Alrighty, let's get going. So welcome to another Intoworks webinar. Today we are going to be talking about AI at scale or another way of thinking about this is how do you operationalize AI to be successful? Solving use cases, productionalizing it, putting proper controls and governance around it, building the skills, the change management, the organizational things like that. So it's a huge topic. We've got a lot of stuff to cover. If whatever reason you have any questions, we have a chat. We have a q and a. Chuck those in there. We won't have a ton of time, but maybe we'll have five minutes at the end of this. I'll swing back around and see if there are any questions. If we don't get to them, we obviously can look in the chat after the webinar, and we're happy to contact you and reach out, with any answers or recommendations if you do have some questions that you want to chuck in there. So always happy to get your feedback. And all of these will be recorded. So if for whatever reason you want to rewatch this or share it with a friend or whatever, since you registered, you'll be alerted when it is posted on our website. Otherwise, all of our webinars that we do are recorded and posted. So you have access to those whenever you like. By way of introductions, a lot of familiar faces popping up into our attendees list, but some new ones as well. I'm Robert Curtis. It is nice to meet you. It is good to see you again if you are returning. I'm the managing director for Interworks. I'm based in Melbourne. I've been with Interworks for about twenty years. So I've got a lot of cobwebs and dust on me from being here for so long. And I look after Asia Pacific. So it is a pleasure to be with you today. This webinar concludes a special series of webinars we have done to celebrate Snowflake World Tour. That includes a white paper that I wrote on the importance of the semantic layer, particularly when you're thinking about it for AI consumption. That is available on interworks dot com. You can find that now. We have done a whole bunch of webinars as part of a series week after week after week. This is the final one of that series. And then we come back in a couple of weeks to talk about a new topic, which is very related, called Five Essential AI Use Cases Everyone Should Do. So that's talking about solutions that are proven, reliable, and doable. So it's not something that is an edge fringe use case. Five things that everybody can do and get significant value out of. So check that out. And again, you can register for those on the interworks dot com website. A little bit about who we are and what we do. My salespeople would be grumpy at me if I didn't spend a little time talking about us. Interworks is a data solution provider. So we do data strategy, data solutions, and data support. A little bit more on what that means. Think about the journey that your idea or your data life cycle might take. So again, let's start with strategy. That could be a full data strategy. It could be talking about frameworks and policies, directions, tactics, assessments, those types of things. We're happy to work with you to chart a course, build a roadmap and figure out the milestones that you need to do to make that successful. Then when we get down to the foundations, let's build the pillars upon which all of those strategy things can be implemented. That could be the actual platform itself, whether it's on prem or in the cloud. It could be your data platform, your lake house, etcetera. And again, we have multiple partnerships that can help you across multiple different tools, whatever your preference is, as well as taking the governance policies that we're building in strategy and building those in so they can be automated within those tools and platforms. Once we have the foundations built, then we look at adding value. We can solve problems, implement use cases with analytics, advanced analytics, augmented analytics, business science, AI, data science, machine learning, you name it. We help you with all of those things. We can help you by building, we can help by co building. So we while we're building, we're showing and teaching, or we can empower you guys to do it. And that brings us down to the next section, which is that sustain and grow. We can support you in several different ways. Like to think of topically as three specific areas. We can help build your community. So there is nothing more valuable than having a data literate culture of people that are using data to make decisions. We can help build that. We can help reinforce and sustain it. We can also help build the individuals in your community by giving them better skills, expanding their horizons, mentoring, workshops, all sorts of things. And then if you actually want us to help application support or support things inside of those applications like data pipelines, we can help you do all that too. Some of our customers use us for all of these things and all in one solution, white glove service. Others have different needs and they pick and select how we fit in like a puzzle piece. We're happy to do whatever. A little bit more about Interworks. We were founded in nineteen ninety six. So if you are doing the math, is thirty years this year, specifically in October of us being in operation and business. So that's quite a long time for a consultancy in that sort of SI side size versus say your global SI's. We love what we do and we wanna keep doing it. So that's why we've been doing it for so long. Same owner, same leadership team throughout. In that thirty years, we had built a tremendous amount of pedigree and that shows up in the big names of the customers we work with. Obviously, we have thousands of customers of all sizes, of all verticals. We write a lot of content to share our experiences, insights, and expertise on our blog, which is on into works dot com. And in addition to this Forbes small giant, which was where we were one of twenty five sort of companies that were quite small, we have three hundred people globally, but punch way above our weight in terms of the amazing things we're able to do. We've got twenty to thirty partner of the year trophies across analytics, ETL, data, platform, you name it. We, we end up collecting those, so much so that I have no idea how many we have. That is enough of the preamble, about