyst or an engineer, I should be able to come into Atlan and find information. So we're going to do is we'll start by looking and searching as if we're a business user. So first thing we want to do is, when you log into Atlan, you may have noticed we have just a big old chat screen here. We actually have our Atlan AI chat built into Atlan, and it will go and search the assets and give you quite a lot of information around how that asset lives, what it is, and what it might be doing.

11:10 So in this, I'm asking the question, can you show me the definition to on-time delivery rate and its associated assets? Now, similar to any good old-fashioned cooking show, rather than have us wait and watch that, I ran the question already. So what it does is it goes in and it will find the glossary terms because I asked for the definition, and then it's going to find the associated assets to that glossary term. So we've got our definition where we can understand the data, the formula, what it is, and then it also will give us those associated assets. Now, for me, I'm going to look at the one that's going to be at certified status is verified. So I can actually open that up and start to review this asset just very quickly from asking a simple question to being able to find the actual asset behind the data.

12:08 We're just going to click over here on this screen. Should be loaded. All right, so then now that we're looking in our orders table, we can see, hey, I know my database, I know my schema, and from here I can then find out a bunch of information about this asset. Because the goal of Atlan is we want to make sure data is trusted. The key component of trust is that verified status. This means that the data owner or data stewards have gone and approved this asset to be used for reports or to be used however your company may like will define verify. This means that the owner of this asset said this data is verified and it's in good shape. From here, we can also see tags and then associate terms that are associated to it.

12:53 So we can look and actually, rather than having to jump back and forth from our glossary, we can go and see, oh, this is the term that is associated to this. I can see, oh, here's my definition of my metrics from Worldwide Importers. Here's my customer category. I've got this to see customers. I can find out that information very quickly. I can even see that these are the certified definitions of each of those because they also have that green icon. From here, I can look through the column information, understand what the columns are, their types and data, and then also review the README. Now we'll dive into the READMEs a little bit more later during enrichment, but just as a secret, this data was actually created and curated.

13:37 All of this was built by Atlan itself. None of us humans actually typed this. Atlan built this out and identified this and created this README for us. We'll talk about how that's done. But one of the other key components of this asset profile is that all of this is now in a machine-readable state. So now when we think about being able to consume this from an agentic point of view, those agents can easily come in, look at this, find the information, and consume it very quickly so that they can make those decisions, make those answers as your process is running in Azure, either whether it's talk to data analytics or operational. All right. Now, so from there, we looked at this was kind of a business user coming in from a chat perspective.

14:28 Well, there's other ways to look through Atlan. So rather than going into chat, let's say you're a business user or you're a user and you know the definition, that you know the term you're looking for, and you want to find information and associated assets to that term. Well, with Atlan's business graph, we used to call it our glossary, but with Atlan's business graph, we can come in here and look up, let's say we're going to look up GPO tier compliance. So what this enables us to do is to be able to go in, get the information about GPO tier compliance, see the actual formula that's been written, look at the selected SQL, and this is something that Atlan will also help you derive through our metrics agents.

15:14 And this way, when someone is talking to a talk-to-data use case, the SQL is already there. The answers to the questions are already there. So if someone has a question around GPO tier compliance, that's great. We have certified information, certified formulas that are going to be available on access so that that agent doesn't have to guess, it automatically knows. And then from that business perspective, we can go understand how it's supposed to be formulated. We can also see what it relates to. So Atlan enables you to build relationships so that we can look and find that, well, GPO tier compliance is related to our GPO contract to GPO tier, our on-contract spend, and even our tier usage thresholds. So you can break down and see that build-out of how this term relates. And this, again, helps our agents to be more specific and answer those questions better.

16:08 And then from our terms, so now I'm looking at the term. I knew I was looking for this term, but I wanted to find the assets. With Atlan, we can link these assets and be able to actually go through, check on the assets, and this is the actual asset that the definition and terms for GPO tier compliance are going to be used from. So now when we have those questions, we can be able to look up, we can look through chat, we can look through our glossary term, find these assets, and then from here, I can even break down and understand the lineage of any given asset, see where the data came from, follow the certification path, and walk through how this data actually originated from our source.

