IBM Think 2025 | Magesh Bagavathi, PepsiCo & Shobhit Varshney, IBM

Clips
More from IBM Think 2025

Magesh Bagavathi

SVP, Global Head of Data, Analytics & AI

PepsiCo

Shobhit Varshney

VP & Senior Partner, Head of Data, AI & Automation

IBM

play_circle_outline Pepsi as a data and AI company serving consumers with products

play_circle_outline Challenges with handling 60 petabytes of data growing at 2X annually

play_circle_outline Transition from traditional to modern data architecture for real-time processes

play_circle_outline Governance and trust crucial for AI and data accuracy

play_circle_outline Value unlocks seen in Gatorade AI campaign, personalized marketing, and digital twin

play_circle_outline Democratizing innovation through central platform for widespread impact

Info
Transcript

Magesh Bagavathi, PepsiCo & Shobhit Varshney, IBM

Magesh Bagavathi

SVP, Global Head of Data, Analytics & AI PepsiCo

Shobhit Varshney

VP & Senior Partner, Head of Data, AI & Automation IBM

Magesh Bagavathi, senior vice president, global head of data, analytics and artificial intelligence at PepsiCo Inc., and Shobhit Varshney, head of data and artificial intelligence, vice president and senior partner at IBM Corp., join theCUBE’s Dave Vellante at IBMThink to explore how data and AI are reshaping business strategy and operations. The session highlights the unique approaches both PepsiCo and IBM are taking to optimize supply chains, enhance customer engagement and drive innovation at scale.

Bagavathi discusses PepsiCo’s significant invest... Read more

explore Keep Exploring

What does Pepsi consider itself as a company and how do they use data and AI in their business operations? add

What are some insights into the use of enterprise data in AI models and the management of digital employees within companies? add

What is the current state of data architecture for PepsiCo and what changes are being made to improve it? add

What is Pepsi doing incredibly well in terms of governance around data and AI? add

What are some of the results and benefits of the Gatorade AI campaign and bottle design capability that was launched almost six and a half months ago? add

What approach has Pepsi taken towards addressing innovation? add

bolt Powered by CUBE AI

Magesh Bagavathi, PepsiCo & Shobhit Varshney, IBM

search

>> Hi, everybody. Welcome to theCUBE's special coverage of IBM Think 2025. We're here in Boston inside the analyst room, and I'm super excited to have Magesh here who's the Chief Data and AI Officer at Pepsi, this is going to be an amazing case study, and he's joined by Shobhit from IBM who's Head of Data and AI at IBM Consulting. Shobhit, good to see you again.

>> Thanks a lot.

>> Magesh, great to meet you. First time in theCUBE. Thank you.

>> Fantastic, Dave, thanks for having me as well. Thanks for theCUBE as well for having us at PepsiCo.

>> So Pepsi, Shobhit was saying, "Look, think of Pepsi as a data and AI company." I'm like, "Well, wait a minute, hold on a second. What are you talking about?" But he's right, it just happened to be in a business that serves consumers with some amazing products. So how do you think about Pepsi?

>> We are a fantastic consumer company. On a daily basis, we serve 1.4 billion smiles, which means there is 1.4 billion consumers and consumer products on a daily basis. We are a company that brings smiles to all consumers and, oh, by the way, data AI enriches how we get those smiles out to consumers on a daily basis. That's how we think about us. Now, we've been in this space from a data enrichment, and even what we consider as traditional AI for a little more than a decade. It's not something that's new to us. For us, it's a continuation of this journey.

>> I was just talking to some friends from Heineken and they were basically talking about their latest data journey, and it started in 2019. I'm like, "Oh, okay. That was before the AI heard around the world." So technology comes and technology goes, as you know. As you point out, you were on this journey well before the whole ChatGPT moment, so I'm sure you faced many data challenges. My understanding is you've got 60 petabytes of data, is that right? Growing at 2X every year.

>> Every year, yeah.

>> Okay. So irrespective of whether it was machine learning or now gen AI, you had a data challenge initially. Maybe we can go back and think about and share the audience how you were approaching that data challenge. What were the drivers that were leading you to where you are today?

