#198 How Walmart Leverages Data & AI with Swati Kirti, Sr Director of Data Science at Walmart

Hi, I'm Maggie, VP of Curriculum at datacamp, and I know firsthand how overwhelming it can be to keep up with the rapid changes in data and AI. But staying ahead doesn't have to be hard. That's why our curriculum team has curated the best learning content to help you and your team stay ahead, covering everything from data and AI literacy for busy executives to cutting edge generative AI for developers, and everything in between. Whether you're upskilling yourself or your team, we make learning practical, engaging and flexible. It just works. Power your data and AI transformation journey on datacamp.com.

From a technology perspective, generative AI definitely is the big one where it's really disrupting the way people are shopping, how people are looking for products. Basically, people are no more looking for a specific item. They want to have more of an idea or intent converted into products that are being surfaced to them.

Welcome to Data Framed. This is Ritchie. Data and AI impact every industry and today we're turning our attention to retail. I want to know all about the different use cases for data and AI to get a sense of how they can be used to improve profits and the customer experience. Since I work for a primarily software based company, I occasionally get a bit of hardware envy, so I also want to dig into how these technologies can be used in the supply chain for moving physical goods around on dataframe. We like to go big, so today I have a guest from one of the world's largest retailers. Swati Kirti is a Senior Director of Data Science at Walmart where she leads machine learning and AI efforts for Walmart Global Tech, a technology incubator. Swattie has built and now runs a department of data scientists, engineers and product managers who create AI solutions to improve customer experience and supply chain. She was previously a data scientist at FedEx. Since Walmart is such a huge company, I'll be sure to ask her about scaling data and AI capabilities too. Let's hear her thoughts.

Hi Swati, thank you for joining me on this show.

Thank you, thank you for having me. It's nice to be here.

Brilliant. So can you just tell me what are the main use cases of data and AI across Walmart?

Yes, I think data and AI play a very important role in the Walmart ecosystem. We have all sorts of use cases around AI, such as we use it for merchandising decisions, we use it for supply chain decisions and how to improve our customer experiences. So anything around how we can provide better automation, better insights to our business stakeholders, or how we can enable better and more frictionless User experience to our customers. Whether they're in stores, in clubs or online everywhere, AI and data play a very important role.

That's brilliant. So it sounds like it's used almost everywhere. That's pretty amazing. Now, since you're a senior director of Data Science, can you just tell me a bit about the department that you run and talk about what are the main goals?

Sure, sure. So I lead a team, a centralized team of data scientists. And, and our main charter is to drive AIML product development to power our international businesses. So Walmart has business in countries like Canada, Mexico, Chile, et cetera. So we as a data science team, we build solutions which are very catered to these markets, we bring our platform solutions to them, but then also we build very unique to the market, very important to the business in the market. How we are able to enable AI for those kind of use cases is all done through my team.

Okay, and what's your team consists of like who's actually in your team and what are the different roles who are using data and AI?

We are a team of data scientists and engineers. Data engineers. So we do end to end product development of AI products, primarily data scientists, data engineers, machine learning engineers are the key competencies that I have in my team.

Okay, and how does your team interact with the rest of the business? So who are the data scientists interacting with in other teams?

So we work in close collaborations with product managers, our application engineering teams, business users and UX designers, etc. So we basically work in a very four in the box model where we are constantly interacting and collaborating with different leaders in different areas of the business and technology so that we are able to incubate the right ideas, do the right POCs, iterate early on and then build a product which is of course AI enabled, but then very attuned to what business case it's going to solve for and who is going to get the real value out of it.

Okay, it seems good that the data team's got a good connection with the business leaders as well. Just to make sure that the data science does have some sort of business impact.

Yeah, and like I said, everything that we build is very geared towards how we are enabling value for our associates in different roles, our business partners, or it's solving for our customer experience, making it more frictionless and more satisfying for our customers. Shopping here.

Okay, I'd love to get into more detail around that about how you improve the customer experience. So can you just tell me how does data get used to improve the customer experience?

So there are A couple of different ways. I would say, for example, on the online channel on our websites, definitely how we curate the information for our customers based on what they're looking for or is of utmost importance. We are able to personalize their shopping experience on our websites, but then we are also able to help them have a complete experience in terms of, hey, if they're leaving our website and they have missed something that they usually buy, we're able to offer them that, hey, before you go, you might want to add these things to your cart. So such kind of different kinds of personalized shopping experiences is enabled through data and AI on our online channel. But then also in store, we want every customer to have the best experience there. We want to ensure that they are able to find the right products at the right time and the right place where they're looking for it. Because stores can be very big. So all that is very data driven and very applicable to the use cases of AI on how we are improving those customer journeys.