16:52 So in this particular example, we're looking at sources coming from Oracle, Salesforce, all going through, making those transformations through Fivetran, and then building out into Atlan itself. We can even see how the data is transformed and how the SQL statements were made. So if you have questions on how this data actually came to be, you can see those transformations and then even use Atlan AI to then explain that. The lineage component also gives us the ability to look and view the impact, so we can see the downstream and upstream versions of the impacted items. So we can look downstream, and we can look upstream and see how this data could be impacted. Now, I just saw a question come up.

17:47 "To be able to deduce lineage if ETL is being performed by two technologies, let's say staging layer with Snowflake and final layer with Informatica, etc." Yes. Etc. You guys got me. Yes, the answer is yes. So for example, if at some points in the stage you're doing and you're looking for in the processes would be like a Snowflake transformation, but then let's say the final stage from silver to gold, you're using Informatica or you're using modeling with dbt. We can go and actually showcase those stages as well. You'll also notice that we can see alerts and warnings, so you can propagate these alerts through lineage as well. So we do have a data quality component.

18:40 So just understand our data quality component is limited to Snowflake, Databricks, and Google BigQuery. But if you have data quality events that you've set, you can actually stage those and have those alerts automatically propagate upstream or downstream. Ravi, yes. The question is, "I assume that all those technologies need to have their own workflows to import the metadata?" Yes, and what's cool is, even, Ravi, let's say in the event that your particular tool is not one of our native connections, through our API, we can make a lot of connections into many of these systems. We also provide our own application framework for you to be able to create a workflow item on your own that will actually run and be in our marketplace, but it'll live on your instance. So yes, they would need their own workflows, and even if we don't have one, you can either use our API to ingest that, and then it would still show in lineage, because we have a very verbose lineage API with a lot of options, as well as use application framework to create your own.

19:52 But from the alerting perspective, you can actually have these alerts propagate downstream. So even if we're looking in a Power BI report, that alert can propagate all the way to the Power BI report. And then the last way that I wanted to show you how, this is not the last way that Atlan provides, but this last way in this particular demonstration that I wanted to show you is from a business user or from a user perspective, how can I get information about Atlan, not just in Atlan, but can I get it outside of Atlan? The answer is yes. So for example, this is a Power BI report. So you have a user in Power BI. This would also apply if you were in Snowflake, if you were in Salesforce, if you were in Looker or Tableau.

20:39 So we have what we call our Chrome extension. So you can get this directly from the Chrome Web Store. But what this does is you'll see the little Atlan logo here. This enables Atlan to be able to make a connection from the Power BI report or from the report in any of those systems in more than I listed, where you can actually see the information around this data. So we can see information about a warning that may have propagated about a pipeline failure downstream or upstream. We can look and get information on descriptions, terms, and information, the tags associated, ownership of this data asset, and whether it's certified or not. We can also go through and expose lineage. Find the lineage for this particular event.

21:23 We can even modify the lineage to show us the max depth. And then if we need additional information, we can actually go right back to Atlan and view that lineage directly in Atlan. But the intent is that whether your business users are in Power BI or they're in Atlan, they should be able to get this information as well. We also made this expose our Atlan chat, so if you want to ask additional questions like we did earlier or get additional insights about your data estate, you can do all that from the Chrome extension. This is also going to be available through our MCP connections into any of your favorite AI tools or harnesses. You can use it in Claude, Copilot, ChatGPT, where you can ask these questions, get this insight as well, while your users are using it.

22:11 We also want to build in, we can connect to Slack or Teams. That way, when there's communication that needs to be done regarding this, you can actually add a channel, you can message an individual, and tag those folks to start to get answers to questions. That then becomes available within our history, and so you'll be able to get a history of this particular asset right there, giving you additional context to make sure you can hit those trust signals. All righty. So from there, what we went through is we went through opening up, looking through things like a business user. We checked out the asset profile, looking at the owners, the certificates, the readme status. We followed lineage.

22:58 We've checked out our upstream, downstream columns and our impact ability, and then showcased how our glossary terms can be linked, as well as how we can use our extensions for Power BI and many other tools. All right.

Daniel Bergquist

23:12 Our next component of Atlan is how can we enrich at scale with our context agents? So like I mentioned in our intro, one of the key findings we found with our customers is just getting the data in an AI-ready state is time-consuming if you're writing that data yourself. But with Atlan, since we already are connecting to your data estate, we're mapping it through lineage. Now we can look at it and actually see the whole story. So when we go to build and enrich your data, or enrich your metadata, we're looking at it from the whole picture. We're not just looking at it from an isolated set of column names and type defs. We're looking at, well, this particular column, it flows all the way through and is used in a metric that's calculated in Power BI.

