All righty let's get going. Welcome to another Interworks webinar. We are gonna be talking about the top five AI use cases that every business should target right now and a guide on how to do it. You can see our little icons for each of these use cases. For those that haven't been to an Interworks webinar, I'm Robert Curtis. I generally present most of these, although I do have guest stars appearing from time to time. My name is Robert. I am the managing director for Interworks looking after Asia Pacific. I've been with Interworks for about twenty years. My employee ID is number three, so I've been here for quite a long time, and I'm based out of Melbourne. I've gotten an opportunity to work with a lot of you, and I see familiar faces in the chat again this time and some new faces as well. So welcome back, old friends, and hello to our new ones. A little bit about Interworks. Very simply, you could say we do data strategy, data solutions, and data support. A way to double click into that a little bit more detail. If you think about how your data travels through your systems out to your end users, it's a good way to think about what we do. So we can start with the vision, the plan, the roadmap in the strategy phase, building the foundations across platforms, governance and data. Governance could be the actual technical implementation of those policies, or could be the frameworks and ideas themselves down to solving problems and driving value. And that could be through analytics, AI, data science, machine learning, advanced analytics, all kinds of things. And then once we get all of those foundational things done and we get real value going, we can continue to help you grow and sustain. And ways that we can support you, we can support you with particular applications that you don't have the appetite or interest in supporting. There could be aspects of that application. So for instance, it could be, your analytics server or it could be your pipelines inside of your data warehouse. We do all of that. We also support the individuals in your organization and we can do that through mentorship and training for particular skills or or applications. And we also support full community. So if you want us to look after your community of practice, which could be ideas on how to engage your community, build the culture, raise data literacy, all of those things, we do all of it. So however you need us, whether you need a full white glove solution or bits and pieces to fit with the things you're already doing, we are happy to help. We've been doing this for quite a while. We were founded in 1996. So if you're doing your math, we are at thirty years officially next month. We're having a big shindig on either side of Australia as well as all around the world to celebrate thirty years of Interworks. In those thirty years, we've gotten to help a lot of folks. You can see we've got a lot of names off of the Fortune one hundred. A lot of the tools and partnerships that we support actually have come to us for work as well. So we've helped data companies do their analytics, and we've helped analytics companies do their data, as well as customers across every vertical, every size that you can imagine. A lot of those stories are told on the interworks dot com website in our blog and in our case studies. Speaking of our blog, it is a great place to get to know us as well as to get ideas on what to do with your data, your governance, your strategy. We get somewhere around 3.5 to 4,000,000 page views per year for our blog. We've got thousands of articles on there, white papers, videos, everything you can imagine. And I've got an award down there, the Forbes Small Job, but we've received a ton of awards in our thirty years. I think somewhere around twenty partners of the year, probably way more than that by now. Don't even keep track anymore. So we've got a lot of pedigree. We've got a lot of experience, and we'd love to understand how we could help you. So on our particular webinar, let's get going. So I've got our five little faces here. It makes me think of Mandarin in his five rings. I'm a nerd. I understand. But, there's some things that you guys can do today very quickly that's gonna get you one, doing AI effectively and successfully, and it's going to accelerate all the other AI that you do. Hopefully you have some indication of why AI is so important. If you've been to any of the webinars that we've done this year, I talk a lot about it. I've given a lot of quotes from industry leaders. This is not an innovation cycle. This is a culture shifting, society shifting paradigm. The quotes that I've seen compare it to the invention of fire or the wheel. That is just how powerful AI, promises to be. So specifically, what we're gonna be talking about are things that you guys can do that are low risk, high reward for AI. So some quick ones. There are ROI positive view cases that are likely already sitting in your data estate in your environment that are just waiting to be discovered, planned, prioritized and properly implemented. They're there. They just need a little bit of insight and maybe a little bit of expertise. AI often fails because people choose tools and don't map them to specific problems, or they don't have departments or owners that are championing it. So it's we're building a tool without a purpose and oftentimes whether that's AI, data, or or any other arena, you are not gonna get back your ROI. You're not going get back your investment. And the order in which I'm presenting these, I'll clarify at the end in terms of sequencing. We'll start with the most exciting, sexy ones. But the way that we do these does matter because it'll be incrementally easier. So the order in which you approach these things is important. But we're gonna start with a couple questions. And undoubtedly, either you as executives have asked these questions of your team or you as data professionals have been given these questions. But where to start? We like AI. It's exciting. We know our competitors are doing it. It's all over the market. But what AI case should we start with? Is it the one that