That's really interesting, the idea of having personalization and then making sure that people can find the right products as quickly as they can. But it also seems tricky to measure. Like a lot of the customer experience feels qualitative. So can you just talk me through a bit on how you might measure quantitatively these experience metrics?

Yes, yes, it's both qualitative and quantitative. Wherever possible, we try to be as quantitative as possible. But of course there is always that qualitative element to every shopping experience. Some of, I would say the metrics, well, there are various, depending on what types of journeys we are supporting in that value chain. Right. Like there could be various touch points, but then some of the usual ones definitely are things like NPS and csat, which helps us truly understand how satisfied our customers are or how our net promoter scores are doing at different touch points in the customer journey. But then there are others as well, like conversion metrics, where we are constantly looking at how complete the transactions are so that we understand from that that if a person came to shop at our store, did they find what they were looking for and if they were able to make buying decisions based out of that. So those are very indicative of both qualitatively and quantitatively on what are some of the things going well and where there are challenges and opportunities that we should be thinking about improving on.

I'm wondering if there's a difference between online shopping and in store shopping. Then some of these things, it feels like online you can sort of see did someone buy something or not. But in store it's like, how do you know whether they found the right things or not?

Yeah, so that's where the different types of surveys and qualitative aspects come in, where we ask for customer inputs on how was their shopping experience and how would they rate them and things like that. It's definitely more challenging in stores. The brick side of the business is definitely more attuned to not having very precise data like online, but then it definitely is very directional in where to focus.

The other thing you mentioned around customer experience was personalization and it sounds like this is an area where generative AI might be able to help. Can you just talk me through what sort of things that you're doing to personalize the shopping experience?

I think generative AI is definitely playing a very important and crucial role. And I don't know if the viewers know, but Walmart was one of the first retailers to actually launch the gen AI search capability on the US website a few months back. And through that we are able to now offer intent or context based searches for our customers under certain categories. So yes, I think that's very much in line with how personalization will work over time. Like more and more personalization will happen through intent rather than through curated, very specific customer journeys. And generative AI is enabling us to do that at scale, scale. It's helping us find different types of use cases around how to not only support customers in what they're looking for to find the right products, but then also offer them recommendations or offer them different kinds of complementary products that users might be looking for, but they might not really be thinking very actively about it. So our generative AI, we're able to surface those recommendations and, and again enrich their personalization journey as well.

Okay. Yeah, I can certainly imagine that as the number of products you have grows to like hundreds of thousands or millions, then search comes really, really hard and that's where you need all the sort of technological assistance.

You should try out our generative AI search on our app and you can find a lot of things on camping, on birthday parties and whatnot without doing a lot of search and you know, going through pages after pages I think, and like that.

Okay, yeah, that's certainly a miserable shopping experience where you like the thing you want is buried on page five. All right, so the other thing you mentioned right at the start was that you make use of data for the supply chain and I have to say obviously you're dealing with a lot of physical goods and the logistics gotta be pretty mind boggling here. Yeah, can you talk me through how data and AI are used on the supply chain side?

I think supply chain is one area where we are really using AI and data in the best way possible to for a couple of use cases, to give an example, like of course inventory planning and management is an important one and demand forecasting is a very important component. So how do we use different kinds of signals to kind of understand the internal and external factors to then make plans on our how our inventory should look like, where it should be placed and the entire supply chain that goes around enabling that right from last mile to middle mile to first mile. So all that is definitely a place where data and how it flows through at an item level, at a geographical level plays a very important role. The other thing I would say is also on and where Walmart is playing a very important role is how we are helping to optimize the supply chain to how we are packing our trailers, how we are using the most optimal route to get from place one to B, et cetera. And actually Walmart just few days back again launched a product called Load Up Optimizer, which is basically a SaaS based product which any business can use now to optimize their middle mile for best route planning. So I think those are some of the use cases where Walmart is really doubling down to have the right product at the right time, but then also do it in a way which is optimal in terms of our supply chain strategies.

That's really interesting. I suppose I was thinking of supply chains mostly about like, well, how do you do warehouses? Right. But actually just figuring out what order do you put the boxes in the truck in order to pack more of them in, that's going to have a huge impact. So yeah, lots of subtleties there.