24:00 We know what key columns we can find. We understand how people are using it by bringing in the query history and query intelligence. So what we're going to do is look at our context agents, how we run it, and then we'll look at a couple examples of what we actually did. All right. So we first come into Atlan. This is back to our homepage. We want to go and enrich data. So in the life cycle of Atlan, at this point now, we've gone, we connected to our assets, we've connected to our ETL tools, and we've added in our glossary information. We pulled some metrics from Power BI or from Tableau. And then now we want to enrich this data. So from here, Atlan by default is going to start breaking the assets that you've brought into Atlan directly.

24:48 It's going to start breaking them down into what we call collections. These collections are going to be based on the related assets and related attributes that we've seen and how the lineage connects, because the reason we break them down like this for you is just so you're not trying to enrich all the data at once. You can focus on smaller components of your data estate and not necessarily have to go and target potentially millions of assets. And then from there, we can select these assets. And we have three components of enrichments that we do. So we'll look at readmes, descriptions, and this includes column and table descriptions, as well as our SQL intelligence.

25:38 So when we go to enrich a description, what we enable you to do is, one, we give you the ability to add in your own template. So you can provide your own markdown template, and then that allows you to curate how you want your descriptions to look. From here, the way we generate them is we look at, of course, the schema and column names, but we'll look at your lineage. We'll look at the query history, the related business graph terms or glossary terms, existing metadata around, because when you're pulling information from Snowflake or Databricks or wherever, you might have descriptions you already have. You may have tags that you've already built and owners you've already applied. So we'll look at that information, and then we'll also review the templates and the instructions that you give us, which I'll show you in just a moment.

26:24 So here's where you can add your new template. You can choose our default template, or you can add in your own to give additional insight specific to your users or your company. And then from here, you can also give us a prompt. So you can apply custom instructions to give us specific industry information, key strengths, your target audience for these readmes. So think of this as being able to give an AI agent prompt to our context agents to help curate this data and this information for you. And then we will also do, and so I'll show you an example of a readme we created. We'll look at the SQL intelligence, which our SQL intelligence, that's where we're going to look at the query history, analyze and kind of... Not kind of.

27:09 And reverse engineer the common queries that are being associated to tables or views, and then add those so that they become, again, searchable in a machine-readable format or searchable for humans to find answers to questions that maybe have already been asked. All right. So here's an example. So this is our orders table that we looked at earlier. And as I mentioned, I didn't write any of this stuff. This was all done by Atlan, by our Atlan Context Agents. So yes, we gave it descriptions for the table. We gave it descriptions for all of the columns as well. But let's hone in on this readme. So when we look at the readme, what we've done here is, of course, we want to provide a summary. What is it? We also, because we know your lineage, we'll give you the source systems and identify those. We'll look at, because we know all the way downstream as well, we'll look at the use case. Where is this being used across your environment?

28:04 And then also, because of lineage, we know what columns are most important, from the primary key all the way down to our salesperson ID and things like that, where we'll go, we'll give you the column name and then the why it matters. Why is this important to my business? Why is this important downstream? And then we'll also be able to identify some limitations, not that that makes the table bad, but there may be a view or something upstream that has more information or more components. So this is this table, you're limited by a few things. And then we also provide FAQs and downstream information, so you can know where this data is actually being leveraged at. And then from here, we've got our readme. This is our usage and intelligence. So each table that you run the table intelligence on, or that SQL intelligence on, you'll be able to get the information around the questions being asked.

28:54 So, we'll pull from the query history, those SQL queries that are being used that are most common, and then reverse engineer that. We'll also note the number of times these have been asked. We'll look for common joins and filters, provide you that usage information about whether it's people or whether it's, in our instance, service accounts leveraging it. This way, again, we're giving you additional insight into this data. All right. I did see a question pop up. Can imported data respect the source RBAC settings? How can agents serve only what users are supposed to see? Great question. So one, like I mentioned on the workflows, if you have, let's say you have in Snowflake, you've applied masking policies or you've applied those, all of that is propagated to Atlan. So if there's columns that we don't have access to or things, that's automatically propagated.

