Video: The CoE Vision: Empowering Self-Service Data for Modern Analytics and AI | Duration: 2556s | Summary: The CoE Vision: Empowering Self-Service Data for Modern Analytics and AI | Chapters: Welcome and Introduction (7.8399997s), Introducing Self-Service Data (148.04s), Changing AI Landscape (228.65001s), Establishing Vision and Strategy (372.32s), User Persona Definition (642.325s), Six S Model (999.99s), Safe Innovation Environment (1710.705s), AI in Learning Development (1821.93s), COE Data Governance (1925.445s), Data Access Governance (2057.065s), COE Effectiveness KPIs (2178.66s), Semantic AI Applications (2287.435s), Key COE Roles (2377.02s), Closing and Preview (2478.135s)
Transcript for "The CoE Vision: Empowering Self-Service Data for Modern Analytics and AI": Hello, everyone, and welcome. We're so happy to have you with us for the first session of our series, building the next gen COE, empowering self-service in the age of AI agents. If you're involved in building or scaling a center of excellence in your organization, this session is definitely for you. As we wait for everyone to settle in, please feel free to say hi in the chat and perhaps share what role you play in driving your organization's COE efforts. Throughout the series, we'll explore how next gen COEs are enabling self-service while maintaining strong governance. We'll dive into strategies for scaling AI initiatives, aligning AI agents with business goals, ensuring seamless integration across the organization, and so much more. Before we could call off a couple of quick reminders, for any of you who have comments in chat, please drop them in the chat, section, and we'll keep an eye on them. For questions, please use a q and a panel. Jonathan will try to address some along the way. But if we don't get to them, we'll reserve some time at the end to answer as many as we can. And lastly, we get this question a lot. The session is being recorded, and we'll send the recording and slides to your inbox after the webinar. So now that we've covered a bit about the series, let's dive into today's session, the COE vision, scaling self-service and AI agents in the enterprise. We're so excited to have Jonathan Tudor here with us. Jonathan, if you could go to the next slide. John is the director of business architecture at AIQ and formerly the senior director of enterprise data technology at GE Aerospace. What's even more exciting is that John is cur currently writing a book on this very topic, which I've had the privilege of reading, and it's fantastic. Some key insights from his book will be shared today, so you're definitely in for a treat. Today, John is here to show us how we can build a COE in the age of AI agents driving innovation and empowering self-service. He'll also walk us through the six s framework, which will be super beneficial for turning your COE vision into actionable steps. So with that, John, I will pass it over to you. Fantastic. Thanks, Renata. Really appreciate the introduction. And, you know, good morning, good evening, and good afternoon depending where you are in the world. We appreciate you, joining the session here today. So just to kinda repeat what Renata said here a little bit. Welcome to our first, webinar here in the series. In this series, we'll be focusing on how COEs can effectively drive, data democratization, AI democratization, utilizing self-service organizations. So I'm a little bit tied to how we work with people, but we'll also think about what how has this changed in the world of AI agents. Just to give a bit more introduction to myself, my name is John Tudor. I'm a director of business architecture here at Dataiku. I lead our program called self-service data enablement, which looks to help our customers really launch, scale, and operationalize self-service programs. And as previously mentioned, I used to work at GE Aerospace. It's been about thirteen years there. I'm over 10 at working in data analytics. And much specific to our conversation today, but a very large, self-service program there with Dataiku and other technologies as well as the data governance program, and then ultimately led the enterprise data strategy before moving off in the organization. So what you'll see in this six part series is a combination of both sharing those past experiences from my previous career, but also looking at the best practices from across over 80 plus customers who now engage with our, self-service program here at Dataiku. So with that, let's go ahead and dive in. So the first thing we kinda want to talk about here today is really the changing landscape. For most of you, if you've been on LinkedIn, really, in any business lately, we're never talking about AI almost every day. Right? So a couple of these quotes you'll see here on screen come from various recent articles related to, you know, economic impact and related to really some of our major partners like such as Deloitte and E and Y. And some of these quotes include, you know, 94% of CEOs are now saying that AI can actually provide more insight than than an actual board member in the organization. Additionally, we're starting to see this kind of delineation between the way that we're looking at using AI agents. Some cases, we're just looking to kinda replicate existing processes. But other times, we're looking at how do we take that further and really enhance how the business process works. Right, piece of different wording of it. How do we ensure the process is correct before we identify it? Additionally, we're kinda seeing a lot of the market move directly into working with AI and agents. So it's anticipated that within this year or 2025, that 25% of organizations will deploy AI agents in their enterprise, and that's expected to grow to 50% by 2027. And then finally, as we go out and survey and look at senior leaders in business, what we're seeing is that it's not necessarily so much a technology gap, but it's actually a data infrastructure and people and cultural change gap that many organizations are solving. So what we can see is that there's many things happening here in the AI space. It's a brand new world. We're trying to figure out what are the best ways to work with that. And importantly, I think throughout is that there's a focus on people and how do people work with AI and agents effectively. So kinda looking at a longer term trend here, one of my favorite things to look at is really the change in decision making jobs really