us. I think I've also delayed long enough to get a bunch more people in the room before we actually jump into the content. Just a reminder, if you have any questions, chuck them into the chat. I will open up the Q and A too, so I can keep an eye on it, which is right there. Great. I will come back around to any questions you have at the end of the webinar. We have a lot to talk about today. So with no further ado, let's get started. I don't think I need to convince you why AI is so special and why it's so important and why compared to other trends, the dot coms, whatever, it is significantly different. It has a very different feel, not only in terms of what it can do, but the scale at which it's going to do it. But I'm going to give you some quotes regardless because quotes are fun. Have a look at those. The one thing that I like the most out of this is this quote here. We are not in a technology cycle like a data warehouse or self-service analytics. We are in a restructuring every industry on earth. Businesses that act now will define the new incumbents. The best way I can think about this, and I've had a lot of opportunities to talk to people about AI, people that are way smarter in AI than I am. I've had ability to talk to tools and vendors and partners and customers. So I've gotten a pretty good view of this. What AI actually means for our world, not just the business part of it, but everything, it's closer to the invention of the wheel or the steam engine or some foundational technology upon which everything else pivots and changes. It's not an overstatement to talk about AI in this way. It is not, like I said, like the first data warehouse, Teradata, or the first BI tool, which I believe was business objects. These are all in previous webinars, if you want to go see those fun factoids. It really is a paradigm shifting technology and will only get more and more powerful. And we can see that with those folks that are on the front line and just how quickly things have changed over the last two years. And it is accelerating. So AI is critical. It is critical for what you're going to do with your data. It is critical for how your business is going to leverage technology. The greatest business risk of the next decade isn't investing too heavily in AI, it's not investing enough. Can't overstate just how important it is. Now, with that being said, that doesn't mean that everyone that spends a dollar on AI immediately gets to see a return, a five to one, ten to whatever. There's a lot of reasons why AI investment isn't successful. And I would say there's probably two common paths. One, the mouse of meh. We didn't define our use case very well. We didn't lock in what the business value would be. What was the actual ROI? Maybe we didn't have buy in from the right people, so we were kind of skunk works, our AI solution. Maybe we had bad data, insufficient data. And when you have the little mouse of, it ends up being a distraction when you could have probably spent that time on other tools, other ideas, other solutions, or hopefully better AI. The other way that this manifests is the wrecking robot, which is we went too big, but we didn't quite have the controls in place. Maybe the controls were bad governance on our data, and we didn't have quality data, garbage in garbage out. Maybe we didn't train our models very well. Maybe other parts of our governance, the way human beings interact with it, or how it was collecting data from sort of that rag external sources. Maybe we didn't have security or privacy set up, but whatever it is, this big monstrous thing that we were attempting to do ended up blowing up. That doesn't happen very often, not compared to the Mouse of Meh. But when it does, it makes headlines. So most of us know AI failure in this lens. But if you think about probably in the baby steps that you guys have been doing inside of your organization or your colleagues have been doing, it's probably a lot of, we rolled out Copilot and three people are using it in marketing, and they're getting pretty good results. That that doesn't that is not organizational change. That is baby stepping when you probably could be doing something not probably. You could be doing something more impactful. And I've got my little arrows around the mouse of messaging. This is where most people spend their time and where they find that failure. So the reasons for AI failure. Eighty to ninety five percent of corporate AI pilots fail to deliver measurable ROI. That's from a Harvard Business Review and MIT study. And the failure is usually aligned to misalignment, poor planning, and the lack of initiative, meaning most AI fails at strategy phase. Anyone that has attempted this knows that there's a very common pattern, at least in the early stages of you guys attempting AI. I mean you guys meaning clients or organizations, whether they're private, public, whatever. And that is some leader, probably at the c level, says AI is the future. We need to do AI. Okay. What do you want us to do with it? I don't know. Just do it. And so AI as a buzzword becomes a destination versus part of a way we can solve a problem, a problem that is going to produce ROI. The other thing that people do is they think of the solution as a product. So I need, this is how things worked for something related. I need data governance, the tool, Calibra, Alisha, whatever is the solution. We buy the software. Now we have data governance. That's not true. You have to have the implementation of the solution in the tool. You have to do stuff, not just have the tool. Otherwise you end up with shelfware. So what happens with AI is people like we have ChatGPT, we have Claude, we have Cowork, we have Genie, we have Gemini. Well, that doesn't mean anything unless you're using it. So you've got to have the traditional piece that go along with the project, you know, project purpose, people, priority, etcetera, etcetera. All of those things are absolutely necessary for AI. And if you don't do it, then that's where you fail before you even actually get implementing. There's another problem here. And that's that not that people are trying and failing, but they're too slow to take advantage of it. So that's sort of reactive AI adoption. Every quarter of inaction means there's an opportunity cost or a value loss. So if you think about