29:48 When it comes to RBAC within Atlan, you have a very, very fine granular ability to create what we call personas, which will then map out not just what assets can be visualized in Atlan, but also what even the UI experience is like in Atlan, where you can curate any of these menus that you've seen, curate even the metadata viewable from the asset profile, even limit what filtering options a particular user might have. So we have very, very heavy-duty RBAC permissions. We also will, of course, link to any of your SCIM and SAML connections, so that we can make sure to apply by any of your current groups or users that you have.

30:34 But yes, then you can scope those personas to the given roles that you require at your business. So then now that we've gone and enriched, we've looked at context agents, we've been able to go and run it. We've looked at the assets it's able to run, and then we can see it everywhere. So now, and kind of the way this would work as far as, well, how do I know if this data is good? This is where Atlan's going to do a lot of the heavy lifting for you as far as curating and enriching the data. But where we need a human is we need a human on the loop. Not in the way of the loop, but on the loop. So that's where, for example, for our orders table, where we get this information, and the way you know that this is approved is now the verified status.

31:24 Atlan's not going to verify things for you. We do have abilities to propagate verified status very quickly if that's something you want to do. But, what we want to make sure is that the data steward or the data owner can come in, rather than starting from scratch, read, review this information, and then add that certification stamp on there so that everybody at your company will know whether or not this data is certified and ready to be used. All right. So we've gone through enrichment.

Daniel Bergquist

32:00 Next step, we want to look at some governing ways. So this is where we can classify data, tag and propagate information, look at a couple policies, and then verify with those certifications. Before I jump, let me answer Kishore's question. Do you have any data on the cost of building a semantic layer, time, effort, people, and measurable benefits before versus after? That is a very intense question. Do we? Yes. Usually, the goal is to have an actual value between the 60 and 90-day mark. And then, I appreciate the smile. But do we have the value? Do we have that? Yes.

32:43 We do a lot of deduplication of your metrics and queries to make sure that we can do that. And then when we come in and you have implementation, the goal is that, especially if you're talking measurable benefits before versus after, where you're looking at that is how we can actually improve maybe agent responses. And so that gets set at the individual level. If you already have an agent, well, when implementation comes in and they go and work with you, you'll work together to set some thresholds to see what the measure is. We'll document the issues you're having. We'll document the goals for the improvement, and then that measure is going to be there. Can I share them right now? No, but if you get on a call with us, we probably can.

33:28 But I appreciate the question, Kishore. So for governing data, we're going to look at classifying data. We're going to look at tags and propagate those tags out, and then see how we can drive a bit of policy. Oh, yes, I can definitely. I'll showcase our term asset links as well, and how we can do that as part of this governing setup, and then how we can certify and announce information. All right. So this is an asset. So you might be like, "Hey. Well, Dan, I want to make sure this is certified and verified, and I want to follow through. Let's make sure this is classified correctly and the terms are linked." Now, I'll show you here as far as our term asset links.

34:17 You can link everything that I can show you, you can do on an individual asset basis, but I'm going to give you insight into how we can do it much faster as well. So for terms, even on a given asset, let's say you have access to add terms to an asset, you can quite literally go through and search here and search in this. You're going to see a lot of glossaries here. I'm an admin in our demo environment. We have a lot of glossaries. We meet with lots and lots of customers across all kinds of different industries, and so we have a lot of glossaries where you guys might not have quite as many. But you can search these terms and add these terms to your assets from here. We can also add them at a bulk level, which we'll show next in playbooks. From here, once you've gone and approved that data, you can go and make sure that your certification stamp is available.

35:09 So you can go and certify this asset, as well as set different stats. So you'll notice, you can even add in comments. All of this is going to be tracked in your activity and access as well. From here, we can also look through lineage and even propagate tags through our lineage environment. We can do that automatically or at a smaller level Where we can go in, for example, if I come in through our lineage, I can go and add a tag. Let's say I want to add a confidential tag. I go in and I add that tag, but then I set propagation. So now I went to the source of this information. I've gone through and I'm setting that hierarchy and lineage.

35:55 So this means it's going to propagate that tag through all the columns, as well as all of the tables in lineage. And then I can update that, save it. So now I'm going to do a big refresh. Go all the way back. So now we're going to check it out. So we've got our tag on Oracle, right? Got our confidential tag. Well, now let's look at the rest of them. So now that tag has been propagated all the way through lineage from that Oracle source all the way through there. Oh, yeah. Yeah, I'm showing you the slow way, so I just saw a statement in there.

