S2E09: Inside Walmart’s AI Journey - From Pilots to Scale | Feat. Aaron Berg

[MUSIC] >> Hello, hello, hello. Welcome back to the Retail Tales. We are going to be a podcast where we exclusively discuss all things AI across the retail value chain. And I'm your host, Sorab. Today's episode is going to be a fascinating one. Because when we talk about AI in retail, few names embody the scale, the complexity and the transformation quite like Walmart. The company pretty much operates at scale where every experiment has a ripple effect across millions of customers, thousands of stores and countless number of supply chain nodes. That makes Walmart kind of a living example of how AI can truly, truly transform retail, whether it is in logistics area or in the pricing domain or even personalized experiences. My guest today is Aaron Berg, Walmart's vice president of digital transformation. He is a leader who sits right at the intersection of technology, operations and future of retail. One has had a remarkable journey within Walmart, from corporate development to finance, to driving large scale digital initiatives that are truly redefining how the company operates globally. He has led initiatives that bridge the strategic with the operational and the analytical with the human. His vantage point gives him a front row seat, if you will, to see how AI data and human ingenuity come together to reinvent one of the world's most complex ecosystems. We're going to talk about his journey, his experience with AI so far, what excites him, and we'll also dig into the lessons that he has learned from scaling innovation at one of the largest organizations on the planet and explore what's next in retail, what's next for retail in the overall era of AI. And we'll also go through a fun lightning round towards the end to get his quick takes on myths, mindsets and maybe even moonshots in AI. So without further ado, Aaron, welcome to the podcast. Welcome to the retail tails. It's my pleasure to have you here and to have you share your perspectives with the listeners on retail AI. Yeah, thank you. I'm delighted to be here. I appreciate your partnership and I'm excited to talk about AI and retail. Love it. All right. Before we dive into our conversation in the specific topics that I have lined up for you, I want to start with your story. Every leader has those few defining moments, those pivotal moments in their career that shape how they think about innovation, how they think about risk, and especially in a company as vast and fast moving as Walmart. You've moved across functions from finance to corporate development to operations and now digital transformation. Some would say that's not a very typical path, but it's the kind that gives a leader both the numbers mindset and the transformation muscle. So why don't we start with you telling us about your professional journey, what has been, what have been some of the key inflection points that shaped how you approach digital transformation today? Yeah, absolutely. So I really think of myself as a generalist who likes numbers. That's been what the through line of my career has been about and AI makes that continue to be just a great way to build a career in retail and across a lot of other industries. My aspiration long ago in undergrad was to become a political science professor. That's what I thought I was going to do. I haven't gotten there yet. Probably won't ever get there. I think I've taken a different turn. But when I graduated, I wanted to go work for a few years before I went back to grad school. In particular, I was interested in studying the private sector in East Africa because I had a transformative experience doing a study abroad program when I was in undergrad. Very randomly, I met a businessman from Arkansas who was starting a coffee company. It turned out to be a coffee company in Rwanda. And he hired me, sort of sight unseen, to go help out. We were just starting the business. There were four or five employees. I moved to Kigali, Rwanda right after I graduated and we started building. It's now one of the larger coffee exporters in Rwanda. And coffee is a big export commodity for the Rwanda economy. I think that that kind of is the beginning of my digital transformation journey because we were standing up a new manufacturing facility in a part of the world that still ran fairly manually. And we needed to hedge our coffee production. Coffee comes as cherries. It needs to be processed and dried. And then you want to manage the risk through that process until you sell it because the coffee price can be quite volatile. We tried to do that using Excel the first year, thinking that it would be close enough. And we had a pretty big mismatch between what we thought we were going to produce and what we actually produced. And so the next year we needed an ERP system. But we were really small and the requirements in that market were a little bit specific. So we couldn't buy a commodity trading system that FedR needs and we couldn't afford to hire a consulting company to customize one. And so I built it. We started working with a tool called FileMakerPro, which is like Apple's version of Microsoft Access. And we built a version of an ERP system that worked much better. We were able to build it exactly to our needs. It was sub-scale, but they are scale. And it meant that we could iterate through it really quickly. That I think is really the thing that has set the tone for me all the way through my journey, which then led me into finance where I did some more things. We would identify a need, find a manual process, and then use sub-scale technology to automate it. And then in 2018, I got invited because at that point I had moved to Arkansas. Once you're in Arkansas, there were all kinds of connections to Walmart. I got invited to join the corporate development