traditional data solutions, If I implement a data solution, let's say there's an AI automation component to it, and I'm able to save us a hundred hours of time every quarter. Well, every quarter that I get to take advantage of that is another hundred hours and probably compounding on top of that, because that hundred hours isn't just saving Susie's time doing an automated task. It allows Susie to probably focus on other things, innovative decision making, expanding value, those types of things. But if I wait four quarters before I implement the solution, that's four hundred hours that I have lost. Now the tricky bit with AI is, again, it is not a linear compounding, or rather, aggregate savings. It's often exponential in terms of how things grow. So the later you wait, in particular, to the rest of the world around you doing it, not only are you losing your exponential value in terms of the efficiency, automation, and innovation, everyone around you that is doing it is already way ahead. So there has to be a real careful thought process of one, we we need to start thinking about this today. But two, what are we gonna do with it so we can make sure we are successful on the first, attempt? Because the longer we delay actually adding value, the longer this compounding debt starts to become problematic. And you could read the other ones, your competitive displacement talent drift. Those are all factors to consider. So the gap between enthusiasm, yay, we're going to do AI versus, hey, we are getting value from AI. So what organizations say, and again, this ties right back to what I was saying. I'm remembering that my silly little headset will shut off unless I have sound coming back. So I'm just going to put on some background noise. Otherwise, you will stop hearing me. Go. Perfect. And a lot of these come from, C level people, people with budgetary authority, but not with the operational knowledge of exactly what the value here is, or how do we get success? So AI is our number one strategic priority. What that actually looks like is one off pilots with no path to production, not a real defined ROI. It is an experiment in a silo. Does not move the needle, does not change anything. Or we're investing heavily in AI capabilities. Shadow AI spreading without governance. We have a data driven culture, no clear ownership, no COE, no framework. We have culture, but we do we have leadership, particularly when it comes to AI. We're moving fast on this. Leadership is waiting for somebody to take the reins, some technical person to appear out of the organization, maybe out of the COE and saying, I'm going to give an initiative. And what leadership C level folks often do is they go to conferences, they go to cultures, they talk to vendors, and that's where they get their ideas from. Versus really smart technical people, like the people on this call, coming up with a path forward tactics and strategy and value that relate to your organization. And a lot of times it's because they have not empowered. So let's ask a question. If we are reactive or we're not ready yet, or we're doing our pilots in a very small little silo for AI, will my community of users wait until leadership is officially ready to launch our AI platform tools and solutions? Anyone that's in technology or in IT will know the answer to this, and it's kind of a follow-up question. When have they ever waited to use really cool tools? Anyone that was around for the boom of self-service know that Tableau and Power BI, they pop up all over the place. They're like termites. Don't mean that in terms of the value of the tool or the usefulness of it or the people that work there. I mean that they just show up whether you intend for them not to. You might be a Tableau organization. You guys have Power BI. You might be a Microsoft. People are using shadow tools. And the same thing is going to be true with AI. We are at a point when it comes to AI. It's not whether we should use it, why we should use it, when will AI be ready. It's only now how we are going to use it that matters. There are basically three options. Some folks are sprinting forward and trying to learn and grab value as quickly as possible for better or for worse. Sometimes they do it smartly. Sometimes they do it recklessly. Some and that's the wrecking robot can be the bad side of that. Some are taking very small steps. And again, measured learning or ineffective siloed POCs. That's where the massive muck can be. There can also be if you are taking small steps in preparation for a big sprint, that can be a very smart approach. And then some are simply waiting. All of these are strategies, whether you believe it or not. But in your organization, everyone is using AI. Everyone. Whether you like it or approve it or not. And there are stats to back this up. This is this idea of shadow AI. It's like shadow IT, but the blast radius, the area of effect is much, much, much larger. So governance is behind the tools. So people are using tools without governance. The value is personal. If I use AI to help me write my emails, my documents, my business use cases, my PowerPoints, It is a actual, genuine, real accelerator for me. And if you tell me not to use it, you're telling me to be less efficient and to use more of my personal time doing menial stuff. Good luck getting employees to buy into that. The failures are invisible. So when personal use of shadow AI breaks, it's often you don't see it until something catastrophic happens. A lot of times it's the little leaks that break the dam before, oh my gosh, we actually have this problem all over the place. It is endemic. It's like the termite analogy again. If you see a termite, you probably have them all through your walls. You don't see them until they're a problem. Here are some stats on shadow usage if you don't believe me. Ninety percent of enterprise AI activity operates outside of the visibility and control of the centralized teams. That's people going to Gemini, the website, and just typing stuff in or Perplexity or ChatGPT, etcetera, etcetera. All of these tools are free and available. And they will take your documents if you upload them and ask, help me make this better. Give me some analysis on this transcript from my meeting I just had. They will all help you do that. The average enterprise use the average