36:42 We have definite ways to go through and we'll show and we'll look at playbooks to enrich at bulk. You can also enrich at bulk via the API. This would be if you wanted to come into an individual asset or set a tag just through one line of lineage, just mostly to showcase that we can propagate this tag. So this you'll see, even downstream in our next table, we can go through and look, and we'll see that tag propagated all the way downstream as well. But a lot of times, you probably have more than a few assets, more than one, more than five. So whether we're wanting to add terms to particular information, create a data product and add assets to a data product, you probably want to be able to act much faster. And so many of our customers will then use what we call our playbooks. So our playbooks, this is a way, think of it as a rules engine that can take action on. And these are all schedulable and re-runnable.

37:45 That way, when you're looking at classifying an entire data estate, you're not just limited to a small scope. You can look at the entire scope, multiple applications even, and not just be able to do it slowly. So, for example, one of our largest customers, they have about 60 million assets, and they had budgeted a six-month window to do classifications and tagging after they got Atlan. They then started using our playbooks, and rather than it taking them six months, they were done with all the classifications in three weeks. So let's dive in and give this a look. So we go to our playbooks. This is an example of the playbook.

38:29 This is for our delivery data, but we could really use it for anything. We're going to be able to use our filters, and we're going to sort and filter information. So in this, I chose to filter based on a database name, but let's say you want to filter across every system. Well, this limited me to this particular table, but if I remove this filter, you're going to notice we bumped up to quite a number more assets. We went from 24 to 205 because now I'm just filtering down to a table with a name of delivery or order. Now think about this from maybe a PII perspective. You could be filtering down and filter by names across every system connected into Atlan by name, address, email address, phone number that you want to have classified as PII.

39:15 And different asset types exist, or different asset filters exist. I promise I know how to click a button. I did, see? All right. But so we can look at the connections. So these are specific connections that you can filter down by connectors or asset types, whether it's a certified asset or even if you've already applied tags. We can even look for tags linked to glossary terms, domains. So there's countless options. And even regex you can use to do custom regex to make sure that you can fit the needs of your classification requirements. From here, once you've gone and filtered down, you're going to want to check to make sure that your blast radius isn't going to go too far. So you can actually verify all the different systems and all the different columns or tables that are going to be impacted by your filtering before you go and run that query or run that playbook. And then we can take actions.

40:11 And for example, in this action, we're adding a tag and note. In this tag, I'm actually already setting my propagation. So notice it's going to say, hey, just so you know, it might increase runtime. It's also going to affect more assets than those just selected. Because just like we just looked over here, when we added the tag to lineage and then propagated it all the way down lineage, that's what this playbook can also do. But you can set this at the bulk level. So you can set propagation, add tags. You can also add certificate status, description information. You can link those glossary terms. So maybe when you have delivery, you might want to add the on-time delivery rate. You might want to add that on-time delivery rate as well as a linked glossary term.

41:04 That way, all of these assets that are going to be applied here, they're going to start being curated and have that trusted and machine-readable data. And so now we've gone and added tags, but here we can also add owners, descriptions, add them to domains, add custom metadata values. So custom metadata values being Atlan, although we have a lot of out-of-the-box values, maybe there's certain information that you specifically need for your data or your requirements or your compliance obligations. You can add those in. That way, you can push out and very quickly push information that's required for all of your data assets, making them trustworthy and searchable.

41:48 All right. Oof. I'm talking longer than I thought. It's 10:43, guys. Okay. So from there, I'm going to kind of breeze through policies. Policies are also something where you can add in a big policy definition. This way, if you have PII, for example, you can create a policy, scope that to tags or rules, and so that this policy then applies to all of those assets with those certain specific tags. And what that looks like in action is we'll actually have linked policies, and then in the asset itself, your users or your agents will be able to see the policy and data applied to that as well. And then you can take action based on those policies. All right. So we governed, we've looked at our information, classified information.

42:34 We've watched a tag propagate. We can drive policy from tags. And then we've seen certification out. So I'm sorry I breezed through that really quick. All right.