team. I had not anticipated and was really surprised to find myself joining a very, very, very large company, much larger than like maybe 100 people in the home office that are other company before. And I was nervous about what that might mean. But when I got to Walmart, I found that it is absolutely a place with a wonderful culture. And it encourages and allows me to exercise some of that entrepreneurial spirit, solving interesting problems at a much, much larger scale than we could ever imagine in the earlier business that I'd worked in. But a lot of the same principles apply. We have the subject matter experts. We have the technology systems. And we have the ability to build really sophisticated solutions, but it often makes sense to start with simple ones. And I believe really strongly that if you can get people in the business, people who are close to the problem to build the first version of a solution, good things happen. Sometimes that's all you need to build. More often than not, that's all you need to build. But even if it's not, it gets you to an answer that you can learn about and iterate through it quickly. And then you can scale up when and if you need to, which is so superior to saying that the problem is really hard. And so you have to take a long time and a lot of money to build the first version of a solution. So that's my career kind of focused at the beginning, but then arcing through it. But really, I'm a generalist of the likes numbers. Now, that's amazing. Look, I've known you for a few years. I did not know about your journey from coffee to AI. Yeah, let me put it that way. But it's fascinating. Thank you for sharing. Let's dive deeper into our conversation. Walmart has been one of the most active retailers in experimenting with AI, you know, from intelligent supply chain systems to AI-powered store operations. And more recently, the agent AI leadership, you know, with the concept of super agents. And now with the partnership between OpenAI and Walmart. But what's always interesting to me, it's not just the technology. It's how leaders experience AI, how they go about, you know, from exploration to conviction. Can you walk us through your own journey with AI? Like how did it move from being an interesting enough technology to something that you saw as strategically essential, you know, for retail and for the organization that you currently represent? Yeah, or I mean, so AI, you know, broadly stated, machine learning sort of statistics with a lot of computing power behind them. It's been a big part of Walmart's journey for a long time. And also, caveat that I have spent most of my time working on the end to end organization. So stores and supply chain operations. How do we serve our associates so that they can serve our customers? And so some of the really interesting things we're doing, I'm somewhat at a remove to Um. day to day, although one of the great things about the home office is everybody is accessible and we talk a lot. I'm going to take your question as sort of, what did my generative AI story feel like or generated switch intake AI, which is the way I think that everybody is trying to figure out how to ride here over the last couple of years? Sure. My my conviction was something that I think I think I really got serious about about a year ago. And I think I figured out how we might activate it within our organization or how we might sort of participate in activating AI within Walmart, maybe nine or 10 months ago. And so it took a little bit of time for me to get really excited about what generative AI I could do. And then after that, a relatively short period of time to start figuring out just how powerful it was. And I don't think that we found the peak of that yet. Like I think that we discover exciting things every week. For me and I think that this is also true for a lot of people. There was one moment where AI went from like, oh, this is really kind of neat and it can do fascinating things to like, oh my goodness, this is this may be sort of a game changer for the way that we work. And we have started to talk about that sort of the spark moment for people. And one of the things that's fun for us to do is as one of our sidelines on our team is to help people have that experience. For me, for me, by the way, is not very common though. Like that epiphany is happening with leaders on a regular basis, even like on a day to day level. Yeah. All right. Gus. So my, my, my epiphany was a notebook LM as a product that Google has put out and continued to develop about a year ago. It was an experimental, basically, toy that they had launched. And its main purpose is to let you chat with documents, but a really cool thing it does. Something that people get really excited about and I got really excited about is it will let you upload a set of documents and it will make a podcast about them. And the podcast is pretty realistic. It's two people talking. It's very specific to whatever the documents are. And it's quite compelling, you know, somewhere between five and 15 minutes. And so as an experiment, just to say, this seems like something that might be interesting to try with AI, I took a personality profile that I had from a Hogan assessment that I'd done at work. I took a copy of my LinkedIn profile and I took my sort of individual development plan and my aspirations for what I wanted to do. I loaded them into notebook LM and had it make a podcast. And when the two hosts started saying, you know, let's type deep into the world of Aaron Berg and then continued, you know, proceeded to, to sort of summarize my, my Hogan assessment fairly insightful in the context of Walmart. That was the moment where I was like, oh my gosh, this is, this is a really big thing. It helped that I had talked to a person inside Walmart who's an organizational psychologist who