enterprise uses over fourteen distinct AI tools, but IT teams are only typically aware of about four of them. Seventy eight percent of employees report bringing their own AI tools to the workplace to improve efficiency. Or if you've given me a restrictive consumption credit limitation, I'll go to the free tools when I run out to save my credits for my for the stuff that's really more complex and harder. One third of AI tool interactions happen through approved enterprise accounts, which means sixty six percent approximately are unapproved. The risks of ShadowAI, data exfiltration, so sensitive data pasted into consumer AI tools that are ungoverned. That could be inconvenient. It could also be regulatory or criminal depending on your organization and the data. Capability fragmentation. If we don't know the use cases because people are solving them outside of our solution framework, we don't know what problems we have and how to solve them or how to govern them. We lose visibility and control. Obviously, the compliance regulatory liability is pretty evident and obvious. And then if you are trying to plan for consumption costs, and there are shadow AI tools everywhere, it's just the same thing that you had in the old Power BI days when they had their virtual servers behind the scenes. A lot of people could start spinning up costs with almost no governance. So people can be through their credit card, whatever, and people can just be off to the races and doing this stuff. Or if you're trying to plan and negotiate your consumption costs with an AI provider and you tell everyone to start using this, you will have no idea just how much AI they're actually using. And then you're like, oh my gosh, we've underbought. We could have negotiated for a much better price if we knew what everyone was using, but they were using shadow tools. Several problems there. So if we summarize the AI challenges that you forgot, and these are the challenges that come without having structure, framework, and strategy place. It is the cost of waiting to react, the lack of strategy, which means leadership, use cases, direction, value today, the challenges of governance, and again, shadow AI, which we've just covered. We need to be realistic about how we solve these problems if we're going to move forward. If we're going to address them, to build the counterpoint to them. So with those things firmly aligned, now let's think about how we're going to be successful. How do we do AI at scale that is predictable, that is governed, that is productionalizable? The first thing we have to do is focus on building the AI COE. You need a center of excellence that have a lot of smart people representing a lot of different parts of the business. That's the data people, governance, technology, etcetera, etcetera. All of these people have to weigh in. And people themselves have to be represented because AI is an accelerator for people. Yes. It does all the automation and stuff, but ultimately, you give AI in the form of agents to people, they're gonna drive significantly more value. It's what self-service dreamed of being, but the realm, the world of dashboards was the limitation. AI unlocks that. It is finding the right tools and selecting them, negotiating them. It is a current state assessment and honest evaluation of where you are and what you need to do and where to go. It is putting governance around your AI agents. In the world of BI two point zero, which is your dashboarding world, people could very easily end up with five thousand, ten thousand, twenty thousand dashboards because people just build dashboards and that's what the tool does. In the BI three point zero world, I've heard of people saying, I think we have somewhere around seventeen hundred agents and I don't know where they're coming from, what they're doing. And that's a real danger because agents can actually change stuff. They can do stuff. They're not just a dashboard reporting on numbers. They build data, they build artifacts, they do stuff that you've got to put governance around that. And then the last one, on top of everything here, you need to build an AI culture that is synergistic with your user community. People need to understand what AI can do, what you don't want it to do, what assets it can build, what data sources are not available. Understanding all of these things together is super important. So let's start with your AI center of excellence. And just like a data center of excellence, you need to have delineation stratification hierarchies. And it starts with your executive leadership. When it comes to building organizational initiatives that are governed and controlled and productive and not a shadow tool, I e, Roundup, mandate and budget and governance must come from top down. It must be, we are doing it this way. This is how much money you have, and you must follow these rules. Executive leadership sets that. Super important. Mapping AI to ROI is a c level conversation. It is a CFO talking to your AI leadership in terms of how we're going to invest and what kind of returns we're gonna get. Otherwise, if you because AI is so consumption based, it can get out of control very, very quickly. Underneath your executive leadership, this is where you have your COE committee. So the operational core of the people that come in and think about building the strategy, executing the strategy, the tactics that make up the milestones within those strategy conversations, best practices, standards, the platforms and tooling, the governance policy, how we prioritize and solve problems in our use case pipeline. All of that stuff comes from here. They get approval, they get mandate, they get direction from the executive leadership, but the work engine of where we're going comes from tier two, the COE committee. Where this works best, and you have to be a certain level of organization. You have to have some certain level of sophistication. Is the COE from a strategy and tactical standpoint is the hub while your sub councils are your spokes. And so sub councils, those could be our our governance council. These are data stewards, domain owners, data specialists, or governance specialists, those sorts of folks that specifically think the PhD level thoughts so they can feed that back to the COE, the thing can be integrated into the strategy. So it