over the last few decades. If you look at, you know, 1960 to about 2018, we've seen decision making jobs go from about 6% of the type of work we do to about 34%. And, really, since that data has continued to accelerate, it go up to the right. And if you can see here on the right side of the screen, what we'll see is about 60% of advanced economies expect to have their jobs either impacted or have their jobs replaced by AI, and this is based off an international monetary fund report. So really when we talk about the space, about enablement through self-service, empowering people to use AI agents at scale, What we're really talking about is how we're changing the future of work. We're talking about how we'll work every day. We're talking about how our employees will work every day. So how do we think about making that successful, conducive, so we're ready for this future that's powered by AI? So with that in mind, the first thing we'll talk about here today is how do you really establish a vision and strategy for a program like this? Because when you're thinking about democratizing capability or asking others in the business to change how they work, you have to ensure that what you do directly aligns with them and the outcomes they'd like to drive. How do you make things resonate with them? One of my favorite quotes that kinda focus on this is actually from Richard Branson, which you'll see in the top right, where he says, you know, to launch a business means successfully solving problems. And while solving problems means listening. So, ultimately, in order to connect your strategy or what you're doing to help the people within your organization and your employees, you have to ensure that the strategy and vision that you build directly resonates based on what you've heard or learned from them, whether it's through design thinking, interviews, etcetera. How do you pick up that information? Now underlying all this is the idea that business itself is in the nature of solving problems. And that's not just true when looking at the external products we sell, like someone like Richard Branson might sell, but it's also looking at the internal type of work we do in organization. Right? When we build something in organization, we're looking to gain efficiency, we're looking to gain more impact, and all those things are focused on, again, on how do we solve problems. And generally speaking, most business problems can be either solved or significantly addressed through data and AI. In today's world, you can use data to make those decisions to accelerate them and make them happen faster. So what we have to think about is what happens when you empower the employees in organization to solve their own problems and do so at scale with data and AI? And what this really leads to is a extreme increase in how you do your innovation, your efficiency, and really your value creation because you're now asking individuals to solve the problem closest to where the work is happening. And through that process, they're actually automating or changing how they do their work on a day to day basis. So when we talk about self-service and AI, really, what you're doing is you're solving the problem for how you empower the human capital of your organization to really solve as many business problems as quickly and as effectively as they can. Now as we think about this and how we take action for it, the focus is really on listening. So the first step to solving any business problem is to do this listening. And and simply to think about this is through employee interviews. So how do you learn about how the organization works with data and how they may wanna work with data in the future? Ultimately, based on these interviews, you can think about, you know, how do you create the vision that resonates around how you use data and AI? How do you build what's called a branded mental model? We'll cover that here a little bit later in the session. But how do you make things simple so folks can easily follow that and then use that within their day to day work? And to find finally, how do you use this information to define what your total addressable market looks like? Who are the people that should be doing things such as AI? Maybe who shouldn't be doing AI in the organization? So generally speaking, there's some key information to collect. First is, you know, what is the general narrative or background of these individuals? Right? So what is your experience using data in the organization? Also, what is their comfort with technology and data? Are they someone who's comfortable logging into a spreadsheet, comfortable with the database? Are they comfortable with prompt engineering from the age of AI? What is their level of comfort? And what are the core needs of data? What are they trying to solve? Are they trying to solve a operational issues, on time delivery? Are they thinking about how do I ensure I'm getting the right market segment? What they're trying to solve ties to many different outcomes, and we have to make sure we align what we provided them with those outcomes. Similarly, there's a lot of frustrations that can happen around, you know, access to data, about how do I use data. So many customers and organizations, how we get data available to a user is actually a pretty big challenge. So how do we do that in effective way? How do we make that easy? Similarly, many users may just be frustrated because they're not programmers. They're not good at coding. Well, we have tools like Dataiku that can help remove that and go to a no code world. So, again, thinking about how you solve for those frustrations. Also, how do you make a wish list for these users? What are people looking to do? What could they wish they could do if they have the right power and technology? Another thing to consider is when you roll up these programs, are the person is the person you're speaking with someone that could be an ambassador or maybe a power user to help strengthen the vision and take it further to the organization? It's a very, very powerful thing to do that. In my own experience, growing some of these programs, well over 50 to 60% of the folks who are growing these are actually from the business side, not just from IT, and they're directly taking us out to the business. So the partnership is very, very important. And then finally, now what are the potentially AI agent use cases for this individual? Had they thought about, well, well, if I could