we're the most excited about? Marketing, finance, product. They've all got their big ideas. If we do this, it'll be a game changer. It's exciting. Is that the one that we should start with, though? I'm gonna contend that those are often the ones that are the most complex, require the most significant investment in data and governance, and often have the highest risk. So I would say maybe we don't start there. Maybe we start with data specific use cases first. And when I mean data, of course, all AI is constructed on data. We have to have governed curated datasets with great context and lineage and all the classification for AI to work. But what I mean specifically is solving problems for your data DevOps, your data team. That is what I'm talking about. And there's plenty of benefits. So first, everything that we want to do with AI depends on data. So if we build foundational stuff using AI to do it, it accelerates everything that we wanna do in the future. It means that the development cycles are shorter, the chances of success go up as a percentage, and the expertise and knowledge share within the organization is already established and ready to grow. Speaking of expertise, the engineers that are building these things for themselves to accelerate themselves, it's going to be invaluable in terms of how they build things for the rest of the organization that have a higher risk of failure. The blast radius of failure and the visibility of that also might be much larger. So let's get good at it in limited trials first, low risk, fast proof. And again, these are the people that are the most data savvy, that are the most AI intelligent. And so it means faster learning curves, greater user adoption, and easier change management if we start with the data folks first. Last thing, it is a compounding investment. So everything that we invest in the semantic layer or governance all of the cataloging and documentation behind this, the quote unquote context, all of this stuff pays off one hour too many as you start to look at additional use cases. In particular, Talk2Data, which is one that we're gonna talk about, is a significant benefactor of these. To also entice you if we have any folks that are finance related or have to defend benefits and spreadsheets, I went up and looked up some industry averages on how quickly you can expect a payback on investment for different types of AI solutions. So if you implement the big, broad, sexy enterprise AI solution, it generally takes about forty eight months to get back to your payback. We spent all the money to build it and the return four years. If we have very targeted automation, here's a process, we're going to automate this step of the process, maybe it's unstructured data into tabular format, whatever, generally that takes about eighteen months to pay back industry averages. But if we put the hands of data people on AI solutions to accelerate them, the industry average there, a quarter. And again, if we invest here, all of the downstream stuff, the targeted automation, the big broad enterprise stuff is faster and more likely to be successful. That's one of the reasons why I'm like these are the five ones that you should start with and you can do today. I picked five that are patterns that are well discovered round and pathways that are well trod, these are highly likely to be successful, less risk of experimentation, you're not on the bleeding edge of what AI is capable of. And to start, you should start where you're comfortable. There's another presentation I did about, your AI risk tolerance. And so I would classify these as low risk. The five use cases, talk to data, using AI to accelerate data pipeline development, using AI to help you with data classification, which is critical when you start to expand the different types of AI solutions that you're gonna implement. Data quality monitoring, again, bad data gets you bad answers. And then AI assisted metadata and lineage, all the stuff that people used to hate doing, didn't wanna spend money on. AI can help accelerate that. And all of these are super critical, particularly the last four in building AI solutions for the future. So as we get into these, you might have questions, you might have comments, please put those into the chat. If you want, could throw them into the q and a. I've got them both open here. I generally just cruise through the webinar just so I can kind of keep my rhythm and then come back around to these at the end. So use the webinar chat q and a. And if I don't have time to answer them during our allotted hour, we always if there's particularly interesting questions, we'll answer them and send them back out to everybody. I guess the other thing is if you do want this recording as per usual with all of our webinars, it takes about two days or so and then we have it right up on the website. So if you want to share with friends or go back and review a point or grab screenshot something that I've got here, by all means you are welcome to do that. We'll alert you when it's up for the folks that registered. So let's get started. Talk to data. I've spent a lot of time on this. Two, three, maybe four different webinars talking about BI 3.0. And this is what I'm talking about. Talk today is an element of BI three point o. So we're gonna do a little bit of a review for those folks that weren't a part of those conversations. Number one, BI two point o. This is the land of dashboards. Everybody can build a dashboard. And to get your answer, you need a dashboard. So this is this idea of dashboard sprawl. And most organizations are dealing with this right now. There's the occasional folks that I get to talk to as a result of my role at Interworks strategy and other things like that, where I find people like, we just went through a really difficult, process where we got ourselves down to 50 dashboards. High five to those folks. But the thing that's interesting is just how quickly these dashboards accumulate. The average size of the enterprise dashboard footprint is around five to 10,000 dashboards. This is the result of self-service and these were amazing