could be the ethics and legal review, the analytics folks, the power users, the AI specific stuff. All of these things need to go into subcommittee or sub councils or breakout sessions so that you guys can have the deep meaningful conversations. Summarize those, synergize those, and then come back at a summary level to your COE. And then tier four, your user community. You could divide this however you want. I think three levels are probably sufficient. So your builders, your leaders, and your practitioners. However you decide to build your cohorts, you need to understand what is their requirement. So do they have a certain level of technical ability that they must have? And how do we get that to them? Two, what is their obligation? What do they need to be doing for us? And three, how do we measure value from them? So builders, leaders, and practitioners, you can probably start to speculate, this is what we would want from our builders. Need to have AI specialization, they need to understand data apps, need to understand vibe coding and And then we get down to sort of our practitioners. They are looking at process workflows, optimization opportunities. There's a BAU function there, etcetera, etcetera, etcetera. You must build these cohorts so that you can build your enablement to get these people properly specialized to doing the parts of the machinery, the cogs and the process appropriately so they all synergistically interrelate to then create an effective engine. So as an example, I'll show all of these tiers. You might you might not land exactly like this, but this is just an example with a RACI chart. And there's other ways you can do this. You don't have to use a RACI. There's other ways sort of assign ownership. But you might say, here's our executive sponsor. And in our organization, it's the CEO, or maybe it's the CFO or whatever. The AI program champion, this is the person that from a C level perspective is the one that's the most intimately involved and they bring back the findings to their C level, to the SLT. The CFO is the budgetary authority, obviously specifically with AI and anything where agents can do stuff, you've got to think about legal and risk. As you go down, you can start to see how we break up the COE. So it's data and AI architects, governance change and enablement, which is the people part of this platform engineering. So that's infrastructure tooling platforms with a COE dedicated leader. I would recommend that a COE person be separate from your C level. You'd want them to have more ability to focus on the COE and the expanded sub councils and things versus being a part time role as a part of a more executive level remit. There's a lot to do here. So give yourself a dedicated COE. And then when you get down to tier threes, and if you are just starting out and you don't have the largest organization in the world, you could probably fold this into your COE. But as you get more sophisticated and you're doing more things and you have more budget, you definitely want to sort of break these guys out. So as we mentioned before, the Data Governance Council, your AI ethics and risk platform and tools, enablement. You might throw advanced analytics in there, etcetera, etcetera. And then we get down to the bottom. We talked about these folks on the previous slide, your builders, your leaders, your practitioners. And maybe even throw in there just for the folks that are users. But as you are building your plan for AI, you need to have these conversations and you need to map these roles to these sorts of responsibilities. However you want to do it, RACI or whatever. When it comes to selecting the tools, it's not really a purely technical decision. And I'd also say the other thing that's worth noting is that this space is developing extremely fast. If you do any of these horse races where you'll see, AI experts, you know, mapping which tool is leading in a particular AI specialization, they change quarter to quarter. So if you've got a really clear idea of what you wanna do, you might go in and say, hey, we're gonna lock in a long term agreement with this organization. And there's different types of AI tools. Let me say that too. There's tools that are highly focused. There's tools that are broadly focused. And then there's the AI itself. So in reverse order, if you think about something like Claude, that's like the ocean. It is very powerful. There's a lot to do, but it is very broad and very wide. And you're gonna have to spend a lot of time focusing it. Is it the best tool to just unleash onto your organization? It is very big and very broad and very powerful. And it's it's where you can also get yourself into trouble. Whereas if you go a layer deeper, this is things like co work in Snowflake. More focused, more structure, more definition. So think about this instead of an ocean, we have a powerful river. It's kind of going in the same direction. There's some risks there. There's some stuff we have to do to layer on top of it, probably to make it work, to make it more usable, to make it more governed. And then you have tools like your BI three point zero tools and something like say Sigma, Think of this as like this is all the plumbing and we've got taps and now we can turn the tap on and we get a really dedicated little feed because the tool has so much structure on it. Now that doesn't mean that the ocean or the river isn't valuable. It means that you need to think about how you solve problems with the types of tools that are best for it. Some other things to sort of call out from here, I would also lean into the financial part of this. There's there's generally two ways you can do this. You can kinda do as a pay as you go, which means you're paying higher per credit cost, but it gives you the flexibility without making a commitment. And so you might you might sort of engineer this for your startup. And then as you get more comfortable and you have a better sense of what your usage might be, you might then lock in something. If you are comfortable doing a multiyear and with a tool say like Snowflake, I wouldn't object to it because Snowflake does a bunch of other stuff. It's your warehouse, lake. There's governance in there, there's data science, etcetera. AI is a component. You can buy your consumption. And a lot of times you can actually