>> See, you've got to start with an end in mind. What are you really trying to do? And what you're trying to do is you're trying to drive high amount of products or the right products for the consumers for the right occasions, and you want to make sure those products are available to them. That's one. What you also want to do is you want to drive an efficient, lean, flexible supply chain. That's two. Third is you want to make sure that from your transportation logistics perspective, we've got the second-largest fleet in the United States. We're making sure that is driving those products and those things, those delight our customers over to them very quickly. So how do you make all of this possible continuously on a daily basis? That's what we try to do. When you go back to how we're set up, what we call is within the technology landscape, we've got 2,400 applications, and on top of that, we've got a huge OT landscape, operational technology landscape that is our manufacturing, our warehousing, and our shop floors. How do you get your data from all of these applications that's humming and running what we call as make, move, sell, and close on a daily basis out to your data lake, out to the cloud, into those applications that are applications of insights and applications of decision-making that actually power the front lines of our organization. That ability for us to move this back in data to the front end of the business and then drive decision-making, the front end is where its art and science is really for us.

>> My question to you is, so how unique is this? Obviously, the scale is unique. Are there patterns in other industries or similar industries that you see that you can share with us?

>> Absolutely. Data is gold for you. We quite often engage with C-suites looking at data as a corporate asset, and how do you stop thinking about monetizing that for competitive advantage, right? Within IBM, we are very strongly believe that enterprise data has not made its way inside of the AI models quite yet. 99% as Arvind mentioned.

>> We heard that from Arvind today, yeah....

>> have not gotten to AI models yet, so I think that's the competitive advantage that you would get. It's like hiring somebody from the outside. Everybody has access to great talent. We're here in Boston, see MIT, Harvard has some amazing talent and you would expect that that person was coming in from the outside has the world's knowledge across all the different, they could have PhDs in Maths and whatnot, right? But when they come and join PepsiCo, they become PepsiCo employee. You go through an onboarding process, you explain to them how things are done in the Pepsi way, you give them access to a lot of different data sets back in control that, and then over time you fully monitor them for how they're performing and you give them the right training and coaching and that's the journey every employee has. Now, when you start to think about this as a digital employee, an AI employee, you're going through the same. You're procuring agents, you're trying to make sure you embed the processes and the enterprise data inside of them, and then you continuously monitor and do performance reviews for them, and then you eventually exit them out of the company. The whole life cycle of agents and employees, all of that requires a lot of data. You need to explain to them how things uniquely happen inside of the organization. We see this across most of the Fortune 100, 200 companies that we work with, and a lot of clients are the size of Pepsi.

>> You've been on this journey for a while. I'm interested in, let's cut to the chase here with AI. The state-of-the-art prior to the end of 2022, early 2023 was you put data in a cloud data warehouse, separate compute from storage, break down the silos. Maybe you put it all in one place, maybe you federate the data, but generally speaking, it's a historical system of analytical record. Now, you're moving to a world that is increasingly, you mentioned your supply chain. We're moving from a world that's pretty much static processes in microservices to one that is this real-time representation of an enterprise where processes maybe not hard coded, maybe more malleable, and of course at the root of that is data. Now we may not be there yet, maybe it's a little bit futuristic, but I think that's the north star that you're heading toward. I wonder if you could talk about the data stack. How do you architect that such that you can accommodate such a radical change in the way applications are being developed?

>> Really good question and radical change on the front lines for our business requires radical change on how we re-architect and the moment you asked about architecture warmed my heart. I love architecture. Let me talk about as-is or where we've come from and where we're going to.

>> Awesome.

