From Data Silos to Intelligent Self-Service with Celso Poderoso

Speaker A: Hi, I'm Chris Adamson, Director of education at TDWI. I'd like to invite you to attend TDWI Transform 2026, an in person education event taking place from September 20th to 25th in Anaheim, California. We have 57 master classes on the Transform agenda taught by 30 expert practitioners and thought leaders. They cover all areas of AI, BI and data management, with offerings for people of all skill levels. These are not short sessions or talks you'll find at other events. They are full courses and they're practical, actionable and vendor neutral. The event also features a keynote exhibit hall, networking opportunities and more, plus a co located 3 day summit for data and AI leaders. Consider TDWI Transform for your team's professional development this year. When you register, use Code podcast for an additional $100 off@tdwi.org Anaheim. I hope to see you there.

Speaker B: Hello and welcome to Speaking of Data, presented by tdwi, the premier data, AI and analytics podcast. I'm your host Andrew Miller and on today's episode we'll hear from Celso Poderoso on moving from data silos to intelligent self service. We hope you enjoy this episode and for more cutting edge information on data, AI and analytics, please visit tdwi.org and with that, let's get started.

Speaker C: All right everyone, thank you so much for joining us for another episode of Speaking of Data. Megan and I are very happy to welcome Celso Poderoso on the podcast Today we're going to be discussing from data silos to intelligent self service and within that kind of underpinning, uh, data fabric and usage of semantic layers. Uh, Celso is going to be joining us for our. Has recently joined us. Excuse me, recently joined us for our May virtual summit, uh, around the uh, next generation business intelligence, Harnessing AI for analytics and self service was uh, fantastic to have you as a part of that virtual summit. Uh, but before we jump into all of the questions, I'd love for you to share a little bit about yourself. So, and your role that you're currently in, um, to let our audience know a little bit more about you, if you don't mind.

Speaker A: Yeah.

Speaker D: Awesome. Hello Andrew and Megan, thank you so much for having me. It's really a pleasure to be part of your podcast. I've been working now more than 25 years with analytics in general, sometimes closer to the technical stuff, but most of the time close to the business. My most recent role is exactly, uh, connecting the dots between the two sides. So I'm very, I have very deep knowledge in terms of technology, applying technology but uh, now I'm more, let's say that closer to the business and understanding how they can use the data, how can use agents, AIs for example, to, to get um, the job done. So it's really exciting being now uh, as a global director in business reporting governance at McCain Foods.

Speaker C: Well that's fantastic. Yeah, clearly there's a reason why we had you join us here today and ah, join for the virtual summit for that wealth of experience and knowledge that you have here. Uh, let's get down to the questions here Salso. As I mentioned the topic from data silos to intelligence self service feels like a transition. Many organizations are currently underway or maybe you know, potentially looking to try and ease into more intelligent self service and less of those silos. Uh but clearly there's a lot of terminology that comes along with this topic uh, in this space. So I want to ask as I mentioned a little bit earlier, about the data fabric. So can you define what a data fabric is and why it's becoming so important for these modern enterprises?

Speaker D: Yeah, that's really a great question because we have so many different definitions that sometimes it gets confusing. Right. But what I love to understand data fabric is um, mix of data governance and data management. So how it works, right. So if you have a fabric, uh, we think that it's very common and implicit on this uh, term that you have sequence, uh, you have a uh, defined sequence of actions to build something. Right. So and that's exactly how the data management works. Uh, sometimes we call the data ops that we, you have all the, the processes defined and so on, so forth but always with governance. So without governance uh, we lose a lot in terms of um, a lot of, in terms of business especially. Um, that's why I love to use these two concepts together when we are talking about data fabric. Uh, in my previous experience what I did is exactly use the Data Lake House architecture to implement a data fabric with governance.

Speaker E: Um, thank you for that definition. I think uh, earlier in the summit I might have heard you describe the unified semantic layer as the brain of the data fabric. What do you mean by that? Can we unpack that a little bit?