say, hey, we're gonna be in a ramp up phase. So work with me on what my year one credits look like versus my years two, three, and four. And a lot of times they'll help you. They might let those credits go across years. They might give you a bit of a ramp up in terms of how you're paying so that you're getting a great discount across all years, but you're not signing up for a whole bunch of credits and you're one that you might not use. Have those conversations with your vendors or ask folks like us to help you. And we negotiate these things all the time that are gonna be in your best interest. So lots of things to think about when you're selecting the tool. From a shadow AI perspective, I think you have to audit before you buy. How are we using? Have a an an amnesty window. So, hey. We we we want an honest understanding of what you guys are doing with AI. And there's gonna be no penalty for things that you tell us that you've done. Now, obviously, need to stop behaviors if they're going to put it at risk, but we need an honest view, an accurate view of what we're doing with AI so that we can pick the right tools, we can negotiate the right prices, and we can understand people process and change management. Fastlane for low risk, meaning we're gonna give people agility versus slow them down with governance. There's value in either or. You've gotta figure out how that maps to the specific use case. You don't want people on the cutting edge of building your products, whether they're data products, analytics products, or actual products, slowing down to wait for IT governance. So you've gotta find the compromise. I think the other thing that's really useful about ShadowAI is it's a signal of unmet unmet need. It's not just something that you must it's not a faucet that you can turn off. People are doing this because we're not providing them with all the tools they need, or we're not making them aware of the tools that they do have that could solve their problem. So it's a challenge for us as the providers of AI to make it more useful. So common traps of tool selection. We are solving today's use case. So we buy a tool that solves today's use cases only. Think about this in a two year roadmap. Again, tools are changing pretty quickly. So if you think a year or two year, that's probably safe enough. And most of these tools, particularly the ones I would recommend look at do have extensibility and interoperability that you have the ability to move them not so painfully. If the tool does require some proprietary lock in, I would take a long, consideration of this worthwhile. Two, letting a single department drive the decision. Marketing is really passionate about this AI tool. We're just gonna let them buy it, then everybody else can come take a look at it. You often end up in a very narrow, great for them, not great for everyone else. You should look at it from a holistic view, which is why tools and platform and governance must come from the top down. Let them be the pump that primes everyone to get excited about it, but everybody needs to have a voice. Underestimating integration and maintenance costs. I'll also throw out there the time to migrate. If you are bringing data from one system to a new one so that you can leverage all the great AI stuff, it takes time to translate business logic. Do not sell yourself short because what you don't wanna do is be paying for two different platforms without planning for it concurrently. We've gotta keep this guy running and this is the new world and how do we do the change management and shutdowns and things like that? So takes a real fine touch to understand exactly how you're going to integrate, migrate, and then do all the maintenance. Treating vendor demos as proof of fit. They will show you the best version. It is a smart thing to do for them because they want you guys to buy. You need to get your data in there. You need to be having your users play around with it. Something like a thirty to sixty day window to go and play with stuff is a fairly reasonable ask, particularly if you're gonna be investing in at at scale for your vendor. Ignoring the talent and people skills required to operate the choice. Do we have the ability to do what we're buying? And if we're going to go buy this tool, is there a talent in market to go grab it? If I'm going and grabbing, oh, I don't know. Let's say I'll I'll pick on Google. Google's the smallest sort of cloud provider here. If I go all in on BigQuery, how easy is it for me to go build a team of twenty data engineers or architects that are certified at BigQuery? It's not as easy as it would be if you went, say, Snowflake or some other like Microsoft. There's just less tools. So you've gotta be well, it's Palantir or whatever. You've gotta figure out how do we actually operate this, maintain it, and get value out of it if we don't have the skills already? Current state assessment. Let's figure out where we are in terms of, in particular, our data. Your semantic layer plus context is the the fuel that pilots all of these things. But AI in particular, it is mission critical. You used to be able to distribute your semantic layer, your business logic across multiple steps of the data life cycle to get a BI dashboard working because it didn't really care. It was just there to sort of take product step step by step and then just regurgitate what it had at the end of this automated life cycle. AI is a consumer. It must see data in a nice curated spot with context so that it can give you the best answer. So I would start by assessing your semantic layer and context layers first. That'll help you prioritize where you start, which department, for instance, or which use cases you're gonna start with. If finances data is all over the place and really ugly, can't really start there. But if let's say supply chain is really tight and they've got a single data source that we know has good quality data in there, we just gotta get it into a cloud platform. Great, let's start there. We can migrate and start adding value. So your semantic layer. Do your data assets have meaning? Business glossaries, metadata, metric definitions, data dictionaries. This is the stuff that AI needs to know so that it knows why this stuff is important and how to use it. Is your data model documented and shared? Can AI distinguish