>> Where we're coming from is a legacy data warehouse, eight different data lakes that we know of, and I would call us data puddles all over the place. Now, PepsiCo of course, is a big data company. However, that's the landscape we're coming from. Where we're going to, significant investment really one data lake completely chosen, a modern data lake platform and capability medallion architecture in terms of how you refine the data from landed bronze, silver, gold, and then how do you make sure that the gold data is truly pristine to be used for your analytical and your AI apps. But in that framework, we also have governance, cataloging, lineage management, and the ability for us to really synthesize the relationships. Think about the world where I have sustainability data. So now I'm looking at a world where I have sustainability KPIs, what we call as golden KPIs, golden reports and golden data. One set for the entire company, not different formulas, not different KPIs that's used by different parts of the company. Remember, we are 320,000 people strong. We love our federation, we love our independence, we love to operate in 200 countries the way we won't operate in 200 countries. The ability for PepsiCo to go through this process where we are refining the data sets, providing these golden reports, golden KPIs, golden data sets for people to consume still in a federated independent manner makes massive, massive difference. It's not only about this data pipeline, but it's also the manners in which you consume this data. Let me also talk about the consumption piece as well. We've got Power BI, Tableau, primary source of reporting. We're going from 27,000 reports to what we would consider as 50 AI for BI consoles in the next two years. Radical, radical shift. You want to make sure that AI for BI is operating in what I call as freedom within a framework. You want to give freedom to everybody that they can ask any questions they want to self-serve, but within a framework, the problem is in the past we've not put frameworks, we've just let people just go off and create the reports. In that process, we see 80 to 90% of the reports having almost zero to no usage. And then you just create the reporting sprawl, application sprawl. That's how tech debt gets created, so you've got to have freedom within a framework.

>> It also, what you're describing, it eliminates when you in a meeting you say, "Well, where'd this data come from? I have different data." And you spend the first half hour of the meeting-

>> Just arguing on the data, whose source.

>> Oftentimes, the meeting ends and you say, "All right, we got to get the data right, then we meet again." When you see this challenge.

>> If you look at the whole spectrum that Magesh laid out, you have a lot of data that was historically on-prem.

>> Yeah.

>> We need to prepare data for the AI, but it's also the yin-yang. You can leverage AI in helping with your data transformations as well. So within IBM consulting for example, we've built this whole data, modern data accelerators that help us understand, do discovery of what data is on-prem. And today, you have about 42% of the data on-prem has been successfully moved over to the cloud-

>> Harmonized catalog.

>> Harmonized, yeah.

>> Lineage managed, 42. And then expectations by end of next year will be 90%.

>> 90%, right? The whole transformation that's happening, we are leveraging a series of some automation, some gen AI across the park so you create a whole automated pipeline that accelerates the time to value. So the data's then ready for consumption downstream for more AI models and so on and so forth. There's a lot that goes in with this whole transformation. If you were doing this two years back, we wouldn't have all the tools to accelerate this journey.

>> Right.

>> But in today's world, we have the right set of tools. And then when you start looking at how do you figure out the different levels of what is gold and what KPIs matter, you start looking at a whole AI strategy. Pepsi has done an incredible job at building a true north of what are the big bets that we're going to go after. 100-plus POCs like every other major company would've done. But they're very, very focused on here are the three, four different value unlocks. That'll be an insane amount of value to be delivered. And if you reverse engineer from that, what kind of data sets is needed to drive the decision to do that? What's the supporting data sets needed and so on and forth.

>> But it sounds like you've done, the key is a key. You've done the hard work to harmonize that data, which is non-trivial. How difficult was that? Was it a lot of human labor, data pipeline, mechanics? Were you able to use automation to streamline that? How did you get from the as-is to where you're going in the to be?

>> It's in a large company like Pepsi, go.

>> Very large.

>> First thing you need is, one, you need the vision, and then you need the political will. And then to a large extent, you need top-down mandate.

>> Right.

>> And in this case, top-down mandate is really from our CEO Ramon Laguarta, and I report to the Chief Strategy and Transformation Officer, her name is Athina Kanioura. That level of top-down mandate/vision in the organization is critical, but of course things that really greases kits is money. So then once you have the vision, once you have the top-down mandate, the political will in the organization. The value story, we are now prioritizing across 28 key initiatives in the company. How do you really go after the value story for these 28? But then how do you also de-prioritize the things that we think is being operated subpar? And thereby now, if you really want to make yourself a data and AI company, you have focused attention on the priorities, you put money behind it, and in this case you have one foundation, what we call as an EDF, enterprise data foundation. We've got 2,000 plus amazing people and partners.