And warehouse is also an important one. I mean that's like I said, there are many different use cases where everything is part of the puzzle, like everything needs to be done right to get the right experience for the customers.

Okay, so if you've got lots of different components there, I imagine you've got lots of like, lots of different data sources all over the place in different parts of this. Does that give you like a data integration challenge? Like how do you manage all the different steps of this together?

Yes, definitely. Data is challenging and I think working for international business where we are not only supporting one market, but many different markets such as Canada and Mexico, there are challenges around how data resides in different sources, how it's generated and then how it's captured. But then over time we are building robust data lake strategies where we are able to bring most of the data, if not all, into our data lakes and we are able to then build AI solutions on top of it. Now this is all done in a very, very scientific way in terms of not just ensuring the data is the integrity of the data, but then how we follow all the policies around data, how we keep it are very transparent and explainable at all times.

Okay, so a few things to manage there, like who else access the data, is it high quality and is it following policy? A lot of governance issues there, I'm sure. So the other aspect of supply chain that I always find fascinating is around forecasting demand. It seems like a lot of the optimization comes from just having the right number of any given products in your warehouse or in the right place at the right time. Can you talk me through like what sort of tools or techniques you use to make product forecasts?

Yes, there are different types. First of all, there are different aspects where we do these analysis and planning. So there's definitely the warehouse component of it, like what's the inbound and outbound from a warehouse. So because that helps us understand how much goods will flow from FCS or dc, how we need to plan it on a given day and then where it will go. So it would go to a store or another fulfillment center or maybe directly to a customer's home. So there are different channels that we have to work through to ensure that we have again the right inventory to support all them, but then also have a good understanding of when the goods needs to go out so that we are able to replenish in stores or even ship it directly from our FCs and DCs. So different kinds of forecasting techniques are being used here to ensure again, a lot of different complexities, but then mostly different kinds of forecasting techniques to figure out how we can best forecast demand of a given item in a window of time. There are many internal signals being used where we can analyze things like past trends, etc. But then many external factors are also brought in to enrich our modeling the AI part of it, where we can build algorithms which are very attuned to these signals and are able to do a better job at forecasting.

Okay. And what happens when things go wrong. So I think like there must be unanticipated events happening all the time. So how do you deal with this? Because you can't predict the future, right?

Right. Yeah. I mean, I think Covid was a big example of that. Like you cannot predict these Things and things can go wrong. And definitely that's why we have a lot of very smart people looking at this data and making these decisions and putting in their feedback into the systems in a very data driven way. So we have a kind of a reinforcement learning going on where we learn from what went well, but what didn't go as planned and then how we improve our algorithms to basically make it more aligned, more attuned to the demand which should actually be captured.

Okay, so when you talk about reinforcement learning, are you talking about the machine learning technique or is this like just.

You have human processes in the loop as well? Yes, of course there are, you know, like some very, like I said, very business savvy people who understand the market, they understand what went wrong or what was the difference, why the difference happened. So of course their inputs are also necessary for us to understand and mimic some of those behaviors which they from their experience are able to identify and share. But we now need to have a way to implement it in our models. So we take those qualitative feedback as well. And of course there is a, like a feedback loop which is more of a pipeline that's built to capture that feedback and then our models are fine tuned or retrained on those as well.

Okay, so you're taking whatever real world or business events happen and then you're feeding those back in to improve your models.

They're very possible. If we know like a holiday is coming in and we know something will be sold more than time, then yeah, we can definitely use that kind of signals to enrich our forecasting signals.

I'm curious how much of the workaround data you do is sort of generic and how much is specific to retail. So are there any challenges around working with data and AI that are really retail specific?

Good question. I don't think anything we do do is generic, to be honest, because we are, we are always solving for the next big problem. And I think all are geared towards providing that holistic and frictionalized experience. So I don't think I'll categorize anything as generic. If you're asking more in terms of what I would categorize as highly sophisticated algorithm versus not so much if that's what your intent was, then I think we try to match the solution to the type of problem. Like if a problem is, can be solved by something predictive analytics, then we don't need a deep learning algorithms to be built to go and solve just that because we want to. But yeah, there are challenges where we really need to go Deep we really need to build more sophisticated algorithms so that because that problem is needs that. So we definitely have that kind of approach. It's not about over engineering anything. It's about finding the optimal way to solve for a problem using what we have. So that's the approach I would say we take. Again nothing generic because everything itself is a big problem and something that solving at Walmart scale is a whole another challenge to go and work towards.