from basic definitions in your, like for instance, customer versus user or any set of transactions? What is sales? And is it, let's say it's, is it revenue as according to finance or is it revenue as according to the sales team? Because they mean different things. One is how do these people get paid on their variable? Because they're salespeople. And the other one is what do we report to the street? And what do we report to our shareholders? They're very different numbers, but they're both very practical and very useful. Have you consistently calculated your KPIs? This is not a AI decision. This is not an automation decision. It's not a machine decision. These are human beings sitting down and figuring out, okay, what does this mean to us? And how are we going to lock in this definition? You've got to inventory this stuff to make sure your AI can help you properly. If you don't, then AI is going to be all over the place. That's not really AI's fault. Context. Is an easy way to think about this is think about this as your semi structured and unstructured data that AI can look at because it can crawl through PDFs and all that other stuff. And then add stuff that maybe not, that isn't in your data that helps it build better associations, relationships, and value. So metadata, lineage freshness, information management policies, all of that stuff is super important to to to combine a semantic layer, which is sitting on top of bronze, silver, gold, platinum, data products next to your context layer. Another way to think about this is this little diagram here. The semantic layers on the left hand side, your context layers on the other. The other thing that's useful about context is this sort of rag, the ability to go outside of our business and look at things. So when I'm using AI to help me build ideas, solutions, or proposals for customers, it's useful to see what is out there. Doesn't mean I take everything verbatim and copy and paste, but I do love to understand what the rest of the market might be doing and then apply my personal expertise as you would for your things in terms of how I make it useful. Those things come together in terms of the AI model and agent. And if it's not ready, maybe we don't have the semantic layer, there's no lineage, there's no KPI governance, we don't have a metric layer. AI will hallucinate at scale and it will tell you that these are right answers because that's what AI does. It gets you answers, but they won't be right. If you're partially ready, well, there's some definitions or insignificant coverage, which means the answers might look close and they might be right some of the times, which in reality means it's not useful any of the time. If you have users guessing which number is right, they're all wrong. You can't trust any of them. And what it looks correct is when you have a fully governed curated semantic layer built off of sound business definitions with lineage, metadata, classification, trusted, documented, contextualized data supported with a robust information management policy. Agentic AI governance. And the first one there is the most important. Agents act. They don't just give you answers. They can send emails, they can modify records, they can do API calls, they can trigger workflows all while they can do it without a human being telling them to stop or to start or to confirm the step. Agents are really powerful and we want them to do those things, but we have to do all the governance around it to make sure they're doing what we're expecting. Potentially very dangerous if they're not. So that means when we're building in controls or accountability, we have to build them as a part of the process. And oversight has to span the full stack, and that's from data source systems down through the agent itself. Risk is continuous. It's not a point in time. We have to be vigilant. It has to be something that we do as a behavior, not a one off activity. And the last one there is while AI and the technology has moved faster than regulation can keep up with, it is coming. And there's all sorts of acts in Australia, in Asia Pacific, in other trade unions that we're probably going to have to go look at a mirror, whether it's in the EU or the Americas or whatever, that are going to change the way that we use AI. So having that governance of that perspective early is good. So some questions you might ask yourself in terms of building up governance, who owns this agent? It needs to have a specific individual, not a team, not a business unit. Roger owns this agent. And if something goes wrong, Roger is the person I pick up to say disable it, fix it, give me a damage assessment. Ownership super, super important. Just like data stewardship, somebody needs to own it. And if it's diffuse, no one owns it. Who touches this data? It's like an RBAC for agents. And a lot of times these agents, you get the most value if they can multiple systems and multiple data sets. You've got to figure this out as part of the planning stage versus an audit of something that went wrong. In terms of the use case, the solution, again, having multiple people sort of look and pressure test and validate that everything that this agent is allowed to do is within expectation. And let's troubleshoot. What's the worst case scenario if something goes wrong? And then let's mitigate that. Who is accountable when it acts outside its bounds? And who has the authority to say turn it off? All of these things need to be mapped and think about. Here's another way that you might think about it. So the compliance layer, compliance layer, data layer, platform layer, agent layer. Who's the owner? And you'd wanna put specific names to these for who represents these roles in your organization. What exactly do they own? And what are the controls that they have? You'd want to build this register. Next, building an AI culture. I'm a big fan of ADKAR from a change management perspective. And ADKAR stands for awareness, desire, knowledge, ability, and reinforcement. Every time that you are going to implement something, it's anything that is an initiative or anything that involves people or a change of process, something that you need to evangelize, you need to be able to answer these five questions. So awareness. Communicate the specific business problem AI is solving for each