PODCAST | The Data Model Behind Mars’ Analytics Evolution

[MUSIC] >> Hello, and welcome to CDO Maggi and interview series. I'm Sachin, Sachin Prabhar, vice president of consulting at Tion Analytics, a leading provider of data and AI services. There's a problem that's plaguing the enterprises today. Organizations have a mass significant amount of data, and yet many leaders cannot act with confidence. There is this chasm between data and insights. Today, we'll explore why this gap persists, how to fix it, and what does it look like in practice? We have with us, Ujjul Sahejal, global head of people analytics at Mars, and Rachel Bellino, HR data officer again at Mars, to talk about this very pertinent issue. Ujjul, Rachel, welcome to the podcast. If you have to think of sort of a solution framework, Rachel, what are the key components of such a solution would be? To realize this kind of a way, generate that you're painting, so for example, something that I have been hearing from some of our customers, is that having a common data model could be a great, great starting point, right? But obviously, that's not enough. So just wanted to hear it as you are solving this for your organization. What are some of these key solution levers that you are pulling, right? And what are your learnings? Yeah, it's interesting because when I started this journey, I actually started very much with the, and again, I hope this resonates with the audience very much with the theory. So I had a conceptual data model, and a logical data model that I have built out for people data. And that's the first things that I've worked on in my role and in this function. And lo and behold, it actually wasn't that useful for us in the beginning, because we needed to deliver products that were solving very specific problems, having that conceptual data model wasn't that meaningful. And now, however, and here's why, because as we've gone along this journey, and we went from very product-specific data to then reusable data that was then common and being used across the board by all of the products in a consistent way, then we ran into the challenge of, while that's great, it's not unified in our data platform, in a way that really can allow us to traverse across, as well as traverse it in time, so backwards, historically, in a very easy way. So that's when I felt that a common data model became a critical part of what we needed. So it's great that you're hearing that section from your other customers, because I do think that it's, when it makes sense for you as an organization, it can really unlock data discovery for your data scientists. And then also, for those use cases that require you to cut across the data domains and/or specific subdomains of the data, and require you to bring that together to have a more holistic package of insights to offer your stakeholders, then that's where I think a common data model is a game changer. Absolutely, absolutely. And I think the point that you're saying, that you had to go through that journey and see how do you solve for it in a way that it can encompass all the use cases. Now, one of the other challenges that I've seen is some of these things takes time. They're just saying that sometimes you have to walk slow to run faster, and at the same time, the business doesn't stop. Business has to be given the value here and now. So, love to hear your thoughts on how you guys have been able to balance those critical asks from the business to deliver the value now versus building some of these foundational stuff which takes time. Which you'll, I'm pretty sure that you would have been in that space where your customers are asking that, you know, hey, I want to have that use case now, right? And how do you deal with that? Now, that's a question we get asked almost every hour if not every day, right? So, that's a very, very common theme that we've seen in terms of how do I get access quickly to, you know, what we are looking for. I think the way we've approached this from a Mars standpoint is primarily keeping the user in the center and really trying to understand, you know, what is it from an empathy and understanding standpoint? What is the goal and the objective that they are, you know, really trying to achieve? And how is this analysis or how is the output really going to help them answer that? The reason why we always start with the user at the center obviously is because A, we want to understand what the business problem is. But then the intention also is how do we design the solution so that it's very easy for the user to clearly navigate through the analysis, get the insights quicker faster and they don't have to really figure out what drop down do I have to do or what filter do I need to run for me to get the answer, right? So, we always start with the end user in the center and design this. We also follow this principle that we've tried to embed into our operating model, which is build the front end while the back end starts to catch up, right? And while we start looking at, you know, what data is needed, get built that whole semantic or contextual layer of bringing that data in, we already have a clear design of what that front end or what the user is really looking for. So that just helps us expedite this whole process because again, some lessons learned where we always started, okay, tell me what you want or, you know, yes, yes, the data I need just do this analysis for me. And when we build those kind of products, we've really seen adoption rates are low because the users can't figure out how to navigate, how what to do, how do they make the decisions. So, you know, some lessons learned from that, which is where then we've turned this around to say, all right, let's bring the user in the center first, understand what they're looking for and then build the solution really centered around