specializes in those sorts of interpretations about my Hogan, not that long before I was trying this. And the podcast that came out was not as good as the conversation with this expert. But it was pretty great for something that was effectively a toy that Google put put on the internet for free at that time. Yeah. And not everybody has access to those experts anywhere. And if the, if AI could get you close enough, I think it's, it's really great for people who don't have such access for free. As you said, I use notebook, LM mainly for learning and development purposes. Because there is so much information across documents, across videos, across podcasts, even to give all of that. And for it to give me like a summarized version of what I really need to know about it top and it just works out really well. Yeah. Yeah. And I think that highlights, I mean, it's both cool with the technology you can do and it's also cool how accessible the technology is and then how it's a technology and this, this came later, but it's a technology that makes other technologies accessible, right? Like the, those are things that I think will come up in our conversation as we keep talking. Yeah. No, well said. Now if you, if you zoom out and look at retail, the overall canvas for AI is extremely massive. You could look at hyper personalized shopping to autonomous supply chains to, you know, to logistical, you know, kind of eureka moments. You could look at, at in store customer experience to associate productivity. Pretty much across the retail value chain is what I'm trying to say. AI is rewriting how our industry works rather than how it must work, you know, in the age of AI. But in my experience, what's most often missing from the conversation is focus. Knowing where AI can create transformational value where just incremental efficiency. So my question to you is when you look at AI's potential across that retail value chain, where do you see the biggest and their most exciting opportunities? So I mean, I think I'm going to challenge slightly and maybe reframe what you stated, which is I think I mean, AI is a tool that has a lot of impact. But, but I don't think that AI itself is rewriting retail, at least not the part of retail that I work in. But, but it is creating a new set of capabilities, both in terms of what it can do as software and what people can do when they use it. And those people, I think, are driving really interesting innovation. And that's where the pace of change is increasing. That also is where I get most excited about what AI might do. I know that there are a lot of sort of fully autonomous or very novel kinds of use cases that can pop up. And I'm excited about them. And I think they'll come where I think the biggest impact is right now is around how AI can help personalize experiences for people who are doing work or for the people that they're doing work for. And in particular, I'm excited about how I at how AI can let the individuals who are already wrestling with that complexity create solutions for themselves and then share those solutions. So I think that what AI does better than any other technology that I've thought before is it helps individuals be more independent in how they go about using technology to solve their problems. And I think I think retail is a really interesting canvas for that to work against because especially in the channel retail, what we're doing is really complex, right? There are a lot of stakeholders or a lot of nodes. There's a lot of sort of lightly controlled or uncontrolled environments because you're trying to serve customers and also associates sort of where they are and how they are in an environment where people are interacting. And AI does a much better job that sort of mapping technology to that environment than a lot of the tools we had before which are more rigid. Yeah, no, fair enough. My, my, the assertion essentially was implicitly driving towards people using AI capabilities to transform the retail as we know it. But you put it in a, in a more accurate, more kind of direct way. So thank you for that. Let's shift gears further because for all the excitement around AI, there is a, there's a quieter reality, which not many people realize there is. The way I describe it is hidden costs of AI. AI at scale is not just about, you know, models and tools. It's about underlying data infrastructure. It's about governance. It's about training and adoption to the people point that you brought up. Integration compliance and even change management on, you know, how you used to do certain activities or, you know, workflows were executed to now how they could be executed with AI. These are the costs that don't necessarily make it in the first year. ROI estimation when you are prioritizing an AI initiative. But these are also the costs that, I mean, in my experience, they make or break long term success. I've seen far too many leaders underestimate these kind of hidden costs only to find themselves stuck in what I also describe as pilot purgatory. So from your vantage point, how do you see these hidden costs of AI and, and you know, that often get overlooked? And why do AI initiatives, dare I say, most AI initiatives, uh, fizzle away? So I think a lot of the hidden costs that you're describing, uh, are, are hidden costs that are associated with digital transformation or just transformation generally. And maybe the scale or the speed is a bit higher for AI because it, it sort of helps to, helps all of those things to move faster and, and further. Um, but I, I think that they're common to a lot of the other transformations. I think that, you know, on the other hand, there's also kind of an inertia or a cost of getting started burden that, that can sometimes factor into those kinds of projects and into AI projects. So I think, you know, it's, it's a matter