Speaker D: Yeah, definitely. And this is really crucial. Right? So um, the semantic layer I would say, and I used to say that this is the brain of the data fabric. Exactly. Because uh, this is where you will connect the technical piece. That is where the data is, what table, what columns we are talking about and how the business will use this column. So you see that the semantic layer will connect both dots, uh so you will have the business logic, the content, the standard definitions, all the metrics. So everything related to how business will use the data will be in the semantic layer. But naturally this data that is used for business purpose, they naturally comes from one specific table, one specific column, and that's exactly where both things connect. Uh, in my experience with that, uh, we had an opportunity to select one tool. Uh, I would say that it would be the 1 visualization tool, but that makes the semantic layer as well. And our approach since the beginning was exactly this. So each column in each table will have only one mapping to our semantic layer. And we will put, we will set the definition uh, of this column into the semantic layer naturally, as you know, right. So it's not everything that is in a database that will be only these will be used when you are building reports. So um, but having this concept, whenever you create a new view, whenever you create a new KPI, for example, you have the opportunity to add the semantic concept for that KPI the same way that you do with any table, uh, in any column.

Speaker E: We see in TDWI data, um, how important the semantic layer is becoming organizationally just because it's really seen. I think in a recent report it was like 70% of people said the semantic layer was critical. A unified semantic layer was critical to success with AI. Yet the stats on companies doing it well and having a unified semantic layer that is consumable both by the business and by AI M and having that, it's way below that. So it certainly is an important topic that we've been hearing a lot, lot about lately.

Speaker D: Yes, definitely it is. Um, because in the end of the day, right, so we need to, to, to um, to break the gap between business and tech and technical stuff, right? So this is something that we need to use and even look, since you, you mentioned AI and agents, AI in general, uh, it's quite important, right? So because it's not only having the data, but what's the context, how you will use this data, what this data is for. So you see that you cannot limit yourself on what is this column, what is the data type, what's the size of this? It's not. Right, so you need to have much more than that. Even in our experience, what we did is, is create on top of this semantic layer, recreate a total, a complete business glossary for the company. Right? So then it, it exactly too because then you can, you can make things make more sense and open the opportunity for agents, uh, for sure.

Speaker E: And one, one area where I know that I've read in one of our reports like that some of the, and I'm not hugely technical, so I bet that in BI systems if they're like sometimes there's multiple systems across a company and those might be getting read by AI and there might be conflicting language in the semantic layer in the metadata. And that's, I imagine that's why having a business glossary with accepted terms and why this is all important because if there's a uh, conflict in how you're describing the data, it can't be utilized probably the most effectively.

Speaker D: Yeah, Megan, this is great because you remind me a very common situation that happened in any company that is sometimes you have a, um, executive meeting for example, and you have for example marketing talking about one specific data. Then you listen, uh, the CFO talking about the same data but giving a total different approach or even different numbers and you have sales with the same thing but again totally different. So it's crucial today because you see that if you are in a executive meeting, you have everyone in a single room. It's not easy to break this gap. It's complicated because everyone will try to put. Okay, no, my definition is correct. No, no, you're wrong because you're not considered this or that. Imagine what one AI agent would like

Speaker E: that a machine in the mix and like can't it, needs it, can't have the discussion about it. Um, so yeah, it's really interesting, um, and I can see why it's so, um, even with my non technical, I'm more of a marketer here and um, you know, helping people learn about tdwi. But it definitely makes sense, um, on why it is so important and building

Speaker C: off of that, I mean you led me so nicely to my next question that I had for you which is like how does a semantic layer help eliminate data silos and create a common business language? So I think we kind of addressed the common business language part of that. But maybe you can drill down a little deeper on the siloed aspect for us here, uh, specifically around how the semantic layer helps solve that issue.