automate this, I get back twenty hours on my week. When you talk to folks, look for the narratives they share and how those narratives can tie to potential solutions. So taking this a little bit further here. You can take those ideas of interviews and take this even further and focus on that idea of how do you do user persona definition, or how do you think about defining an abstracted view of how you look at the people and groups within your organization so that you can work with them more effectively. So with this in mind, user percentas themselves do a couple of things. Similar to our last page, they help you look at the total addressable market or because you can look at the full list of user personas and add that up to figure out what are the right people for me to go after in the organization. You can also use what I would consider observing or kinda design thinking approaches to really understand, you know, when how user percent was operating. The interesting thing about people is that we tend to say we do one thing, but when we actually do something, we may do it a little bit differently. So using an observability approach or watching someone can really tell you, what is this persona struggling with? Where do they get stuck? Where do they get hung up as they work through processes? And then ultimately, the value of a persona is they tell us where the program focus should be and what our backlog should look like, and then really what actions we should take. We know there's a particular division or group of people need a certain kind of help or need a certain kind of work. We should position the program work that we're doing to align with that. And that can mean, you know, enabling one group with LMM type capabilities and certain models or underlying infrastructure. Could mean for another group, maybe going through specific training how to orchestrate AI agents. But it's important to align what you provide directly to what those percentas are needing. Now one way to think about this is how do you actually organize personas? And some popular ways of doing this include level of experience. Is someone, you know, newer to working with technology? Are they more advanced? So think of someone who's using spreadsheets in the software for thirty years versus maybe a more, advanced data scientist who's writing in Spark and Panda data frames. Right? Those are different levels of expertise. You also can think about your business division. Ultimately, those folks in your supply chain division will probably have different needs than folks in supplier sales and marketing, which is different than folks in engineering. So how do you think about what each of those divisions is looking to do and map it into your personas? Again, thinking about the types of data and AI work they may be doing. Are these individuals doing more data prep and engineering and data visualization? Are they gonna be constructing AI agents or building out their own models? What are the types of things we'll be working on? And, again, my answer to our last slide, what are the types of AI agent use cases that also tie up their persona? Again, going to the supply chain world, I can see many use cases where we're talking about efficiency, run time delivery. We're looking at our supplier networks and trying to understand the predictive modeling for AI, how how we make the right decision. Those are different kinds of use cases and say someone in engineering who may be looking at how do I speed up fault tolerance testing or how do I speed up my experimental testing. Those are different types of use cases and assess to a different maybe underlying technology or approaches to data. Then finally, what are their general types of needs? So what are the types of things these folks need to see on a day to day basis to help make them successful? Now the percent attributes you'll see here, you'll notice these tie closely to the interviews. So with most customers, what we recommend is taking what's collected in interviews and rolling that directly into the percent of definition. And as you'll see in the top right, the goal is to make that percent of definition be a bit more humanistic. Right? We wanna make it something that you can feel, you can understand. So it's like you're working with a real person. So it includes seven things like your basic details, you know, a picture. Right? Your role, location, page, maybe a little biographical narrative to make that interesting. Again, talking about what their comfort is with technology and data, what are those core needs, what are those frustrations, and then finally, maybe what are some of the AI agents or even possibilities with those agents. Now as you'll see in the top right here, this is a sample card. So it's actually looking at Star Wars from the twenty tens. So the, the third iteration of Star Wars, if you will. And And what we did, and this is actually based off my experience in working at GE at the time, we actually broke out our personas into alignment with Star Wars characters to make them more interesting for others to pick up and understand. And I give a simple example down here on the bottom right. Within the organization, as we looked at naming more beginner users, they may have been more on the supply chain division. Right? So in that group, they're pretty used to using Excel, and they probably wanna move those Excel processes into more of a standardized pipeline. So in data accuracy terminology, utilizing preparation steps. Right? So you move from that Excel type basis to maybe a no code basis for data engineering. Within our organization, those type of personas were largely called like a Han Solo persona. Someone has a lot of experience. They know the business very well, but maybe they haven't been as exposed to more of the recent data technology. So how do we bring them into that? Another good example of them were the intermediate layer. Now within that organization, you have aerospace engineering type work. And you may have, you know, younger professionals or early career professionals who are interested in scaling, looking at this space. You know, they've taken some previous classes in Python or SQL or things like that. So with that in mind, these type users could be a couple ways. Right? They're intermediate skill. They could start with no code and then they can move into more code based type work. They may cover multiple disciplines. They may do some data engineering and prep. They