Daniel Bergquist

42:42 And last but not least, I wanted to show you an example of what Atlan can do when that context gets connected. So, for example, we're going to look back at this GPO tier compliance metric. So think of this as a diagnostics company, and they're doing a group purchasing organization, so health insurance of some kind. Where are the compliance and how do we meet that information given that GPO tier? Well, you might have the definition, but you also may want to actually get the data and the numbers for that particular asset. Well, what you can do is whenever you go and connect Atlan to, in this example, Claude, we can start asking questions and getting that information and actually having Atlan go first.

43:34 So in this particular example, I asked the question, which customers are below their thresholds for their contracted tier? Well, from here, I went and first thing it did was it pulled information from Atlan's MCP. And then after understanding the right table to query and the right query to run from Atlan, it went and ran that, gave us the information where now we can actually pinpoint down the different customers that are within that contracted tier and the ones that have met it, the ones that have not met it yet, and get that insight. And then as a business user, you could continue, or as an executive, continue to ask questions. I do have a few more options as far as questions we can look at, whether this same thing, what's our on-contract versus off-contract spend?

44:20 It went and found the definitions for on and off-contract over in Atlan, pulled that information, was able to deliver that insight to you. And so and then just so you know, for those of you that maybe came to last week's context conference- There's even more that Atlan's bringing to the table as far as how we can gain and build context from your actual insights that you're getting. So where we're pulling in traces and getting in more and more enrichment of context. That's not what I'm demoing today, but please go and look up that link for our context conference. It's recorded and it's available for everyone. I highly recommend it, and you can see where Atlan's going with our Trident and our agent registry options.

45:05 Okay. I really appreciate everyone's time. I'm sorry I went over by a minute, but let me look. Let me see.

Daniel Bergquist

45:11 I think there's a question. At the beginning, you mentioned two use cases, operational and analytical. Can you give an example of each? Is it different type of-- Oh, that's a great question. So one is definitely different than the other. So an analytical agent could be a talk-to-data agent or something to analyze documents or analyze something, and so providing context. So a lot of our customers, they start with an analytical agent or a talk-to-data agent because they're trying to get certain questions asked or answered and certain context added. So that's going to be our first, and almost everybody starts there, where they have a specific question they're looking at or a specific use case that they want people involved in. Versus an operational use case, that is an instance where you're wanting an agent to actually make and be autonomous in its decision-making process.

46:06 And so we have the ability to provide that context so that that agent has the insight. So for example, in a refund situation, we've worked with customers to automate how a refund works. Well, you have to explain exactly what's required in order for a refund to work. It has to have access and the understanding of what each of the datasets means in order to make that autonomous behavior. So maybe a refund or an automated refund policy would be your information around an operational type example.

46:49 And then we've worked with a lot of customers, including a lot of different food industries and other things. We have a customer that's big into foods. For those, they needed to understand, their salespeople need to understand what orders and what kind of insights were coming in, both an internal sales agent and an external agent, so they can make deliveries and promises and things like that. Can I show an ontology graph looks like and how our different business concepts connected with across terms? Yes, but I'll try to be very fast. So for example, let me go to one that'll give you a real pretty insight.

47:45 So we do have additional updates to this. This is our current-- No, it's on this one, sorry. This is our current working of relationships. We do have with what's releasing, it should be into next year, we do have a different type of agent, but I'll show you what we have as far as relationships and ontology type information from here. So for example, if you're looking at a customer, we can build in relationships, and this can be done through mappings of different formats, leveraging our API, just depends on how you want to do it. But we do actually have the ability to build a full mapping of relationships from customer. Well, what makes a customer?

48:30 It has 11 properties from email address, date of birth, age. There's nine metrics associated to your customer. So again, enabling it to know what metrics make a customer, and even one of my favorites is, for example, a customer is going to earn a loyalty transaction. Loyalty transaction is going to be earned on customer orders, et cetera. So all these relations can be derived. It just depends on how you, for this instance, how we're populating that. Is it from fields in Power BI? Is it from an out formatted ontology that we go and map to our API? Versus what's coming is our ontology agent, which is not something I can demo today. That's something that we'll be able to build out, that you'll be able to get more demos of here very soon.

49:17 All right. Well, I went five minutes over. I think I have done my darndest to answer a lot of the questions, except for your really hard question earlier, Kishore. I was vague on that one, and I'm sorry. But I hope you all have a wonderful rest of your day, and thank you so much for joining.