Speaker D: Yeah, we can list several. And I totally agree with you, we talked a little bit about that. But look, one uh, thing that is, uh, again my previous experience was exactly on top of that. Uh, and by the way, just to explain, we created first the semantic layer, then we created the business glossary. So, so you see that perhaps if we did it um, in the same time it would avoid some mistakes that we made in the semantic layer. But at that time we didn't Have a data governance console, for example. So we needed to use the definition. Whoever requested that data, we get the definition from the user. And that's exactly when in one point, uh, we saw this, what I said happening, right? So same metric, same data, but with different uh, visions, different interpretation. Right? Uh, and then we decided to create a data governance console. Exactly. To bypass this situation. And then, okay folks, we have only this. This is coming from that again another concept on the governance is extremely important because you must have the lineage, right? So usually we, we said, okay look, this is the data in the source app. Ah, and this data is here. We cannot have two definitions for this. So you see that because it's only one and it's coming from this source. What, what I sometimes see is exactly, uh, and that's how we, we broke the data silos internally is exactly like this, right? So we must explain them. It came from here. We must define once here, if you want to have a different view, probably you are adding some other variables that we need to consider and we need to make clear for everyone. So that's where the business glossary help us uh, a lot. Right, so standardizing metrics, breaking down definitions. Um, right, so yeah, basically that.

Speaker E: So once you know, you've got the business glossary, you've got your semantic layer in place, you know, you, so you're all speaking the same language now about the data, then how do you decide, you know, in self service and making data available to people like, how do you um, balance getting that broader data accessibility with governance security compliance requirements? What's your experience? How have you done that?

Speaker D: Yeah, that's really great because when I just arrived at the company at that time, uh, one of the main uh, requests that I got is okay, we need to make data pervasive for everyone. Or in other words, right, we need to promote data democratization internally. Uh, it was before the Genai explosion, right?

Speaker E: So I mean public service and democratization has been a key initiative for a long time.

Speaker D: Exactly. And this is exactly the point, right? So uh, we needed to make everyone uh, aware of the data, but we have both sides because if you open everything to everyone, it can be risky. So you need to balance the compliance, you need to balance security. It's quite important. So what we learned is that in the very first moment we open to everyone and then when we notice that we needed to take care of part of the information, we use the business glossary to filter and to make uh, into their tool compliance, to make things more uh, manageable or go over medicine.

Speaker E: M. Um, we, you know, we had a conversation, uh, a few days ago with one of the other summit speakers on generative, um, AI and the role of generative in bi. Now do you have any perspectives, uh, on that and whether it's helped, like do you see it helping accelerate democratization of data?

Speaker D: Yes, yes, I do, I do. Um, and mostly. And that's why, um, that's why the data fabric concept is very important, right? Because uh, it's not only governance by governance, it's not only process by process, it's a mix of both that will make things um, clear, available and governed. So that is important because uh, in just reinforcing the message, uh, it's extremely important for any gen AI initiative or agent AI initiative that we can give high quality data and context on this data. Otherwise uh, if we don't have internal, uh, understanding on the data that we are producing in a daily basis, uh, don't expect uh, any agent to be ready to do that. That's where hallucinations happen. That's where people get frustrated on some results and you lose a great opportunity to get a big, very big uh, step, uh, on this side, the agent side.

Speaker E: It's funny, some things in our 25 plus years and don't change, like data quality is, you know, at the heart of it all there, um, and then context becoming so increasingly important.

Speaker C: Well, and sorry, go ahead, Salsa.

Speaker D: No, no, no, please, please.

Speaker C: I was going to say like building off of your story about how you, you know, implemented a data fabric semantic layer business glossary, uh, for folks and organizations out there that might maybe just be starting this journey, uh, from your experience, what were some of the biggest mistakes or maybe some misconceptions organizations, uh, have when pursuing the data fabric or semantic layer initiatives.