may be building their own, you know, machine, surveying, and data science models, or they may even start to build their own AI and build up their own AI agents within the organization, especially from a no code perspective. And then finally, you may have more advanced users, folks who would have been from maybe our commercial organization. You can think about this as your your data scientist. You're a folks who've had a lot of experience doing this kind of work. Right? They're comfortable with code. And we used to call those kind of, personas or YODAs. Right? Those ones with the most experience. So this is a very simple view. We talked about, you know, these kind of three personas. So a Han Solo, a Yoda, Ray, right, different ways of looking at it. If you can imagine this being multiple parts of the organization and multiple levels of skill, maybe management and leaders, you can see those can get much larger, with many organizations that, you know, we've worked with as part of our self-service programs here. What we've seen is some groups go with personas that are even 30 or greater, and they tend to align these with their data literacy programs and how they're looking at changing their data culture. Because, ultimately, they want to align the type of work they're doing directly with the actions for those users. So turning us forward here a little bit more, we would like to talk about next is introducing what we call our six s model, which is really a model that takes a look at how do you simplify the idea of looking at your self-service data capabilities or AI capabilities and making them simple to share with the organization, and how do you brand that. And the reason you do that is that for many individuals, right, especially if you come from a technical background like myself or you're an architect, it's easy to get wrapped up in talking about what we have this data lake infrastructure, and we have these particular flows and all these tools and ingress and egress. Right? And if you're a new user just trying to solve a problem or just trying to look at the, you know, day to day work that you have to solve, that can be pretty cumbersome or a lot to look at. So it's important, especially if you're someone who's a technology steward or someone who's CEO, you're trying to lead these efforts to really simplify how you think about working with no data and AI. Now the success model that we share here is an example used across many customers. It's also one that I used previously in my work before joining Dataiku. And you should use as an example to share the customer and say, hey. This is one way you could look at doing this. But it's important to think about how do you define something similar within your organization. What are the right things that are gonna resonate? And what is that right kind of branded mental model that'll help you do this? So with that in mind, we'll go ahead and spend a few minutes here talking about the success model and maybe some of the questions that get into that. So first and foremost, this is model is meant to represent what are the kind of basic steps that every single person goes through when solving really any data and AI problem. Whether they have been using spreadsheets for many years or they're trying to build out a brand new, you know, AI agent or build out a new LLM. They're roughly still doing very similar steps. Right? So first, you need to kind of search for or find the data you're looking for. So if you can't find datasets or the underlying data that might go into, say, an AI agent or a LM, how are you actually gonna build that to begin with? So this really comes to data discoverability, how do you reuse things and elements in the organization. In the context of using Dataiku, you can do this through Dataiku's data catalog. You can do it through various re reuse functions related to project workflows as well as your analytics and models. But the idea is to make your AI assets, your data assets reusable throughout the organization and easy to find so they can help fuel the cycle. From there, many groups tend to do what's called stitching next, which is this idea of how you prepare engineer data or how you stitch it together. For those who have done this a while, it's very similar to doing a simple SQL join. Right? But it's the idea of how do you actually start to combine these datasets in an effective way so that they can be used across the organization. And it's where most, folks from a self-service perspective or from a business line tend to start. Right? When they're in a spreadsheet, they're usually bringing in multiple data sources, entering and pivoting that data a little bit to build an outcome. Well, stitching helps to do very similar work that does it within a broader enterprise data ecosystem, such as a data lake or data mesh. In the world of Dataiku, now we have many no code or code based steps that actually allow you to go through and directly take these kinds of steps, but do so from a business perspective without having the inhibitions of code around you. Another similar step is the idea of science, right, which is really the idea of taking historical data and then predicting a future outcome. And that can be building analytic models, that can be building predictive models, but, ultimately, you're trying to do that future prediction. And many of us may have heard the term citizen data scientist. So when we ask someone who's not from a data science background to kind of build these data science models, this is really where this concept comes in. And, again, coming back to more of the Dataiku world, we have various no code, you know, based machine learning steps that allow those type of system users to do this kind of work. Similarly, we have code based steps that also allow someone from more professional background and data to do the same thing. Now the next item on here is a little bit newer, one that's been reintroduced in the last year or so as we dove deeper into AI. That's the term synthesize, which is really the idea of applying maybe another s from this model to the idea of how AI and agents work. Right? And, really, when you think about AI or LOMs, what's happening is that when you get a prompt, you're asking that, system to say what predict what the next most likely word is going to be based on what I provided you, and it continues to predict those words. Right? And it does this using an underlying corpus of data and