tools and they were super super important to get where we are now, but they're considered legacy at this point. If every answer requires a dashboard, you're behind. And quite frankly I've seen I've talked to organizations that are 10x 20x this number. They accumulate very quickly and you can imagine the organization the organizational burden that it takes to maintain these dashboards and the user conundrum in trying to figure out which one actually hits the right one to get your answers and the staff and resourcing it takes to maintain all of this. All of this has got bloat that AI is going to make way more efficient. So what is Talk to Data? It's known by a couple different names. You could see it as T2D in short or more professionally, more scholarly is conversational AI. But basically what it means is you can use natural language to talk to an AI agent and the agent will then query that data very, very quickly and then come back with answers. You don't have to write SQL. You don't have to wait for a dashboard. You don't have to have any technical knowledge. It does help to understand how to ask questions with context, but very quickly, that will be something that every business user will intuitively know. Just like every person in the world knows how to search Google. That was a skill that was acquired when search engines came out. So in summary, self-service for everyone, legitimately everyone. Self-service in BI 2.0 land was if you want to take the time to learn how to build a dashboard, then self-service for you. But conversational AI talk today to me is if you can write a question, you can get an answer. And it also means you have the ability to ask additional questions. So here's my first question, get an answer. Ah, that's interesting. What about this? More context here. Can I dig a little deeper there? This is what Talk2Data offers. As an example, in our interface, and this might be Snowflake co worker, it might be Claude, it might be Sigma, we might ask ourselves, you know, what were our sales by region q three? And if we have good data, then we could get good results. And this is just sort of true of any data product of which AI is a data consumer. AI might come back and say, here's your revenue by region. And oftentimes, it'll ask you, would you like to compare q two? Would you like to do this? And there'll be a prompt for you to continue to interrogate to get better and better answers. And if there's something that you like, AI is really good at helping nontechnical people build more persistent views or applications that help them get this data that they think is useful and then come back to it. So I kind of mentioned this already, but in terms of the gap of what exists, 5,000 is actually pretty stock standard. I think more likely we're looking at 10,000, particularly when you aggregate across tools. 5,000 might be in Tableau. There might be another 5,000 in Power BI. I've seen worse. If you do the work to trim down everything in there from test cases, sprawl, we created six different versions of this dashboard for the six different business units that probably need it, but we're not super comfortable with our RBAC and giving, making sure these people can't see that data, etcetera. And you limit it down to executive reporting or operational reporting, really 50 is about what most folks need, A tight 50. Everything else should be pushed into a different tool. That could be Talk2Data, it could be a data app, but you don't need as many dashboards as we've got now, a 100%. So to make TalkToData work, there's a couple of big things that you need to be true, and then we'll go into a little bit more detail. Obviously, you need a real semantic layer. And what I mean by real versus imaginary is most people assume, well, I have a dashboard that gives me answers every morning, and the answers are right. So that must mean there's a semantic layer. It does not mean that. Most organizations built the equivalent of a bronze layer data asset, and then the analytics people cobbled together the rest in the analytics layer, in the workbook, in the worksheet to get the right answers. But from an AI perspective, that's not great. We want the actual curated data, the business logic to be in a consolidated location that AI can leverage easily. Because it's again, not a human consumer. It wants all of the logic in one spot. That's what I mean when I say a real semantic layer. Governance that travels. So depending on where you're interacting with the data and what other things might be brought in to help contextualize the answers, the governance needs to be pervasive and consistent. Unstructured data is a big part of this. Data quality and lineage. Well, agents need to understand how these things interrelate. They need to understand actual definitions of what a customer is, sale is, revenue is, etcetera, etcetera. The good news is is that for for all three of these, the out of the five use cases that I'm presenting, three of these address these dead on straight straight to the point. So AI is going to help you build these to help you get to a Talk to Data AI solution. So some additional things to think about. I already talked about the semantic layer, but metrics, dimensions, and joins in the governed layer. So we're getting everything up to gold and then building on top of that and then adding contextualization next to it. If you go look at the last webinar we just did, which was AI at scale, as well as another webinar talking about the preeminence of the semantic layer, I go into a lot more detail on this. There's also a white paper I wrote that you can find on interworks.com. Glossary, a data dictionary. The business needs to agree on basic business terms and what they mean so that AI knows and then AI can respond accordingly. From a scope perspective, I think this is something that's a little bit of a think of this as a best practice or a little secret ingredient. There's going to be different places that you should start. And I mean, let's say, let's take business domains like we've got here. Your finance people are generally gonna be very data savvy and most