that. So that's the approach that we've taken your ad models. Yeah, I think that point on the year's interest city, while simple, but at times gets missed, right, I have seen it definitely first and in many situations. I really like the point that you made about building the front end and let the back end catch up, Rachel, I'm pretty sure that mostly you end up with that back end catch up thing, right. Tell me how do you deal with that. I do, but at time, you know, I think the first time I heard usual say that is quite controversial to my ears. But it's absolutely spot on and I'll tell you the benefits of it and the benefits of it is that, you know, a lot building a foundation requires investment. Very significant investment for your organization, both in terms of money and time and prioritization. And therefore, it's really important to demonstrate value tied to that investment. And so if you are focusing what you're doing on the foundation on what has been proven successful in terms of the pilots or the solutions that, you know, we've put forward, especially with our front end, then it's very easy, I would say, or straightforward to justify the investment that you're making in the foundation. And I think for any executive or leadership team that's very important to show them that, because I know that many of us live in a world where, you know, our analytics functions as well as any other functions, but we all want to demonstrate the value that we're bringing. So I think that's one of the ways that that's definitely a big advantage and a benefit of the backend catching up. I see the other advantage I would say where the backend catches up is, you know, frankly, it's kind of nice. It gives me time to be thoughtful and mindful and plan and come up with what is that roadmap that builds the foundation to be scalable in support of what is being driven by our stakeholders and our users and that front end. So, you know, rather than, you know, putting the cart before the horse, you know, I mean, okay, this is what we've seen this is what's happening. These are the priorities of the business coming through and showing up in our products. Therefore, this is the roadmap that we can do to scale and support that. Got it. Got it. No, very, very helpful. And again, I think this definitely is something that I'm pretty sure our listeners will also kind of read. try to practice and take a take away from this conversation. Let's talk about agents now, right? That's all the buzz right now. And while there has been a lot of excitement about bringing the agent ecosystem into the play and doing a lot with the external side of the equation, what I mean by that is like, no, like maybe optimizing or marketing function, right? Or how you engage with your consumers and things like that. But there is also a huge space ready and right to be disrupted, which is that internal ecosystem, right? Just the way we do things, the way we do our engineering, the way we do kind of manage our data. Love to hear how you guys are leveraging some of these newer constructs, right? Again, in the interest of serving that original goal of bridging the gap between the data and the insight. Rachel, maybe I'll start with you and then I'll come to you next. The buzzwords are all out there. We're all familiar with it. My favorite is the Agente AI framework. And here's why it's my favorite because I hate black boxes, I hate high words. And it's I love digging into it and figuring out what's in that black box. So I found out what's in that black box behind Agente AI framework. And I think that's really important is anytime you're presented with the hype or the buzzwords. I think what we owe ourselves as data professionals is to get behind that and really understand it. So break it into the different components and educate ourselves and make sure that we know everything about it. And with the right partners that are partnering with you, you can do that. I think the important pieces that we learned in the past year to really double down and focus on in terms of our education and our capability build. And also turning that black box into something that I know every single component and have peace and am able to establish some visibility into. There's several pieces. One is the context layer. So it came extremely important last year when we were enabling AI for BI. So conversational AI to understand what is the role of the context layer and how important context is and everything. And one of the design principles that I made sure which we've always done along our journey here in this function is how what do we need to set up in order to make the that happens. So in this case the context layer, what does that look like? And so that's really important, I think, to have a technical blueprint. So make sure from an architecture standpoint, you have the very detailed technical blueprint that can explain that to you and you can also explain it to others. And then that way again, you don't have this black box. So the context layer was a big piece. The second is the prompt engineering. So everybody talks about it, but you know it's really important to know what's going on in that prompt. So to have that full visibility and I think you have to focus in on the mechanisms to show you that and make that visible all the time. And to have the roles and capabilities within your team, do you have the right technical skills as well as functional skills to understand what's going into the prompt? Are you able to validate that what is then returned from that prompt is actually in line with what you're expecting? So I think that's the second component. And then the third is getting back to the Agente AI framework. And I'll tie it back to what was covered I think earlier in the podcast about the personas. You know, it became very clear to me that in order to get the