Speaker D: Great, great question. Because you know that I would be lying if I said that we did everything correct. It's a big lie. No, we, we made some mistakes and I think that this is part of the, the learning curve. Right? So you, you will make mistakes. But what I could say that would be the biggest one of the biggest one is. Exactly. If you try to use the term data fabric just because this is. Oh, everyone is talking about that. Uh, uh, no, you need to understand exactly what is a, uh, data fabric. Right? Don't, don't put labels yet. Understand exactly what you want to get from a data fabric. Have your methodology in place. And this is another point that is extremely important. Having a methodology, having ah, um, consistent architecture. As I said, we use the data lakehouse concept Exactly. Because, uh, it was something that we, we knew at that time. Uh, and something that we could track each step. We could insert our methodology in each step. So the first thing that I would say is exactly this. Please have a methodology. Don't, don't try to do just because Data fabric is a great name or let's do it. No, no, no. Understand, uh, exactly what it is. Have a methodology input in practice. Second big error. Uh, don't use semantic layer. Oh, we will do this later. That will be really tricky. Especially if you have a, um, big, large environment. It will be tricky for you. Right. So, um, even us, that was not at that time a huge environment. Environment. Uh, when the small gap that we had between semantic layer in the business glossary. M. We needed to do a lot of meetings just to reveal all the things and we need to spend a lot of people hours to review everything. So you see that uh, if you start doing the right thing, it's much better. I would say that this is really important. Um, don't try. Another thing that is really important is that, um, look, the business is always right business. They are our clients, so they are always right. But, uh, sometimes we need to put some. And say, folks, look, it doesn't make sense. Let's discuss better. And for that it's very important having a good sponsorship because in the end of the day, uh, someone needs to make the decision. And this is really, really important. Very close to this one. That is exactly. Uh, build a common language. Don't think that this is not important. You must have a common language internally. What I'm seeing sometimes is, um, I would not. It happened with us as well. That's why I'm saying that it was not perfect. But in one point in time we asked LLMs to define things for us. Yep, big mistake. Don't do that. You have internal knowledge. Right. Use it.

Speaker E: Yeah. So how hard is it to like the development of that business glossary? That's like the coming together of lots of people's different interpretations of data and how they describe it and how they describe the business. How hard was that? Like, I've never been through a process like that. So I don't know. Is how hard.

Speaker D: It's hard. Trust me, it's hard. M. Yeah, it is my recommendation. One error that we made as well. Small, uh, groups. Small groups. Right. So try to use kind of hierarchy to do that. Uh, who is able to define that concept or that term? Uh, yes. Take this person into the the room and let's discuss. Remember. Right. So if you have 20 people in a room, it much complicated to get. Yeah, I can only imagine it's almost impossible. Right. So what we learned in, in the very end of our, of our process, uh, we had very few people, but people that really were SMEs, right. So they are, they are uh, expert on that subject. And then even because they are SMEs, uh, other people will respect what they say. Yeah, so use it. Yeah, exactly, exactly. So use it in your favor. Right? So use it to get things done quicker. It's important because if you have a project and then you, oh, it will last six months, eight months, uh, my gosh, don't do that.

Speaker E: Yeah, that does, that does not sound like fun.

Speaker C: Uh, I'm curious. You know, it sounds like so much of the time and effort was put in place obviously getting this off the ground. What did the rollout look like for. We're talking about like bringing in the self service aspect here. When, when it was established internally with those small groups. How difficult was it to get people on board, uh, to even figure out that these new terms were what they were now defined as? Was that also a difficult task?

Speaker D: It is, it is. But again, uh, a good sponsorship can solve this type of problem. Right. Because they will say, okay, this is important. But I think that even, even in we uh, run in parallel, uh, data literacy program as well. That is exactly to show the importance. Right. Why it, is it important for the company. Right. So everyone, uh, usually, right, people try to, to see only what is close to them. Uh, so you need to offer a different approach. You need to, to show them that it's not only them, what they are doing will reflect on other job, on other people's job. And you need to, we need to work together. So you see that I think that this is, this is a quite important point as well, Andrew.

Speaker E: Um, well, thank you so much for coming on. The thing that I always appreciate is people love to share their successes, but I also love when people share the mistakes they've made because it's so important for those that follow behind you that are just getting started to save them time and money and all of those things. Um, our last question, like sort of a future looking question, um, how do you see the semantic layer data fabric shaping the future of AI ready self service in business? Um, and what excites you about it, you know, into the future.