information. So, really, what you're doing is you're synthesizing the underlying information. You're predicting what maybe the right next answer would be. And if you extend that to AI agents, what you're really doing is saying, hey. Using the information you've had in previous decisions, predict what you think the most likely next decision is, and then ask agent to potentially make that decision. So we use the word synthesize to kinda take those ideas, which to many can definitely be a very high level kind of technology construct, and simplify it to be something easy for someone to understand. You're synthesizing data and saying this is the next most likely outcome based on historical data I have and then taking an action on that. Now within Dataiku world, and I'm very happy to say this, we have no code encode based capability that allows you to do your own prompt engineering and to go through and do your own AI agents through our AI agent connect. So based on these technologies, you can now extend this out to the broader business community and have them do this kind of work. Next kind of step, usually following all these steps, is the idea of showing or visualizing. For many, the I the idea of a final data product or outcome is showing the visuals around that product or what the known information is that's coming from that. So this could be, you know, data visualizations within data. It could it could also be visualizations in other pieces of software, but But it's another step and usually one that's much more well known to most users. And then finally, kinda coming to the middle here is this idea of sharing or driving reuse. Right? As you create each of these kinda data artifacts and outcomes, whether it's a new kinda prep data engineering dataset, a new model, a new AI agent, whatever it may be, how do you ensure that's being shared back, it's being documented, and being well understood? So, again, if you're within data accuracy context, how do you ensure there's right project documentation for the entire project flow? How do you ensure that any, machine learning models within your registry for your API node has the right material metadata on it? The idea is to consistently be driving reuse understanding around your data so you can drive those next steps. Now I've kinda shared an example here of what a model like this looks like. And as mentioned, some customers follow this model, meaning customers create their own based on what they're looking to do. But I think the most key part of this, if we redirect back over to the it's kind of the why side of this, is that you build something like this so you can organize and bring consistency to how you talk about a program within an organization, so how you talk about your self-service or your AI democratization program, and to help ensure that your communication and your documentation and really even your user journey all align as simple as it's easy to understand. So think about this as an abstraction that you use to easily communicate. So if you're talking to leaders of various business teams in the organization, you can show them the simple model and say, here's how we think about the flow of data, and here's how we think about how your team can engage in this flow. And it becomes a very visual way of working, and you can take that even further. You can embed it into how you document, explain users, go here if you wanna synthesize, go here if you wanna do science work. But the idea is to use this as really a branded way to talk about the work you're doing and to really simplify the underlying technology and architecture. Now, also, as part of building this, you also define parts of your program what your strategy will look like. So as an example, you will be defining kind of the data capabilities that you want your organization to do, including which ones you may wanna have within, self-service and which ones you may not. Right? This gives you a chance to think about, do I want the whole business developing AI agents or maybe I just want them doing data prep work? Maybe I've gone ahead and engineered datasets, maybe go on datasets for consumption, and I just want these users to focus on doing machine learning models on top of that data. This is where you get a chance to really define what those capabilities are. Now, of course, you can also think about what are the steps of those capabilities. So if I start here in this tool, I need to move this to our technology to do this, how do you document that and make it simple for a user to walk through? And then finally, and importantly, you know, what technology support each of these steps? You know, today, I've shared how Dataiku can work across all six of these. Many customers may have other ways of doing this. Now it's common for us to work with enterprise data catalog providers that provide the search and kind of sharing steps as part of this. It's common for us to work with data visualization providers. They'll take the outputs of Dataiku and put it into a visualization. But it's really important within your own organization to define each of those capabilities, what tools support them, and they make that clear for users. So they're not wondering in their head, when they need to go do some work, which technology am I picking, or what should I do next? So I think based on what we've kinda shared here today, we'll be taking a pause here and jumping to some q and a. And I think that's what we have here for today's session. As mentioned at the beginning of the call, this is the first part of a six part series that is going to focus on the idea of, you know, self-service, empowering organizations in the age of AI. And those sessions will go a little bit deeper into things such as data management architecture, talking about people and the structures around organizations, things like governance. So looking forward to have you join those future sessions. But with that, we'll start jumping over some of the questions. Yeah. Thank you so much, Jonathan. That was a fantastic walk through on how to scale self-service and integrate AI agents as part of a COE. As Jonathan mentioned, we'd really love to hear from the audience. If you have any questions, thoughts, or ideas, please feel free to drop them in the q and a panel. We're happy to dive deeper into any topics you'd like to discuss further. Fantastic. So I'll look to, to read each question out and then jump into it. So I believe from Jonathan. What is your view