likely that data is gonna be pretty clean. Yes, they're doing all this great stuff in Excel land. It's inevitable, it's finance. But there's gonna be different data domains that are stronger than others. Start there. Do not try to do everything for everybody all at once. Test, limit, govern, train, and then once you get comfortable, then start to expand. It's much easier to deal with a single fire rather than setting fires all over the business. And and and that's just a natural part of testing and validating that the data is correct. From an evaluation standpoint, 50 to a 100 real business questions that you think that domain would typically answer or rather ask. And then the answers, let's make sure we have a dashboard that we know that this is the right answer. And then we can go and test and validate. And you'd be testing and validating those questions over a period of time to see how the engine, how the AI bot agent, whatever changes to match with the real answers. Do that with a limited test group from your domain and then expand slowly as you are sure the data is accurate. And then guardrails, row level security, all of these sorts of things. Also want to put limitations in terms of how people can spend credits because you don't want to blow out a budget on a POC or spend way more AI consumption or data consumption than you were intending. So monitor that really, really carefully and put guardrails in place. Number two, AI accelerated pipelines. I am really excited about this. I've been spending a lot of time thinking about this, and I'll probably chuck another white paper together next year to talk more about this. But to understand what AI is going to do for people that are building data pipelines, let's start with software engineering first. Software engineering was the natural place for AI because AI can scan the way that programs are written and find a lot of reusable stuff. There are certain patterns or data sets or information basis that AI is really, really intuitively gonna be good at. So probably less good at navigating New York City as a taxicab. Autonomous driving has been more difficult than say really good at legal precedent and in the law studies. In a similar way, AI has been much better at software engineering and slower at data engineering. Software engineering is much easier for AI to understand. And some rough stats, industry averages, 55% faster task completion in GitHub's controlled Copilot study, 21% to complex task. An estimation, AI roughs roughly 20% to production code. Vibe coding is now a universal term that even my 10 year olds know about. We have people on our team in sales and marketing that are building applications that have no coding experience whatsoever. And this is a rough number, 10% sustained organizational wide uplift across millions of developers. That's about four hours saved per developer per week. So we're talking four million hours minimum, probably more in the neighborhood of 20,000,000 saved per week from a software engineering acceleration from AI. I'll tell you anecdotally, mentioned, as I've been into the works for a very long time. One of my colleagues that used to build applications for us, and now he's in our executive suite, used to build stuff and he said, Rob, just so you know, that application back in the nineteen nineties, it took me three months to build. I could probably build in a weekend now. That is significantly bigger than those numbers, but I wanna give you industry averages. Anecdotally, we're seeing a 20 x return on software engineering. Is that true with data? Not yet, but it will be. I'm gonna break this down for you, but I want you to understand some basic concepts first. So medallion architecture is think of it like a assembly line of how we prepare data. So bronze, we stage raw data into a into a single location. So we're bringing it through APIs or from this database or whatever and dumping it into a place where we can get it all together, and then we can start to do more things to it. Silver. Once we have it all staged together, then we can cleanse it, de dupe, conform the data. From a from a semantic layer perspective, silver is a minimum. Then gold, you start to add facts and dims, start to add more business logic. This is less abstracted than say bronze in particular, but silver as well. And then when you get to the semantic, that's when you're virtualizing the gold layer. You've defined all the relationships. You've got your metrics locked in. You might even be partnering it with context. So it's got this really rich business ready, user ready consumption layer. So think about this as like the assembly plant for data products. So right now, and these again, industry averages, I'll tell you in conversations I've had with customers as well as vendors, there are people that are significantly beating these percentages. People are seeing a 20% acceleration as a coding accelerator. So if you think about something like Cortex Code and Snowflake, which is also known as Coco, Coco is fun to say, on average, you're seeing a 20% uplift. So in other words, if every four hours I spend, I getting a fifth hour efficiency. And there's people that are significantly crushing this number even more than that. So coding accelerator, again, AI draft SQL DBT pipeline code, whatever I review test and merge. Analytics engineers, this is important. Because AI now makes it easier. I don't have to have fully credentialed data engineers as the exclusive people writing to my semantic layer or building data products. I can take data analysts that have good understanding of SQL, and more importantly, a clear, strong understanding of their business and the domain logic that they're working in. Generally, they're embedded in the business unit and they can write to the semantic layer. Think of this as sort of a data meshy style approach, but the elevation of analysts into engineers from an analytics perspective. AI is really empowering that too. Let's get into the individual layers here. So Bronze. Bronze is the one where AI is most suited for right now. Because if you think about it, we're just getting