of degree more than it's something that's unique to AI. Um, I, I think, I mean, again, I like data, I like numbers. And so I think having good accessible data available is, is one of the keys to, to most digital transformations, but also wanting to have, have perfectly controlled or perfect data before you get going is a major hindrance to things that could be good. Um, I think that finding that right balance is, is what helps you to manage the, is this project going to be successful or not successful? And, and for me, I think it's sort of going back into that AI helps people use technology in ways that they, they personally couldn't use it before. Um, I really get excited about AI projects. projects that can stand up to fill gaps that just couldn't be addressed previously and often at a very small scale. A lot of the risks of governance or data quality get smaller if you're talking about small things and at least in the organizations I've worked in when it's relatively expensive and slow to do digital transformation. You end up working on big projects that get good outcomes for most of the use cases but I believe a lot of edge cases behind or make it difficult to the average and those things just continue to either not be addressed or consume a lot of time as they're worked through and to me there are a lot of no-brainer early AI solutions that target those areas because you have fewer of the costs that you're describing and they're usually completely sub-optimized because they haven't been the focus previously. That's how I think about it. When you take that lens, a lot of the problems start to soften from my perspective. You have, if you talk about governance or control, a lot of the AI projects that I get excited about initially as people are going on their journey are replacing things that were manual or not happening and where the assets that were being created were Excel or PowerPoint files being shared via email. Although governance is a really important thing, there are a few systems that are less governed than that. If you're building something, even if you haven't solved all the challenges of AI at scale, you're still probably improving on what was there before. Yeah. I'm going to go off script a little bit and ask you about governance. Do you see governance as a necessary evil or a foundational capability for any organization to get on with as soon as possible, especially for their AI journey because trust is something which is unique to AI in my mind, which wasn't really top of the conversation when it came to earlier digital transformations. Yeah. I think governance is really important. I think that anything that's deployed at scale, you need to make sure trust is paramount. I think that Walmart does a great job. I think a lot of organizations are a really good job maintaining that. We've got it appropriately at this center of what we're willing to launch. I also think that what governance you need, what controls you need for something that you're going to deploy at scale in the world is very different from what you need to get going and experiment. Once you've experimented and prototyped and built the thing that you want to scale, often the specifics then of what you need to govern to scale it are a lot less daunting than trying to govern things in general before you've started. I also think that with any technology, and again, AI is a good example of technologies more generally. One of the best ways to make sure that you get to a good outcome is to make sure that as many people who are involved in those journeys as possible understand what it is they're dealing with and have the knowledge they need to help supervise and guide that process. I think that one of the other things that's important to me when you think about governance is really literacy or data literacy. How do you get business partners who maybe wouldn't have been super involved in the details of how something was implemented in the past, aware enough of how things work that they can assess what's going on. One of the best ways to do that is to teach them to build stuff that's not going to ever be launched to everybody, but it lets them see how the machine works. That opens their minds to a large extent on the capabilities itself, whereas if you just ask somebody called them out cold and ask them, "Well, what can AI do for you? They will come back to you with their current frame of reference and could be a task level activity at the end of the day." Let's talk about your hopes and fears. I've had some very experienced leaders on this podcast so far, and even beyond this podcast every leader that I've spoken with on about AI in retail, every leader carries a mix of optimism and caution. On one hand, there is an incredible promise, the promise of creating better experiences, creating that enhanced productivity or efficiency, improving decision-making, empowering employees even. On the other hand, there's a fear of bias over reliance and losing human connection that retail was built on. I want to know your hopes, your fears, and maybe even your wishes for AI in retail. What gives you most optimism when you think about AI the next decade from that lens and what keeps you up at night? Yeah, decade is a long time, especially now. I can talk about my hopes and fears for the next six months, but I think. Let's start there and get it going. You can extrapolate out, but I think one thing that we're all learning is. What you think is going to happen over the next 36 months changes about every six months now and in ways that have it started, which I think is exciting. For all of this, I'm much more an optimist than a pessimist around what AI can do. Where I think we're at a really neat moment and my hope that would lead me to continue being an optimist is I think that AI makes technology accessible, whether that's