>> Thank you.

>> Shobhit and others really helping us with this data foundation. Then you start seeing the value unlock, right? So for a period of time, is it painful? Absolutely, yes, because you are telling no to a lot of people and you're moving everybody in this one path. But the moment people start moving this one path, you start seeing 1, 2, 3 wins. For example, integrated business planning, IBP, we started seeing wins in Turkey, in France. Mosaic, which is really a financial cockpit where we do cash flow forecasting. We started seeing wins in many markets. Automatically, we started seeing the population and the convergence starting to happen, right? You've got to see wins.

>> Penguins off the iceberg, first one goes.

>> Yes.

>> Dave, I'll just highlight one other thing that I think Pepsi is doing incredibly well is all the governance around this, both on the data side and the AI side. If you're doing AI for BI, you lose your end audience. The trust that they're building if you get one data point incorrect, so if you think about all the data is in one EDF now and you are asking national language questions going from national language to a SQL statement, to then fetching the data and bringing it back. The incredible thing here is Pepsi has years of data of usage patterns and how humans that were writing these SQL statements, how do they optimize it? For example, if you have a seller in Texas and says, "Hey, I want to have sales for, give me me sales for Paris," we know that the context is that the seller has access to only Texas territory, hence we're talking about Paris, Texas. We ain't talking about Paris in France, right? Bringing in that context to improve the accuracy of data is amazing.

>> Harmonized data.

>> Yes, so you can unlock those kinds of things that a generic model can't. So they've done a really good job at the data catalog leveraging the past experience of how humans have done this in the past and then driving that AI to be as you start to reduce those thousands of reports.

>> So many questions, but let's talk about the value north star. If we can convene in say 12 or 18 months, what are those tangible business outcomes that would let you know that your program at PepsiCo was working as intended? Those KPIs that you talked about.

>> You're right. I work for the 120,000 men and women who are on the front lines every day, right? That's my job, that's my mission. And why? Because they're out there in the field selling to our customers, engaging with the consumers, and really operating supply chain at scale and then doing transportation logistics to get the products to the right place so we work for them. When you talk about value unlocks, I'll give you a few examples. We launched almost six and a half months back the Gatorade AI campaign and also on gatorade.com. We've launched the whole capability associated with you can actually go design your own bottles. And by the way, it's a billion plus.

>> Show me the demo, it's amazing.

>> I wish I got the bottle, but it's really cool bottle, right? The unbelievable reception we've seen for that because we're one of the first in the market to show the fact that it's possible, but it's possible at scale and it's possible for consumers and we can actually drive top line revenue out of it. Think about it, we're seeing 165% uplift in terms of sales. We're seeing a doubling in terms of brand loyalty and Gatorade loyalty. We're seeing 113,000 unique designs that have been created as a part of this process. And then we start seeing one very important thing, athletes who are the core of the brand. All of us are athletes, by the way.

>> Tech athletes.

>> Tech athletes, but I'm sure we all do our own thing. I do yoga, I'm sure whatever, right? We all want to now engage with the brand in a very differentiated way, but also in a very personalized way. That's just one example. And we're doing the same thing in spaces like marketing, in supply chain, where we are now really unlocking digital twin. We talked about in the IBM conference where procurement's a big unlock in terms of what IBM's done.

>> Massive, yes.

>> We're now looking at what does the procurement AI brain look like for us? And then how do we unlock value from there, but actually having all of our buyers and our suppliers engage in one space. These are all huge unlocks that can drive not only productivity, as has been talked about quite a bit, but also top line growth as well for the company.

>> Company. And these are part of the 28 prioritized use cases.

>> Correct.

>> That sounds like a combination of obviously productivity value unlocks, but also revenue generation.

>> Yes.

>> Brand loyalty, real value drivers for the organization.