Actually just to follow up on that. What helps you scale things? I know you do it like everything has to be a gigantic scale. Are there any sort of tips or tricks you have for scaling your work.

With data for scaling? We definitely have seen that learning more about machine learning techniques, machine learning engineering in itself becoming more and more important where we it's not just about building the right algorithms but then how do we take it and roll it out in production which can then serve millions and millions of customers in a way that is providing the best experience in terms of being real time or near real time. But then also helps us manage our resources like compute resources and our and the amount of effort that goes into supporting these capabilities. So it's definitely challenging at this scale. But then I think machine learning engineering is becoming more and more pertinent as we go roll out more of these AI solutions.

I think one of the biggest complaints I've heard from people working at large companies is it takes longer to do things and makes it harder to stay cutting edge. So since Watermark is still about as big as it gets, have you got any advice for how people at larger companies can stay up to date with their technology, with their data processes and capabilities?

Yeah, I don't know if you saw that, but I think it was yesterday or day before yesterday Walmart was named as the fifth most innovative company in the world. And we came this is in comparing ourselves to Google or Amazon, Microsoft, et cetera. So I think it very much tells you that even at scale scale being very relevant in terms of the next technological advances as possible and with generative AI search I think Walmart was one of the first in the market to do that. So I would say that is definitely hard but I think good and focused intention of being the best in technology in retail we are able to do that and we are able to bring in that best in class technology to Walmart and then build solutions which are best in class for retail. So I don't see that to be a challenge where I am and in my team now.

I'd love to talk a little Bit about skills. So within your team or across Walmart more in general, what are the most in demand data and AI skills?

The most important skills I would say is definitely those for data scientists and machine learning engineers. So skills like data analysis, statistical analysis, predictive modeling, good coding experience with technologies like Python or et cetera are definitely very relevant. I would also say like there is a lot of these no code or low code AI tools that are coming up in the market and a lot of things can be done using those tools. But I still feel like having very solid coding background, critical thinking, analysis experience will go a long way just because it can enable you to build many more bespoke solutions which these tools might not be able to do that. So I think those are some of the skill sets I would definitely say are very critical.

So you mentioned a few different sort of tools there and techniques. I guess because you've got so many people working data at Walmart, there must be a sort of variety of things going from one team to another. How do you synchronize skills and synchronize tools and things like from one team to another, Is there any attempt to do that?

Absolutely. We do have like every organization have their hiring standards. Like for this kind of role I want we will need this kind of skill set or this kind of a profile of a candidate. So those kind of I would say normalizations happen, especially like if I'm leading a team, I'm leading multiple teams, but the skill sets needed in different teams might be different. People who might be building search and personalization solutions might need a different skill set versus somebody who might be focused on building more demand planning solutions. So those kind of variations do exist. But at the same time having a good benchmark of what are transitional skills versus what are very specific skills for the job helps us normalize it across different teams and different organizations.

Are there any skills that data professionals need like to work in such a large organization? Like I imagine it's a very different vibe to work in a startup.

Yeah, yeah, I would say, you know, like I mentioned those tools like Python and deep learning algorithms, predictive analytics, those are definitely some of those. I'll put them as more transitional skills because if you know how to work on them, then you can pretty much apply them to a lot of different solutions. But then I think for companies like Walmart, I think a lot of critical thinking, making sure you are really a problem solver and you're able to communicate these complex concepts very well to a large and a very different set of Audiences who may or may not speak the language of a data scientist is very important. It's ultimately about creating value. So if you're able to articulate your work in a way that is understood by most people out there, then I think that is definitely something that an advanced technologist should work on, especially in the data science and AI space because things get really complicated here. So how do you make sure that you're able to represent those complexities very well to others is very critical to how you make them understand what you're doing?

It seems like communication skills are very important whether you're talking to a technical audience or a non technical audience. Is there anything that you think is helpful for being able to navigate a large organization?

I think just be a confident technologist. I think if you know your work, what you do, and you have, you are very clear about what your strengths are, then I think it's no different than any other place. It works in a very startup mode. We are very aggressive about how we roll out new products and solutions. So I don't think it's going to be any different, but it depends on how self assured you are in your own technical skills.

I'm curious about if there are any innovations that are happening with data or AI in the world of retail. So what are you most excited about right now in this area?