about by agents versus or capabilities versus billed internally? So I think it depends on what do you consider to be unique IP for your organization versus what you don't. Right? And the way I put this is for many companies, there's a unique kind of, capability or IP that they're building that makes them unique within the market. There's other things that they do that are pretty repetitive across our industry. So if that's an example, if you're an organization whose primary job is maybe not sales, right, but you do have a sales component, it may be very easy to take it off the shelf, we'll say agent, say Salesforce here, and leverage that agent to have the same parity of capability that other customers will have doing that same kind of work. That said, if you're looking to drive something that's maybe more internal to your IP or value that you bring to customers, that's where you're gonna wanna think about how do you kind of build your own internal agents. Because, ultimately, you're working with data that's your own unique IP. Right? Think about it this way. If I've got data all over the world that's being picked up by tools like ChatGPT and every single company is using that data, then there's nothing really unique in what's being produced out of those agents. Right? But if I start to take those agents and connect them to my own internal datasets, data that's unique to only me and my organization, now I'm building something around unique IP and potentially unique processes. So I think we're gonna enter a future where differentiation comes from the idea of using your own data capabilities and your own data that you have and leveraging that into your potential own AI agents and models that give you that unique position within the market. Great question. Yeah. Next one here. Okay. So this question's from, from Kieran. Jonathan's view on data management strategy versus use cases and maintenance of an optimized agent. So, I kinda go a few different directions. So if I'm reading this correctly, what I'm kinda reading here is the idea of having a lot of upfront data management. We'll call it certifying datasets, doing a lot of predata engineering work before align the consumption of data, right, a lot of from strategy versus maybe focusing a bit more on the idea of how do we run after potentially new use cases with that kind of ongoing maintenance. Right? So to me, I'm actually a big believer in grassroots type work, and, you'll notice this throughout the series. My focus is on how do you ensure that work happens as close as possible to where the physical work happens to reality. And that's because the individuals who are doing that work are also the ones who know the business process the best. They understand the decision making that goes into it. They understand the data that goes into it. Right? So what I believe in having happen is building, we'll call it an environment that's safe, an environment that allows people to do maybe the right things, but do it in a safe way. So I could tell folks it's kinda like running a playground. Right? How do you ensure people are having fun or able to build the outcomes they wanna build to keep them, you know, safe from either hurting themselves or others? And then when it comes to some of the content we'll cover in later sessions with the IT of architectural governance. But if you build the right architecture, the right guardrails, the right technologies, you can empower the organization to very quickly innovate, try many use cases, try lots of new ideas, and you can track and see what are the most popular ideas that are happening. What are the ones that get the most traction? Right? Maybe this AI agent that was built is getting used five times per month. This other one that was getting used a hundred times per month. Well, the business is kinda speaking and saying, hey. I wanna use the one that's been used a hundred times per month. So maybe that's the one you start to promote up the stack. So when you think about that and you start to look at that overall adoption, you allow that grassroots work to happen. Eventually, you can then move, let's say, the most popular or most used data products from a self-service perspective into the actual IT teams. So when you think about data management, one way to do that is actually have your datasets be engineered by your self-service population, find the best definition of those datasets, and then move that into your gold or certified data to be managed by your IT teams. In that way, you're building an innovation funnel of sorts. We allow new ideas to flourish within a safe environment, but then capitalize on those best ideas and fold them back into your data management overarching strategy. Great question then. Alright. And then I think I get another one here from Victor. Have you seen agents used in the context of learning and development? I think we're very much, starting to get there. I have some customers talking about those use cases, and I think we're starting to see it. And I think part of that comes down to the idea of how do you start to look at your now you have to be careful, like, because HR data always has a lot of sensitivity around it. But how do you start to look at even potentially where folks may be successful or folks may be struggling and start to coach them how they can even get better. Right? I'll even give you one example. Here at Dataiku, we actually have a piece of software that we use, probably the software's name out, but actually can record our presentations of the sessions that we did either the customer's live or potentially just internally on our own practicing. And throughout those sessions, the tooling uses AI to actually track our behaviors during the call to look at how our eyes kinda focus on the screen, speed of our speech, how do we engage with fictitious users, and it uses all that to provide feedback directly after the session in real time. What's really interesting about that is you're actually using AI as almost an adviser and assistant that's tracking the type of work you're doing and then directly giving you feedback for how to improve. And I think you can take those ideas and go much further. You can look at someone's overall body of work and say, hey. Maybe we could work on your, you know, succinct to some writing skills. Right? This is something I could probably poke myself at a little bit. You could also have AI take a look and say, hey. You see you've done these