using AI to build things that aren't AI or using AI to build things that have AI embedded in them. Both of those things. I really am floored by what AI can do to help drive that. I think the other thing that I'm really excited and optimistic about is. Generative AI as it exists today, the models, the words, language models that I spend a lot of my time thinking about. They're not magical, but they do a really, really good job of summarizing stuff. The number of problems that I see across our business and I think across a lot of businesses that can be stated as, "I know what exceptions I'm looking for. I can find a lot of them, but there are too many to handle one by one. What should I do?" Right. I think a lot of problems of that class and I think that for the next little while we can use AI to just really take care of a lot of those. I think you can use AI's summarization power as well if you know how to think about it to take human knowledge and use it along with that summarization capability to create really good outcomes. That's the short term. Those are the two things that I'm excited about really soon. My hope projected farther into the future is that if you get more people or people who are closer to the customer connected more closely with the technology and the technology continued doing more accessible, then that has a very long run. You can keep doing that and you can keep following that. I think that's where AI potentially could be the most helpful and the most useful because right now leveraging technology remains a somewhat higher friction activity that I'd like it to be. I think that if you look back 10 years, that friction has been decreasing and I think that it could decrease even more and even faster. Where I get concerned, and I guess one last thing there, the concerns around cost and margin and all of that, to the extent that you can use some computing power to replace some human time. That ROI looks good. There are a lot of situations where you look at it and you're like, hey, the solution we're building is extraordinarily expensive from a computing standpoint. But it's not expensive if you look at it from a human standpoint and I could take the human time that we can save and use it in ways that are much more valuable than the cost of that time. There are all kinds of new ways to serve customers. You end up in a virtuous feedback cycle there where you're saving time on things that you'd rather not do so you can create time for things that you really would like to do. Where I get nervous and the thing that keeps me up at night, it's less around. There are, I certainly have some concerns around the runaway things. I think the philosophical implications and the potential risks of AI are interesting. What I think is the more pedestrian but much more worrisome thing is that for whatever reason concerns about AI, sort of inertia and ways of working, AI becomes a tool that isn't democratized and isn't used widely by people who are close to the customer. That failure mode to me looks like we mostly just don't move fast enough. You have this really powerful technology but you use it the same way that you use the older technologies which means that you move closer to the pace that you move for the old technologies and you don't get the value that you could unlock out of the system. I think the other kind of flavor of that is if you have specialists who are trying to develop AI solutions for non-specialists, I think the most likely outcome is that you just don't get very good outcomes because communication is hard. I think another possible outcome is the specialist in AI become good enough at the work that they're helping to replace that they actually take over that role and I think that's really bad too because we've got, I mean, there are wonderful experts all across our business and I would much rather help empower them with this new technology than help build somebody up who needs to learn the technology and what they do and then have that tension in the world. I mean, clearly you think a lot about AI in retail and I love it. That's why our conversations, I kind of and I look forward to every time. Let's do something fun. So I'm gonna introduce a new section that we have not had in the past on the podcast. It's something different, you know, than what I've done with my other guest interviews. It's a quick lightning round of questions. It's called the lightning five, five rapid-fire questions, short, sharp, spontaneous. So are you ready? Let's get into it. All right, so here we go. One myth about AI and retail, you'd like to bust once and for all. - I'd like to bust the myth that AI is not something that everybody can build with, right? I know it's something that increasingly everybody can use, but I want to bust the myth that it's something mysterious that only people who know AI or know technology can work with. So democratize it, got it. One unpopular opinion you hold about AI. - I don't, I mean, so unpopular opinion. I'd reference the other one. Like I think that there are circles we're saying that everybody should be using AI and should be building with AI is a bit controversial. I think I'll probably stick with that one 'cause the others are more just sort of philosophical speculation. So I think specifically I would say an unpopular opinion I hold that includes AI. His people should learn to write enough code to work with AI, no matter who you are, right? People should be building agents using Python, not using an aging building tool that allows you to drag and drop stuff. I really would love drag and drop tools to just go away. - Go away. Well, it is definitely an unpopular opinion. I can tell you that. - Yeah, okay. - Well then I'll own that one. - If you had, let's say, 10 million bucks to invest in AI tomorrow as an entrepreneur, where would you allocate