>> And Dave, the way Pepsi has attacked this problem, if you centralize a platform, then you can democratize innovation everywhere. We don't use the terms center of excellence because that gives you the impression that is only excellence happens only in the center. Pepsi thinks of it as the AI acceleration team because you're enabling everybody else in the organization. And if you do this on a good platform that scales cases like gate rate, right? If you're using a platform, you're able to govern and audit and control and drive security. You can imagine somebody creating an image on a website, it could go viral if it's not for the wrong reasons and stuff. So you have to have all these layers of protection. Now, here's the kicker, right? If you are a marketing team, a legal team, an R&D team, if all of them start to go work with 1,000 different vendors and it's a wild, wild west, then we are not able to guarantee the brand loyalty and what the brand stands for. But if you're leveraging explorators, in this case, Pepsi has this amazing platform that they've built. If you're leveraging the platform, your time to value, you come up with an idea and you're able to get to market before Christmas because you have all those checks and balances already baked into the platform itself.

>> Yeah, governance at scale is critical. We can't repeat what happened with the big data days where everybody was off doing their own thing and then people realize, well, none of this is governed and/or we're going to bolt governance on as an afterthought and the security, it just doesn't work. It doesn't scale. So you had to design that in upfront, didn't you?

>> Absolutely. The good thing is that this is a team sport and everybody, I've got a lot of amazing partners in the company. So from InfoSec to infrastructure operations to cloud team, to the data analytics AI team, to the entire application team, it's a team sport, plus what we call as transformational leaders, right? Leaders from running finance transformation, supply chain transformation, commercial consumer transformation, HR transformation. This is a team sport, and that's how we've come together as really one team to drive this. But the advantage that the team has is it's also how do we drive it for consumption as one team, but how do we also drive it as one team with centralized platform capabilities? That's data platforms, AI platforms, cloud platforms, all operating at scale. IBM's been a huge partner for us-

>> Thank you so much....

>> as we've expanded these kind of platform capabilities.

>> What organizational deltas have you had to either architect or address? I was interviewing Marc Benioff the other day and he had just written an article in The Wall Street Journal saying, "I'm the last generation of CEOs that will be managing human only employees and workers because we're going to be hundreds of thousands of agents." And so you begin to think about, okay, well how does that change how I actually manage the organization? What have you learned from this journey so far in terms of just organizational structures, skills, management changes that you've had to deploy?

>> I'll start off by sharing first is let's talk about operating model. We're very fortunate that we're now moving to an operating model that we think should be the operating model of the future. Athina Kanioura, our Chief Strategy and Transformation Officer really start moving into this model where it's what we call a capability teams, transformation teams, business teams, all under one roof. So now, you can prioritize, you can focus, you can drive outcomes. And it's not like I was in IT, let me find out what marketing wants to do and then we'll play ping pong and figure out how we get the right finances. It is a huge challenge. If you've got all these groups in one organization, you can prioritize, you can focus, and you can get to direct execution because you know where the values come from. That I think is one huge unlock. It's very, very difficult so I got to give a lot of kudos to the leadership team for really moving us in this direction. Very painful, but that's what we've done the last year and a half. Second piece, if I would call it right, is how do you move end-to-end, PepsiCo end-to-end processes? PepsiCo is a relationship-based company. We love our consumers, we love the bodega store down the street. We want to create that heartfelt relationship, that's how we are as well internally operating within the company. But now, how do you still get end-to-end process working in the company? So for example, source to pay, that's hard because you have to say no to a lot of people that been used to doing the way they've been used to doing in respect to markets. So saying no and then really moving everybody into centralized process streamlining, very difficult. That's the journey that we are doing and we are choosing the top six, seven processes that we want to go and automate end to end. Third is how do you really start moving into a persona-centric approach? This is something that I think across the board, most enterprises have not done a great job. Our organizational tools and capability shows up on the front line. No. What is needed for the front line needs to be how you create capabilities, right, in your organization, how your organization should be or reorganized as well. That's not how we do it. So persona-centric approach is something that we are not fundamentally rethinking because in an AI for AI world, personas matter. The warehouse operator, how does the day in the life for him matter? Someone who is selling Gatorade in the field, what's the day in the life and all the tools, all the capabilities, all the processes should be in service of him or her and what they do on their job. So those are I would say the big analogs. The last one, I would say the fourth one is really a digital backbone trying to get to this centralized, governed, managed digital backbone that becomes a platform that everyone can build on and use. We've had to say a lot of nos before, now we're starting to say, "Oh, that makes a lot of sense."