presentations. This was the attraction or adoption based off those. How does that fall back into a training plan for you? I think we're starting to see organizations consider that more and more, again, with a very specific caveat of make sure that you're not getting into specific PII or HR data for the individual. Okay. Lots of good questions. I'm loving this. Alright. So, we'll go with, I believe Koichi, if I said that correctly. What are your thoughts on how what are your thoughts on how to push a expertise from the COE to each so I believe if I'm reading the rest of that, I'll read more. One second. The new interface. To reach operational unit so that each unit is doing AI right and maximize value to the unit. Yep. Awesome question, Koichi. So I think I started to address that a little bit here a few minutes ago, and we'll definitely go into this more especially in the next up two coming webinars, which will focus more on data management, architecture, and really governance. Right? To me, if you think about a COE function, the main functions of the COE in today's world from a data perspective are ensuring that you have a, we'll call it, enterprise source of truth. So how do you manage references to data across organizations so they're easy to use? How do you enable the organization to work with data? So enablement technologies such as a Dataiku. Right? And how do you think about putting the right governance and guardrails around that so that people can effectively work with that environment? To me, once you set that up correctly, you create an environment where people can innovate and build new things. Once you've done that, you can ensure that folks are starting to build, and we'll call it AI the right way. Now the definition of right can mean many, many different things within an organization that's unique to each company. But you can build those guardrails to help ensure they're building it the right way. And then when you say, hey. You wanna go to production with, you know, we'll call it agent or LLM, then we'll have the ability to ensure that you're following the right standards to do so. And the reason I like that approach is you are not necessarily dictating whatever unit does and every single project is. Right? What you're saying is we trust those business units or individuals to know where value is, and we're gonna give you the right tools and technologies to build towards that value. Now, again, we'll cover this in a later session, but, Dataiku LM mesh technology is fantastic for this. Right? You can set up the right routing for specific AI prompt request to the right agents. You can help ensure that you're managing your cost over time. You can help ensure that you're removing toxicity, all these different things, but you're doing it at a routing layer where you can have really anyone across the organization accessing and doing that work. They have the right filtering and control on top of that to help ensure you're keeping things safe and running productively. And I think to me, that's a great example of how you can kinda enable that democratization, but also help ensure units are building specifically to the value they need because they know the work they do the most. Okay. I'll go ahead and jump to I guess, looks like Michael's at the top here. So from Michael, Wong in COE Models, who typically leads MVP development for self-service or VPN free data access, the IT team for integration, and the business team for use case definition, how is well clarity maintained? So if I'm understanding this correctly, there's actually a couple components here. Right? There's that kinda we'll call it MVP development. I'll rephrase this a bit. I think the question is asking, how do we get access to those datasets in a more governed way across the organization while allowing users to quickly kinda MVP or build things quickly? So I think there's a couple of key ways to look at this. First and foremost, if you do not have a we'll call it a source of truth for your datasets, data lake, data mesh, etcetera, I'd actually recommend just initially starting with one of those key value use cases in getting the datasets for it because it can take a long time to build that kind of reference layer for data. But once you've got a bit more of that happening, you're starting to build a broader repository for data analytics. There's a few ways to look at it. One is, are there datasets in your organization that can just generally be viewed by most employees? You'll find in most companies, so I'm including Raican from GE Aerospace, that there are datasets that can be more open and datasets that need to be restricted. So it's important to send those classifications. If datasets can be open, you know the business is gonna use them. I'm a fan of actually enabling access more broadly to the company, maybe through one particular role or ID to see all that data. Now if you have data that has certain classifications, in my experience, expert control is a big part of that, so data can only be seen by certain residents in certain countries around the world, then you wanna make sure your underlying data and your access technologies such as Dataiku are aligned correctly business permissions and roles. The other things that I would need to consider are the specific kinds of roles you enable. Now who owns datasets after they're created? If you enable self-service and you write a dataset back, who actually owns that dataset? How is that classified and managed within that data lake ecosystem, etcetera. What I will say is access controls around self-service and data is actually part of the next session. So, Michael, I would recommend coming back to that one as we'll cover that further. So keeping track of time here. I'll jump to, Jonathan here at the top. So it says great stuff. What are the KPIs you would recommend manage the effectiveness of the COE? Fantastic question. We'll actually cover that directly in the final webinar of the series. That said, there's a couple ways to think about it. I tend to look at it in two different ways. One of the value KPIs that you use to communicate value to the organization. The other ones are really your day to day program management KPIs. Right? I'll kinda focus on the value ones here. The top three I see are no. I think first to everybody is return on investment. So how do you show the actual value coming out of these data products and solutions within the business? The next, I would