Speaker D: Yeah, that's really great. Personally, I don't know how the data visualization tools will reinvent themselves because in the end of the day if you have a very uh, consistent business glossary. And again it's mapping correctly to its source. Conversational, uh, analytics is something that is very natural. So you don't need to, you just ask the data and the data will come to you. Right. So if, if it's. We could experience this there. Uh, because yeah it's not, it's not only self service analytics anymore. You see that it's much more than that. You, you can, you can get insights from agents, uh, you can get job done by agents, uh only having a uh, consistent semantic layer, having a consistent business glossary. So you see that, I see that in our experience we had this opportunity to get the conversational ready but even uh, this is a side of our conversation but just to understand better because we had a very uh, well defined process, uh, we created agents to support people. Our, my team at that time, uh, to do the business analyst tasks, data engineering tasks, tests. Right. So the DBT tests. So yeah, everything was exactly. Because you have, you have everything controlled, you have everything uh, with a good knowledge over that. Right. What I would say is exactly don't expect agents, AI and AI in general, uh, replacing us on everything. Right. So and for us by the way, that we have a, we should have a very great methodology. The results come even faster.

Speaker E: Yeah.

Speaker D: Because they will just follow what we used to do. Right. So and that's why it's important And I always talk about this uh first step, organize yourself, organize your environment. That, that's the first step. Then try to use it uh, and the results will come even faster. Right.

Speaker E: So yeah it is pretty, I mean it is amazing thinking about the possibility when you start to see when you are using. I mean I myself have used you know, natural uh, natural language interface to query like well structured data and have it deliver insights and at the most basic level. But when you start to do that and like have a conversation with the data so much I, I've watched sit like it generates all these other ideas and then you're, you're asking it more questions and like I am so excited to see what happens in the future because I can see so clearly the potential and with agents, I mean I've built a few of uh, my personal agent. I mean I have an agent that helps me prepare for this podcast based on all of our research data at tdwi and I you know pass it to Andrew and I have it myself and um, I just. It's an exciting time, exciting time to be in our space and and so I thank you so much for joining us today and for being a part of the TDWI summit. Um, and you know, um, for those of you, our listeners today who are really interested and excited about things we talked about semantic layer, data fabric, just generally creating a stronger, more value, creating sort of AI ready data foundation with the things that Salso talked about, you know, a mix of data management and data governance. I would encourage you to join us. Um, we do have TDWI Transform in Anaheim which is running September 20th to the 25th salsa. You should come join us there as well. Um, but lots of these topics are on the agenda. We have um, a data and AI leader summit that focuses um, solely on data foundations for AI. Um, so that's going to be a great day focused on a lot of these topics topics as well. So hopefully all you listeners out there can join us. But Salso, Andrew and I are so appreciative for having you take your time to talk with us and I um, know our listeners are in for a treat with this episode.

Speaker D: Thank you. Thank you so much again for inviting me and it was great pleasure talking

Speaker C: with you both before we let you go. Soso, if people listen to this and they're interested in asking a question or following up with you, what's a great way for them to, to get in contact with you?

Speaker D: I think the LinkedIn today is the best one. Right. So, um, my LinkedIn is Celso. It's not very common here in North America but it's C E L S O. Uh, and my last name is Poderoso. So P O D E R O S O. So it's uh, the best way for us to get in contact.

Speaker C: Fantastic. Thanks again Celso. Really appreciate your time.

Speaker D: Thank you. Thank you so much.

Speaker B: Thanks again for tuning in to speaking of data presented by tdwi. If you'd like to engage further with us, please check out our website@tdwi.org and

Speaker C: please don't forget to follow us on

Speaker B: Apple, Spotify or wherever you get your podcasts. And if you enjoyed today's episode, please take a moment to leave us a five star review. Take care and m we'll see you soon.

Speaker D: Sam.