data out of source systems through APIs and staged into whatever warehouse we want it. So let's say we're taking it out of Workday or some financial system or SAP or whatever Salesforce, and we're dumping it into Salesforce. Most APIs are really well documented, which AI really appreciates. So right now, AI is probably looking at like a 40% acceleration of just getting that data staged in. As we start to look down the track, that number is going to significantly improve. So let's say, I think anytime I put an estimate on here, AI always goes and does it in like 70% the time it would have normally taken. So if I say in four years, and undoubtedly I'm going to be talking about this in three years, it's been, hey, we've got the 80% efficiency. So a copilot experience for common sources, common sources are the biggest, most popular repositories of data that you want to get out. Landing routine work is generated. Engineers are approving rather than having to primary authors. That's where we're going to get that 80% efficiency. So instead of us spending four hours to do five hours worth of work, that's us spending an hour and getting five hours of work. On top of that, all the automation that comes along with that, quality checks, optimization, debugging document, all that stuff, optimization, you name it. Now what's interesting, and I won't cover this and this is just the role, we're gonna talk about engineering here, but there's a significant role that your solution engineers, your solution architects, your data architects are also gonna play. Look for that in a future webinar. When we get to silver, we still find really strong AI opportunity here. Gold is where it'll start to become a little bit more difficult. We'll talk about that in a second. Data silver is where data comes to become trustworthy. Trustworthy, conformed, more usable. And we could even point power users from our analytics at Silver, they'd probably be okay. We're looking at that 40 to 40 to 60% efficiency. And again, just like Braun's 80%. AI suggested relationships conformity dedupe's human curated. And again, as we do this time and time again, the AI will get stronger because you're gonna be training it. Proactive recommendations, prescriptive, all of these things AI leading with human beings being the beneficiary of it. Gold is where it gets a little bit more complex because gold is where we have to make business decisions. We can't use documentation exclusively. We can't use precedent exclusively because human beings have to decide what's important to the business. And then that gets factored in how gold, your gold layer gets built. So your marks, your metrics, your semantic definitions for BI, talk to data, it would be a heavy consumer of this. So when we look at the opportunity, it is a little bit less, but still significant. Again, we're talking 50 to 70%, and that's looking down the track from 2026. When we get to '27, '28, '28, those numbers will undoubtedly change as AI gets faster and better. So the team's role here. Engineers own the semantic layer and its governance AI executes. This is what turns Talk the Data from a demo into a product. I don't know if I fully agree with that. I mean, think the idea here is that talk to data requires a significant amount of energy behind the scenes. It's like the legs of a goose swimming nice and peaceful, a swan swimming across the but it's underneath that the legs are working. That's what Talk to Data is today. So I wouldn't call it a demo because it is a functional product. But what this is saying is is that AI will be able to do a lot of the engineering work. So the pedaling, the duck's feet underneath the water, will actually be pedaling less because AI will be doing more of the work. But very, very exciting. Cross bronze, silver and gold 20 to 30% efficiencies more so in bronze than gold, but they are coming and they're getting faster. If you talk to the people that have been using these code accelerators like COCO, you'll find that they were trialing, fixing, debugging a year ago, and today it is off to the races, and it's only gonna get better. How to accelerate pipeline safely? Profile the backlog, repetitive connectors, cleansing modeling work is where error removes all the hours. So if we have to do, if we have to pull this table from the same through the same API, you know, there's thousands of tables in SAP, that type of stuff. Repetitive work AI is really, really quick at learning it. Naming conventions, modeling patterns, test templates, all those things that you can establish the precedent that AI can just follow, great. Give it the context and AI will shine. Documentation and standards and that kind of stuff also help consistency. Which pipeline goes first? Again, just like anything like what we talked about with Talk To Data, start with the stuff that you already know the most about. Don't start with the stuff that's high important experimental that we're learning as we go. Save that stuff for later. You want to establish the patterns, get a proof of concept, really validate that you've got is working. And then you can start to expand to higher risk effort. Sensitive data classification. So the next three really talk about governance. And it's important that I put a little bit of a color on on governance. So prior to about 2024, governance was the thing that every company knew they had to spend on. And some some verticals like healthcare or government had to spend more because of regulatory and compliance. But nobody liked spending money on it. And most of the time what organizations did is they bought a tool. So I'll talk to people and I'll say, okay, what's your governance look like? And like, well, we we need to do this, this, and this, and this. And I'm like, what's your tool? And they'll say, x. I don't wanna pick on any governance tool. And I'll say, do you guys use it? And inevitably, 99% of the time, they're like, no, we don't. But the reason that, governance is so important now is because without governance, AI is fraught for disaster. You must have governance now to limit the data that AI consumes. So you don't