function. Not AI is the future, AI is valuable, etcetera. But this is going to change your month end reporting cycle for you collecting these data assets and building this integration from four days to forty five minutes. This is the specific process that we're automating. Bang, there you go, awareness. All the way down through desire, knowledge, ability, reinforcement. Again, what most people miss is it's not any sort of change management is not a clash of symbols were done. It is a repeatable, reinforced messaging, marketing campaign, ad campaign to users to use these processes and tools. You don't put up a billboard and call the marketing campaign done. You are constantly trying to find different avenues to communicate your message, and that's exactly the same thing that change management is. You need to post it on your user portals, your intranet. You need to have conversations about it in your community of practice. You need to tie the reinforcement into managerial conversations whenever there are employee reviews. All of these things work together to holistically move people in the right direction. One size training fails. Different roles, we talked about different cohorts. If you say we're gonna treat everybody the same and write the same owner, the same authorship or rather say again. If we write the same type of training plan as if everybody was the same, it will fail. For instance, let's say I am a data engineer and I need to be trained on a change management piece for this new AI tool. Would you want me to do the same training as you'd want your CFO? No. Very different. One person makes way more impactful decisions, but they're working with trends and aggregate data. Me as an AI engineer, I am way down in the detail and I have the ability to mess things up more granularly. So different messaging, different training, different outcomes, different goals. The other thing that's important is there is a onboarding training and then the reinforcement. And then when things change, we have to do an update. So again, not a one size fits all training methodology. So when you're thinking about how you build this training model to fit the folks that you have, think about how you define your cohorts, project roles, people roles, process roles. Who's doing what in this particular limited set of work? We're building stuff and there's a front and an end. People roles, management, leaders, HR, team leads, department heads, etcetera. Process, they do this thing to help this thing move. All of those things might create or pull people into different cohorts or define the things that they need to understand. So things that you can do in the next thirty days. Inventory and assign ownership. We saw that AI schematic of who owns this process at the different layer, risk of the data, etcetera. Run a use case pressure test. Take your top five AI data priorities in the next eighteen months and pressure test them against your platform strategy. Do we have the right tools? Do we have the right people? Do we have the right skills? Do we have the right measurement in terms of how it's gonna be successful? And then pick a cohort and start a seventytwenty build. And what seventytwenty means is it's an educational model, a training model. Ten percent is classroom or more formalized instruction. Twenty percent is sort of coaching and mentorship. And seventy percent is let's get their hands on the tools and let them learn. Give them a sandbox, give them a safe environment, and have them just start experimenting and building. All of those things super easy to do. And then that will get you going. It'll start that AI to scale. That AI COE will start to take shape. You'll start to have a bit of a direction. If you want help, we're here for you. There's a lot of things that we can do versus say strategy or helping to find use cases or helping you map ROI or helping you discover what use cases you could like, hey, let's go talk about your your top ten biggest problems and process flow them and say, bang, right there at that step, that's where AI can take two days a week out of your workflow. We can help. I think the easiest thing that we can do is a DART assessment. What the DART is, is it's a data analytics review and tactics. We have a broader one. If you want to go into much more deeper detail, let me know. The free version is pretty succinct, so it's not a massive time commitment for you. But as part of it, we do an assessment of five different technical domains, which is culture, analytics, data, governance, and infrastructure. And from there, we're able to extrapolate your AI readiness. It's a great place to start. Let's start with the assessment. Let's measure first. And then as we do stuff, we can come back in a year and have a deeper look and say, how did we improve? As I mentioned, there are five different maturity frameworks there. If you are interested in the Dart or just having a conversation about AI and how AI can help you, contact us. We have got amazing salespeople. We've got great solutions people. I like to jump on these calls and and talk directly to folks too. Any way that we can help, whether it's just giving you some ideas, maybe giving you a bit of an audit assessment, how have you how have you guys done, but some things that you guys could think about sort of patch up some risks or some areas you haven't thought about, areas where you could get more efficient? Happy to help. Scan that. That'll take you to our contact us page. We'll also be emailing you after this session probably in say forty eight hours to say, hey, the recording is live. You can go and access it on the website. Would love to chat more about how we can assist. So that is the presentation. I don't see any questions in the webinar chat. If you have any, chuck them in there. We've got three or four minutes, so I'm happy to grab a question or two. Hopefully, you you found this useful. If you did, I'd love to see any perspectives or thoughts on the presentation in the chat as well. Thank you, Rafael. That was very nice of you. Let me know if you do have any questions. Otherwise you are free to go and we hope to see you, I think on what was it? The twenty second when we talk about five essential AI use cases that everybody should be doing. Alrighty. No questions. Thank you so much. I'll see you guys next time.