say, is really, the idea of active users. So one of my favorite terms is that people build with their clicks. So to us here at Dataiku and then many of our customers, what's important is that that somebody just has a license. What matters is that someone's logged into that software and actively using that software. Right? We wanna see you be active on a a daily, a weekly, a monthly basis. So I think to me that activity is important because, again, you wouldn't necessarily click into something unless there's value. And I think the last one is really business process adoption. How do you track the actual adoption of something in a business? How do you determine if it's actually making a change to how the business functions? And, again, based on my prior experience, I usually tend to look at that as about 50 unique users a month, whether that is a dataset, API to a particular data science model, visualization, agent, whatever it may be. Once you've got about 50 or so unique hits either from machines or users in a month, it's very likely that that particular data product has got a lot of adoption, and that's a good point to think about what that business process adoption looks like. As mentioned, we'll cover more of that in the sixth webinar. But, again, great question. I think with that, I only have one more on here, I believe, from Kieran again. So use of semantic AI within LMs and its effectiveness. I think it can definitely be, be useful. Right? When we're really thinking about semantics, so the idea of tagging or how do you drive how AI makes decisions, I believe that's important in general. Right? So if you think about even, the idea of just a self-service user, remove AI for a minute, if someone's searching for the dataset they need, they're gonna need metadata around that dataset to find or to use it. If you just see a, you know, Oracle date, table database name, you're not gonna understand that. But if you see a name that says, hey. This is your parts and sourcing data, you probably will. So to me, metadata around data is incredibly crucial, not just for your, you know, AI systems, but actually for your users. You have to have that in place for the need to understand how to do that work. Now if you expand that further, having those same metadata and tags around your datasets can also make AI smarter. So if you go back to, again, the examples I gave really about expert control, if I know datasets can only be read in certain parts of the world, not others, but I'm gonna have an agent that's working with data and users in one part of the world, where it shouldn't be accessing data in a different part of the world, Having that right tagging information can help me make sure that my agent or my LLM is excluding those datasets. So to me, there's many, many use cases, but I'd say a primary one is helping ensure that you're not using the wrong data for maybe a purpose you're trying to solve and that you're only using data that won't give you additional risk. I'll keep going unless someone stops me on questions, but I got another one here at the top. What are the most have skills and roles in the COE? Great question. We'll cover that more in our fourth session, which we'll talk about people and structures. Right? What are the right roles? I'll just cover a few of the key ones here since this is, you know, getting out of your session for q and a. But some key ones are one, I believe very strongly in product ownership. I think any COE should have product ownership that is focused on working with the business, working with internal teams to help ensure that what's being built and developed is aligned with where the business needs to go. We've talked a lot about here. They've got personas, setting aligning a vision and strategy. That product owner should really take the helm to ensure that the work the team is doing or the COE is doing is direct line with the business value. Another kind of key role I would recommend is really someone who's like an applied SME, or within Dataiku's terminology. We call it data scientist in residence. This is someone who directly knows how to use data technology, use Dataiku, name how to engineer datasets, build agents. They're an SMS. I mean, right, and their role is actually to work out walk with the business, to help walk them through how do you actually use this technology. They're the ones who do maybe some of the hand holding. Maybe they hand run the open office hours or they answer questions. Right? So these are applied folks that can really help users. I'd say the last one is really a platform architect. So this is the person that you might think of as the admin, right, or someone who's actually running a specific technology. Within Dataiku, we call that a technical account manager. That's someone who is actually ensuring the platform runs appropriately, that you have the right features installed, that you ensure you have operational consistency and things like your architecture. Now without going deeper, there's also additional related to business, there's roles related to training. There's lots of pieces that go into that, and we'll cover more of that in the fourth session. I think we might be through all the questions. Looks like we've done plenty. I think we're also a little over on time. So with that, I'll pass it back to Renata. But, thank you again for joining, and I hope to see many of you next week, next Monday for the next webinar. Yes. We are slightly over time, but we love the excitement and the engagement. So thank you everybody for asking your questions. Speaking of continuing the conversation, Jonathan, if you could go to the next slide. Our next webinar in the series is coming up next Monday, May 19 at 11AM eastern time. In that session, Jonathan will be back to walk us through how to lay the technical and metadata foundations in your COE from building resilient data architectures to ensuring your self serve service platforms are scalable and well governed. You can scan the barcode to register, and if you can't make it, don't worry. We'll send you the recording as long as you sign up. And so if you're looking to build a strong sustainable foundation for your COE, this is a session that you won't wanna miss. And with that, thanks again, everyone, for joining today's session. Thank you so much, Jonathan, for your time and your expertise. We look forward to seeing everyone next time. And other than that, have a great day. Thanks, Roy. Have a great