want it to pull in PII information. You have to give it guardrails so that it is using it within the policies that you have for I'm restating myself what I just said. I'll switch to you for things. So in terms of data quality, you have to make sure that the data that you're receiving is good. You want to put metadata around it so that it understands the relationships and classification and lineage. All of that stuff is super important. So the next three use cases that we're gonna do address those things. And by having AI do stuff that human beings hated doing, it's a win win. Sensitive data classification. So detecting and tagging personal, financial, health, confidential, children, information, sensitive all of that stuff is what's in scope. Manual audits is how this was generally done or manual scans reflect only a point in time and almost immediately they're stale because the data estate is quite dynamic. You can do regex catches. It might look for specific things, but as we've noted here, there is a lot of ways that sensitive data might pop up. On the very next slide, we'll talk a little bit more about that. Why it matters. So compliance exposure grows with every unclassified source. Regulatory standards are going to only increase for AI as well as brand, financial, and performance ramifications if you don't do AI correctly. You must classify this data, make sure you're not using it inappropriately. Financial penalties or brand damage could follow. So the risk. It is no longer just what's in your database that's actually at risk. Most sensitive data, most of your data period sits in unstructured files. And and now AI data sources and AI can consume all of that. So roughly 80 to 90% of your data is actually in unstructured sources, files, documents, PDFs, whatever. And on average, about 25 to 30% of that is actually sensitive data on average. This is done by studies and etcetera. So that means there's actually a blind untapped minefield that you could potentially be walking into if you start to allow AI to go through and add all the value it possibly could without data classification. So what AI can do here? AI scans data as it arrives. It looks for specific things that you flagged that might be personal data sensitive, financial, HR, payment, salary, whatever. Tags it automatically and then goes into your cataloging system. Compliance is current as new sources land, auditing prep time shrinks because AI is doing it consistently. And then you apply it to an automated governance, RBAC privacy security plan. The other thing that's super important about sensitive data classification is you really wanna do this pretty early. So you'll get a sense. You're probably guessing that if I were to sequence these use cases right now, this would probably goes up pretty high and it should. You don't wanna do Talk Data without real control and understanding of what's fields you might have that Talk to Data is actually gonna be leveraging. So define your sensitivity tiers before you do your scan. Obviously, you could do a bit of a discovery, but I think having that good understanding of, hey, these are the things we're worried about, and this is how we're going to classify within our RBAC or data governance policies is important. Classify at ingestion. If you can integrate this as part of your bronze, silver, gold, you know, sort of the data dev ops cycle and automate the classification as well as some of the other things we're gonna talk about, it means it's much, much easier and it's a part of a process for some sort of audit and audits are always gonna be out of date. Enforcement is important. So if your system picks something up that says it's PII, you don't want it to be flagged. You want it to automatically be swept into permissions and authorization on who can see it. And if there's things where it's uncertain, so these low confidence tags or there's a percentage of, hey, this looks suspicious, but we're not sure, then you overlay humans to go and actually validate. As a result, human beings are doing less work and they're adding the deciding vote with human context. So these are all things that are super important. Use case number four, data quality monitoring, the whole thing garbage in garbage out. We need to have consistently authorized, verified, excellent data to power all of our analytics. But on top of that, and more importantly, because AI cannot make human intuition like, oh, that number looks wrong. AI will give you an answer. Data quality becomes super, super critical. So some static rules for AI, let it learn the baseline distributions, freshness relationships so that normal it understands versus it's just having to start guessing from scratch. Prioritized by impact. So lineage is super important so that if it says, hey, this thing is not right, but it gets to see how it impacts everything below downstream of it, it can elevate it and say, hey, this actually affects multiple reports that go to executives or go into other automated systems that have agents acting off of them. So let it see what's downstream so it can prioritize. If you allow the AI to make corrections, then you don't have alert, alert, alert, alert, alert. And everybody has been in an organization where you end up with a 100 to 200 extra emails a day telling you of something bad that's happened and you'd start to tune it out. There's all sorts of research on road signs. If I put 45 road signs on that say, be careful, people don't pay attention to any of them and accidents actually go up. Same thing is true when it comes to alert monitoring. Fewer incidences, reaching users, faster root cause, less engineering time. This is one of those ones that quickly pays itself back. So in terms of quality monitoring, start where the data is most used. This is generally where the most pain is going to be experienced. So identify those 20 to 50 tables that feed executive reporting, any AI use case, anything that has significant blast radius and reusability, I'd start there. Let the AI learn before you throw it out to start doing stuff or alerting. Make sure your governance policy has technical