>> And we have the backbone to do it, create these personalized workflows.

>> We are seeing the same exact thing in our own business too. In IBM Consulting, we are 160,000 people globally, and each one of them is a very different persona. If I'm working with a large bank, they will have different kind of roles that we're staffing, versus in healthcare, and they have different roles from marketing to procurement to finance to software development, whatnot. We have this 160,000 people who want to do their job every day and they're looking for acceleration in the way they deliver. So we are on this mission of creating 10X agents to help current 60,000. That's 1.6 million agents that we're on a path to create, right? I look at myself as the leader who's growing in a world where for every employee, team member, we would have 10X more assistants and agents. And the aha moment here is when we ask one of our team members to go do finance procurement for our client, we are expecting that that person will have the assistants and agents that are tailored to the way they do their day-to-day work. Today, we are not trying to force a new method of doing it. You can't have a set of brilliant kids go overseas and go look at how things are done and five days later tell you how to do your job better, does not scale, right? We need to get to a point where the end user has a particular way how they trust because they're held accountable for the outcome they're generating. They have built a trust mechanism of saying, "When I'm asked to do X, here's how I get to it and I know that I'm validating it and the answer's correct." So where we are infusing surgically AI into it, that has to play to the strengths of that particular individual. Empowering the individual in their particular workflow, the way they do things, AI should be asked invisible as possible to then 10X accelerate that.

>> Okay, last question. Let's say we're at Think 2026, year from now, what do you want to be able to say? I guess that you cannot say today?

>> Truly AI at scale.

>> Right.

>> What we mean by that is all of the business use cases now really, really, really bearing fruit, right? We've invested, we've planted the seeds, we've conditioned the soil, and we are now the plants are starting to grow And then they start flowering, and then hopefully at that point in time we can start smelling the roses. Across different parts of our business, whether it's make, move, sell, close, we want to tell stories, but we want to tell big stories, huge value-added stories. That's told by the 120,000 men and women who are in the front lines of our business, and they're seeing the outcomes for themselves. That one, AI is not scary, but it's very real. Two is we've democratized knowledge, we've democratized AI and then now the warehouse AI supervisor of the future or the sales rep or AI sales rep of the future, they're now having something very, very powerful in their pocket that is now helping them drive and curate their entire day. Think about somebody who's a sales rep starting their morning, start their mornings very early, 4:00 AM in the morning, they get home very late. They're traversing multiple different routes, going to multiple different stores. It's a challenging job, and then they have to take these products and put them on the shelf. Rack and stack, rack and stack, rack and stack. For them to be able to now go back home two hours quicker so they could see the loved ones spend time with their family, that I think is a very positive side of what we could really be unlocking before.

>> That's the best KPI I've heard all week. Shobhit, your thoughts?

>> On, I'm on a very similar note. Magesh and I share this quite a bit, that both of us are very focused on people and lives and smiles and what it means to people. Mike, I would love to come back where everybody has an Iron Man suit on where you have these superpowers where it's just amplifying what you're really good at. You want to get to a point where you are doing things 10X faster and you're unlocking time in your day so you can go focus more on families and be able to get the smiles in the evening back. I think we are getting to that point.

>> Gentlemen, thank you both for spending some time here. Think 2025, really appreciate you coming to theCUBE.

>> Thank you, Dave.

>> Thanks so much.

>> Really appreciate having us.

>> All right, and thank you for watching.

>> Thank you.

>> Keep it right there. We are back right after this short break from theCUBE at Think 2025. We're going to move over to theCUBE Studio. This is Dave Vellante, right back.