it? - I mean, from somebody who likes to think about sort of return profiles, I think it's a very strange time. And I mean, it's hard to think about the valuations and that's not my business. - Yeah. - If I think about what would have the most impact, it would be around helping people to access these tools and really right now would look a lot like education, right? Like it would look like how do you help bring new types of problem solving and the capabilities to work with code? Maybe for the first time to a large group of people in business settings, right? Like I think it's an interesting thing in general, but I think that it gets real for people and you can get people to get over that learning curve much faster if you can go and you can say, you have a real problem that you get paid to solve. Let's work together to figure out how you might use these new technologies to do it in ways that you didn't know where possible. - Got it. Bitwors is by, where do you draw the line for AI capabilities? - So I think, I mean, my opinion is you should buy the models, right? Unless you have an extraordinary amount of money, a lot more than $10 million, you shouldn't be trying to build your own AI models. And right now it seems like frequently it's not, you shouldn't prioritize fine tuning them. - Yeah, after that, I think you should mostly build, like you should build using tools from other people. You frameworks are wonderful, SDKs are wonderful. I've been really enjoying building with the agent development kit, which is a framework that Google open source to a while ago. But I think that once you've got kind of the base model available to you, you should build. - Amazing. And then the last one, which retail function or domain do you think will be the last to truly feel AI's impact and why? - I think every, I don't think that there is gonna be, I don't think that there's gonna be a meaningful, interesting part of the business that's last, because I think everybody's already feeling the impact. I'm trying to think, and I really think the impact might be quite significant across many parts of the business. I mean, you'd have to look for what parts of the business aren't digitally enabled today, because those would be the parts where it would be hardest to bring AI directly to bear, although large language models even make that less of a barrier than it was before. And one of the things that's really wonderful about how Walmart has been approaching technology for most of its history is a lot of our system is digitally enabled. And I mean, we've got pretty much everybody at the company using sophisticated tools and we're building tools to help them power them to do their job. So I don't have a good answer for that one. - No, that's fair. It was lightning. It was quick. - Yeah. - We're coming to the close of this episode. But before we wrap, I wanna end where many of our listeners are right now, either at the starting line of their AI journey or maybe in the initial starting phase. And this phase, they're convinced about AI and they're convinced about AI matters. They've maybe done a few pilots, but they're still trying to figure out how to move from experimentation to enterprise scale adoption. What's your advice to retail leaders who are in this phase of their AI journey? - I would recommend thinking smaller and more. And I also would recommend sort of thinking about the people who could learn those capabilities. I think that, again, this is, maybe this is an impact of our opinion, but it's definitely a very personal opinion. I think that a lot of leaders are approaching AI as something closer to a capital investment project where you need to build a very detailed business case and you need to expect a fairly large investment and a fairly large return to make it worth pursuing. While I think there are lots of cases where that sort of cost benefit exists for AI, if you're just getting going, I would start with something that's low cost and low risk and it doesn't need to have an extraordinary return to create a really good ROI plus you'll learn, right? And your associates will learn. And so my advice would be figure out who might be your early adopters and then give them the tools and challenge them to solve any problem they want with AI without putting a lot of guardrails or barriers around the need to scale, the need to have tightly-governed it. Again, I'm not suggesting that you would take that solution and scale it without dealing with those things, but sort of lean into where the value is and look for places where maybe those things look and feel like pilots, but they actually pencil. I'd go back all the way to the beginning, like there's lots of software that if people can build it for themselves or for their peers, you don't need to scale it up all the way. It doesn't need to be something much different than a pilot to be working at scale and delivering it to value. That's amazing. We did talk about the hidden costs of AI, but the point that you're bringing up about the hidden value of AI as well, which is learning within the organization sometimes or most of the times goes, you know, kind of under-noticed, but that's incredible. And this has been an amazingly insightful conversation, at least for me. I think our listeners will really appreciate how candid and practical your perspectives are. And, you know, especially around the balance between ambition and responsibility when it comes to AI. Thank you for joining us on the retail tails and for sharing your experience. Yeah, that's my pleasure. Thank you for having me. And to the listeners, if you enjoyed this episode, don't forget to follow the retail tails on Spotify and Apple podcasts and connect with us on LinkedIn for more conversations at the intersection of retail and AI. Until next time, stay safe, stay healthy and stay curious.