data owners mapped to those assets as well as data stewards so that those folks can be alerted when something needs to be reviewed or there has been something that needed critical implementation or intervention. Again, connected to the lineage so that people could see the downstream impact. And this is true for every data person, but report the saves have celebrate the wins so that you can actually track an ROI on, hey, look what the AI did for us. If you just let it operate blindly and only look at the things that went wrong, you're kind of missing a bit of the story there. The last use case is metadata, lineage, documentation. AI is phenomenal at this. I use it all the time. So manual documentation versus AI assisted. The old way docs were written once and they were stale. I wouldn't say within a month. I mean almost immediately. And the truth is, no one reads this stuff. You should be writing your confluence pages for AI so that people can ask questions and AI can get them the answers they want versus the needle in a haystack approach of we've got 12,000 pages of documentation. Good luck. Lineage drawn by hand by the audit or crafted in a tool. Tribal knowledge lives in three people's heads. You're lucky if it's three, most of the time it's one. That person's never allowed to take holiday or PTO. And out of the 400 tables, who knows which one does what, which one is experimental, which ones actually relate back to these. The understanding of how these things all flow together, quite difficult. Using AI descriptions generated and refresh the scheme is changed so it's a real time document. Column level lineage parse from code automatically. Business context captured, searchable and shared. Again, write Confluence for AI so that it becomes a searchable context engine. Every asset has an owner, a description, and a trust score, which AI can help you assign. It can help you write. And then it makes everything that it uses downstream beneficial. So some other tips here how to document the estate with AI, generate first then verify. Let it generate the columns, have some human editing, and let it learn in terms of how you are guiding it. Parse lineage from code. And again, this is another one. This is true for all of governance. And I think this is a big problem with traditional governance tools, is governance generally goes in existence in an application. And that might be Calibre, Alation, Atlan, whatever, but people don't work there. So if you can integrate where people actually do their daily business, so in the way of working in the pathway of them actually doing stuff, that's where it's going to be the most beneficial, which is why building an AI interface that it can see all this stuff becomes so important, particularly with a Talk2Data. And Talk2Data doesn't have to be, I'm asking questions at the business. It could be asking questions about governance or the validity of the data set or the freshness or errors or data quality. All that stuff put it into the interface. So let's talk about since we've gone through all five of these and I've given you some tips, I'm going give you some idea of sequencing. I'm also going to give you some recommendations on tools. There's a lot of tools to do this, but I'll tell you the ones that we like the best. So talk the data probably has the most immediate obvious efficiency gain. Because if we don't need 10,000 dashboards and we can go with 50, that means we could probably take our data analysts and reduce that headcount and move those folks into other things like analytics engineers or data stewards or something else. But that does require the most work to get broad application across the enterprise. Like I said, start with the domain and expand. So that's where we're gonna put that one last. Let's start with the stuff that's a bit easier, that has incremental value to the foundations that you're doing, which is these first four. So metadata, lineage and documentation, let's start there. Then we can start with sensitive data classification as you can see how one feeds into two. Knowing what we must protect, tagging on arrival, and then opening it up so that there's wider audiences and consumption, both humans and agents. Data quality monitoring. And then number four, building faster, more efficient pipelines using AI acceleration. That's the sequence I could do. That's the sequence I would recommend. You can change it however you want. And again, if you want to get your executives excited and metadata lineage and documentation is not a way to start, you might start with a very narrow POC with Talk2Data. We're gonna get the finance data in there. It's gonna be very specifically. We're gonna be very controlled, but we're gonna show people what the future looks like. Great. Start with Talk2Data, and then you can backfill one, two, three, four, and then expand your Talk2Data to other domains. So let's say there's only one way to do this, but for incremental value, this is the order you'd wanna approach it. Oops, a little note there, which I just covered. Talk to data goes last because it benefits from everything else that we've just done. So some key takeaways. You can do AI today. You do not have to continue to wait for the big bang silver bullet use case. If you put AI in the hands of your data people, you will get significant value right now. So much so that you can you can look at getting the return on investment in a quarter. And honestly, if you have a tool like Snowflake or some of the other cloud native data platforms, those tools are there and ready to use. You're just going to be spending consumption credits to get them going. So you might be able for instance, like with cocoa, get that return far faster. Sequencing is super important. So again, that order that I presented, have a good think about it. And if you want help saying, you know, we've got some unique things with us, we might change the sequence here or come back to something later. We'd be happy to help. We do strategy engagements. We give you tactics. We give you a roadmap. We do that all the time. I love the saying measure twice, cut once. Well, every