I'm really excited about how of course generative AI that I mentioned earlier, that was just one example, but we are doing a lot of innovation within every other aspect of retail in different domains, in different problem areas, where we are looking at different ways to again use generative AI to either provide automation, get to better insights or to just enable, better associate or customer journeys in their processes, whatever they are working on. So that's definitely one area where I'm excited, but I think I'm also excited about the fact that we are using really machine learning and AI to actually solve for some problems which at least I feel very passionate about. Like one of my teams is really solving about how we reduce waste in our stores using AI. So we are actually looking at waste metrics of if you walk into a Walmart store, there are a lot of things that are being produced in store like pizzas and chicken and bakery items and whatnot. So how do we use AI to basically ensure that we are producing the right optimal quantities at any given time. So we're not able to just provide fresh products to our customers who are walking into the stores, but then we're able to reduce our waste because that helps us to really work towards a better environment. Our sustainability goals and which are I think very long term and something that we need to solve for now so that we see real value in the future. So it's all innovation. It might not be the most clear problem to go and solve for, but then applying AI to actually go and solve for these problems which will make a long term difference but also help our customer experience. Those are the innovations we are talking about and we do that in many other areas of the business.

So that actually sounds like a very fascinating problem. Reducing waste because it is going to help both the customer because they're getting precious stuff and there's helping Walmart because you're wasting less money on throwing things out. I'm curious to set how you can go about like measuring. That is like someone just going, just weighing the trash that gets thrown out every night or something. What goes on there?

Very good question. I don't know if somebody actually measures it, but there are different ways. Like for example, we do look at our sales metrics on okay, how much did we produce? How much did we sell? And was there a difference? Because that is an indicator if you're overproduced or under produced. Right. Like not under, but overproduced. Definitely. Like we bake more cakes than we are able to sell. But then yes, there are different measurement techniques and there are different ways in which stores measures their throwaways. So we are able to capture that data. Now, all of it might not be attributed to just how much we produce in store. It could be due to other things as well. Right. Like something is not of good quality or whatever reason that could be. So we try to attribute what we can to waste due to fresh. But then we have some data as well on how much did we really produce on a given day. And we have other information like seasonal products, etc. So we know that we were going to probably sell more during certain times or less because of weather or whatever. So we do bring in those signals again as well to make it very informed forecast so that our associates in store can use it to actually go and produce those units.

Okay. That's like all the people who buy the leftover chocolates like the day after Valentine's Day or the day after Easter.

Yeah. Hopefully they find the freshest products. Yeah. All right.

Are there any other sort of trends or things that you think are changing.

The world of retail From a technology perspective? Generative AI definitely is the big one where it's really disrupting the way people are shopping, how people are looking for products, basically people are no more looking for a specific item. They want to have more of a idea or a intent converted into products that are being surfaced to them. So I think that is definitely going to be very revolutionizing because it's about again, the customer experience, but also helping them find the most relevant product as quickly as possible. So I think that will be really disrupting. But yeah, I mean, retail is a fascinating space and I'm having the privilege of working in different markets. So I can I see how Canadian customers are different than Mexican customers and then customers in South Africa. So I think it's an ever changing. It's going through its own evolution and it's an exciting journey.

Okay, that actually sounds fascinating. Are you able to share any insights on like what the difference between a Canadian Walmart shopper is from a South African Walmart shopper or vice versa?

Yes, I think everyone is in their own journey adopting technology to shape their shopping experience. So like a Canadian customer would be, I would say, much attuned to a US customer in terms of how they are using online apps to shop versus actually going to store or how they're using our newer offerings like pickup in store, deliver in store, things like that. Mexican customers, we operate through different banners over there. So again, within the banners we see a lot of variations in what shoppers are really looking for, what are their preferred methods of shopping with us or interacting with Walmart. And then South Africa is in. It's an old journey of having their own different banners operating different African countries. So they'd also have their own differences in how personalization works for them or how they would want to get their products to them either in store or online. So I mean, there are different kinds of uniqueness that each market has and it's just really very fascinating to learn and to build products which serves the wider need of all these different markets.

All right, wonderful. Just to wrap up, have you got any final advice for people working in retail on how they can make better use of data and AI in retail?

There are many ways in which we can harness data and use it, but I think it will be very important to understand how we can use data in a very ethical way, how we put governance on that data that we capture. And even if we are building AI models, then let's make sure that we are very transparent in our approach. It's very explainable on what kind of analytics or AI we are building on top of it, because that is really important from a customer experience perspective. And I think ultimately retail is all about serving to our customers. So we should definitely keep that going wherever and howsoever we are developing using data in AI.

Okay. I like that. Focus on improving the customer experience and helping your customers. That seems like a good moral. All right, thank you very much for your time, Swati.

Thank you.