right responses or I would say the more the suitable responses or the appropriate responses to the questions that we were asking of the conversational AI is you really need to first start off with the persona. Who is the persona that is asking the question? And then that way the Agente AI framework that you have with the orchestrator and the different agents that are brought into play with that orchestrator that it's able to focus in on the persona and deliver the appropriate responses because there's a path that has to be followed. But in order to follow the path, we have to say, okay, it's the path for this persona or that person. So those are the three things that I learned last year that are really important aspects of this. And so that has to be set up to be scalable, repeatable, dependable in order to have an AI ready data platform. Wonderful. Wonderful. In fact, the species around persona really resonated with me because it ducked into what the Ujula saying earlier that he not putting that end user into the center and designing everything around that. Which was anything to add from your site on how you guys are leveraging agents and the whole agentic paradigm. Absolutely. So you heard racist speak more about the fundamentals and the workflow designs that really happened at the back end, right? But to an end user, the end user is really looking for is the AI elevator. A lot of the end users really don't understand the fundamentals of getting the right data and having the right workflow etc. But as an end user, I think what users need to also really understand is the AI truly able to give me or shine light on questions that I myself as a user haven't thought of. Because many times when we talk about personas, many times, it's like everybody has their blinders on if I may use that as an example to say, if I'm responsible for talent acquisition as an example, I'm only going to think about talent acquisition. But the beauty of bringing in the common data model and the whole agentic framework is now saying, let's not just look at talent acquisition because talent acquisition is just one part of it. But can the orchestrator actually give me reading or lagging indicators that might be relevant for me as a persona. But I'm probably not even asking that question because I'm only looking at talent acquisition as an example. But a classic example could be turnover has direct implications on talent acquisition. But if you really look at a talent acquisition manager, they never ask questions around turnover. They're all worried about, how many new rx are coming in? What is my time to hire? How quickly can I actually fulfill those requirements? But if I get these leading indicators, and I'm calling turnover as a leading indicator because turnover leads to a demand in acquisition. So if I already know who's going to leave or what my turnover is going to look like, I can actually start prepping my TAT to say, all right, these are the roles that I would want to fill in. And really, that's where the whole AI solutions really need to come to life through the agentic framework by saying, don't look at it, we're only based on the blinders on or looking at only what's relevant for you. But there could be other aspects that are having an influence on what's relevant for you as well. And I think that's really what we are trying to bring through life here, which is to look at the entire ecosystem. Again, not just, you know, what's relevant for one persona, but then what can influence that persona as well. And that's where the whole power and the beauty of AI really comes to life. That's actually very, very profound with Jules, right? Because I think as you're speaking, I was just thinking through that it's not about having a point in time in sight, right? But looking at the persona, it's an entirety and on its adjacencies, right? And then through that, I try to find the insights at the intersections if you've well, right? I think that's actually very, very profound because I don't think people always look at the data and insights that way, right? And maybe using some of these newer constructions, they can actually stitch that together and that could be very, very powerful. Exactly, exactly, which is again, user-led, bringing the user in the center and in some ways forcing them because it can be a very uncomfortable conversation for some of those users because, you know, you're kind of asking them to think through what and how and all of that and that can be, you know, a little bit intimidating for them because, you know, they've never been in those kind of situations in the past, right? So I think that's where the whole paradigm is shifting as well. Wonderful, wonderful. A small plug from Tiger Analytics here, because we are talking so many of the personas and as our audience is also hearing, right? The personas are really at the center of realizing this kind of solves. We do have a solution called "GeroShotProfiler". The idea of this "GeroShotProfiler" is that for any persona within the organization, right? It can very quickly create a full profile of what are the kind of questions that persona need to ask. What are the kind of decisions that the person needs to make? What are the kind of data elements that that person needs to look at? What are the kind of insights that they need to kind of review on a daily, weekly, monthly basis? This is being achieved through a curation of our understanding of the marketplace and designing several solutions over time. If any of our listeners are interested in knowing more about that, they can always reach out to Tiger Analytics and we'll be happy to do a demo. Well, this was great talking to both you, OJuel and Rachel. Thanks for your time and we're looking forward to kind of collab it further. Thank you, Sachin. Thank you. Appreciate it. Thank you so much, everyone.