hour that you spend building foundations, ripple effects to two, three, four, five times the ROI as you get into other sorts of solutions that take advantage of the data foundations that you're building now. And Talk2Data as an enterprise level tool is built on all of these other things and it benefits from it, which means practically more people, more domains, more datasets, more questions can be leveraged there if you do the work before. So in terms of tools that we like, we work with a lot of tools. Snowflake and Sigma, we love these tools. The reason that we like Snowflake now, obviously, we we work with all kinds of tools on data, on analytics, on ETL. The reason we like Snowflake is it's the speed to value. The development cycles that some of these other tools require or the number of hands on to perfectly set up the way that it computes and consumes and everything. The expectation of speed to value is significantly shorter than it was five years ago, ten years ago. We like Snowflake because of the low effort, the low management and you can get going really quickly. Sigma we love because it's the closest thing out there to a genuine fully conceived BI 3.0 tool. It is mature, you can build data apps natively, it allows finance people to do spreadsheet style interfaces that is governed and controlled with the full power of a full analytics tool versus Excel. And on top of that, it's got talk to data functionalities. It sits natively on top of the warehouse so you can surface up all the great stuff that Snowflake can do as well. We love Informatica for a couple different reasons. But primarily, we love the fact that as you do some of these things like ingestion, you can start to catalog, classify metadata. It builds a workflow that compounds on top of each other using Clair, which is their native AI engine that automatically starts to build incrementally all of the governance stuff on top of it. We love that. Again, remember what I said earlier in this webinar. If you can add governance, so cataloging metadata classification as a part of your ingestion and data DevOps cycle rather than a separate effort that's manual or bespoke or discrete, it means that it becomes native, intuitive, and way more useful. And we love Informatica for that. So hopefully you found this useful. We can help. If there are questions that you have on how to get started, or which tools are going to be right for you, and not every tool is perfect for everybody, or a combination of tools, depending on what your use cases or your vertical or your data requirements. Or maybe you understand what your plan is, but you want help expertise in helping implement it, or building communities and getting user adoption and change management for the solutions you have built. We can help you do all of those things. Scan this QR code or you can visit the interworks.com website and enter in to contact us. You'll have communication from our people telling you, hey, the recording is up or following up afterwards. You can just simply respond to that. We've got lots of salespeople here in Australia that can assist, but we'd love to hear from you. We've been in business for thirty years and our number one priority is building long relationships, mutually beneficial relationships with all of our customers. We put a priority on people and how we hire and the partnerships as well as the clients that we work with. So we'd love to have that opportunity with you as well. That is our webinar for today. I'm going to give you guys the opportunity to ask questions if you've got them. We've got about five minutes. My voice is holding out, so I'm more than happy to jump in and field some of these. I'll stall for a little bit so people can start typing if you want. Otherwise, if you're good to go, you are more than welcome to continue on with your day. And for those folks that happen to be in Melbourne like me, it's a short week. Thursday, and then we have the grand final. Yay, football. No questions yet? Might give you guys a countdown of about twenty seconds. Oh, question. Oh, not question. Thank you, Andrew. Thank you, Rafa. So good to get the feedback. Happy that you guys are enjoying these and find these useful. If you guys do have other topics that you think we'd be interested in, let me know. I'll preview. There's two webinars that we've got planned for the rest of the year. One is how to survive and thrive in the age of AI disruption. So basic decision making and what we would think would be guidelines to help you. And the other one would be how AI is remaking the data landscape. And in particular, I'm thinking about people resourcing, how to plan projects, how to build processes and think how about platforms. So that is a transformative thing. I'm very excited about those two webinars. Got a question here from Srathi. I wanted to see what AI solution fails and where in the five use cases. Srathi, you might take another crack at that question. I don't know if I understand what you're asking. On a surface AI fails. And I answered this in a I put a version of an answer to this in a previous webinar. But basically it's if your solution is too narrow and doesn't actually track to a real problem, that's what I call the mouse of meh. Like we did AI, it didn't make a difference. Who cares? Most of the time, that's how AI fails. The other way it fails, which is far more spectacular and far more newsworthy is the wrecking robot. We went too big with no governance and it exploded. And now we've got a lot of people with a big job of cleaning up the brand damage. But if I were to simplify it, would say it's either a lack of vision of where you're going to use it, a lack of empowerment in terms of the people that would be using it, bad data, or a lack of training, I. E. Governance. But that's kind of true for all projects. Right? It's that idea of a purpose, people, process, platform. Hopefully, Srathi, that helps you with your question. Alrighty. I think we'll call it there. Thank you so much for joining today and please keep an eye out for future webinars. Reach out if you need anything we're happy to help.