Seemantini Godbole, Executive Vice President and Chief Information Officer, Lowe’s
Dr. Katia Walsh, Senior Vice President; Chief Strategy and Artificial Intelligence Officer, Levi Strauss & Co.
Moderator: Jonathan Vanian, Writer, Technology, FORTUNE; Co-chair, Fortune Brainstorm A.I.

Fortune Brainstorm A.I. 2021: Reimagining retail

Conferences

thank you so much for being over here. Um The retail industry, I can't think of an industry that's been probably hit so incredibly hard with Covid and with changing customer behavior. So I think it probably the best way to start off C Monti tell me about maybe how you had to adapt. Maybe your models are forecasting system with so many differences I imagine happening. And yeah I think that's a great question. First of all you're right. We have gone through a tremendous amount of change homes mean much more to our customers than they did. They always meant a lot. But in the last two years we have all been exercising at home, learning at home, doing everything at home, kids that at home everybody is at home. So it means much more. Um So I feel like our forecasting models had to account for much more than just historical sales because historical sales were not the only predictor of what's going to happen because lots of our services just went through a tremendous amount of change capturing that data in real time from the way users what they're behaving in their home, how they're behaving how the online demand. Um You know online acceleration. The other thing also is you know, customers greatly depend on devices and data. They are standing in front of a plan a gram in in an aisle and they're flipping out their phones and they already have the data um on the at their fingertips. And lastly the associate expectations around data around intelligence around um all the tools and assisted ai has gone up significantly Kathy how about you? You're coming in, you're you came two years now into your position over at levi. So that's got to be really interesting. You're coming and what a time to come into a role with the onset of covid. Tell me maybe how what you did coming in that might have been different if you would have started pre pandemic times. Well I can tell you exactly what I did, that was different because I've done this kind of work. My entire career is about using technology to help companies gain a competitive advantage. So I stood up ai in various companies including Prudential Financial and Vodafone group in London and all of a sudden I find myself in the original san Francisco startup, levi Strauss and company but in the time of a pandemic. So what we did differently in this case was to go immediately without looking for how do we build this and How do we find the best talent? We just, it was 12 people because I had just come in a few months before that 12 people including me and my assistant and we set up a swat team and we were in front of the executive team every week with specific predictions, having the benefit of being a global company uh and having operations in china. We actually could see what was happening in china, how it was moving through europe, how it would then hit the United States and Canada north America. And we would use data from china to predict what would happen in europe to predict what would happen in the US. Uh and we were able to pretty much in real time optimize our operations, get the best pricing we could possibly get when other retailers had to discount because they were facing masses of inventory in the closed stores. We did not have to discount anything. We charged full price because our models told us that we could and we have the healthiest margin that the company has had. We're using machine learning and this forecasting or this like help me help me understand where machine learning comes into play with forecasting because I assume people have been doing forecasting for years. What is different now. Sure, I'll tell you that before. So the before which still exists even in our company, you, it takes time more than three years to completely digitized company. And remember this is a para retail, it's not just retail. So this is an industry that has not changed in 100 years. Uh it has been analog. We're now making those digits. So it has been manual. We're now making this automated, it has been creative and intuitive but that also means imprecise. So we're making that precise. So planning which is the original data science function in a company like uh retail or apparel company planning has usually been done with lots of spreadsheets, always lots of spreadsheets always starting from scratch regardless of how many times the same plan. I have done the same forecast. We have now done away with the spreadsheet. Not immediately, not for everyone, but we have created a live streaming demand repository where we have collected all kinds of data. It doesn't have to be just about transactions and consumer behaviors. It's also epidemiological models, economic outlooks, whether climate all over the world, some of it may seem irrelevant. It's amazing how you can be sometimes the data integration process, you know that sounds too fancy. We actually did not have time to make it to have this big aggregation. We literally would take anything, we could get our little hands on and start pouring it into this repository and then applying machine learning to that. So more data than we had ever used before and applying machine learning including computer vision for new product That we had never felt before. A lot of work for your 12 person team. Yeah, well I you know what I would act to that forecasting. You asked the question Jonathan What changed? And I think what changed is forecasting. We were always doing it But here is the fact that every store. So if you have more than 1800 store and every story is carrying about 100,000 items, a lot of them are unique When you look across the chain. There are 90 million combinations of item and stores And if you're really going to crunch 90 million combinations do all these permutations and combinations on top of that. There are translate weather promotions. Um you know you are looking at the economic data exactly like Cartier was saying. The other thing you're looking at is mobility data because mobility affects retail trends. So when you're looking at all of that, our ability, because of the compute storage, cloud machine learning algorithms When it's multiplied, greatly accelerated. And that's why we can't even afford to do machine learning for forecasting which was not possible four or five years essentially negate the idea of spreadsheets because you're talking about 90 million variables. I can't even imagine, let's go to row zzz zzz zzz. I mean how I think what we are, you know, we said some of that in the conversations prior, it is explain ability is one of the hard problems of machine learning and ai you know when the model is computing for 90 million combinations, it's difficult for us to answer questions such as it is possible but it is expensive to answer questions such as well. Row L columns, easy, what's going on here. Um and I ping we have to have a mature mindset about how to test all these various combinations, how to do some dipstick reviews, how to look at overall trends to see we are going in the right direction and it is possible to do all of that. How about let's go to computer vision, that's where we've seen a lot of the big breakthroughs and deep learning, you know, identifying objects and images. And I'm sure both of you have very interesting use cases Katia, maybe if you want to start, if you're looking at a computer vision. Certainly. So at levi Strauss and company, we talk about applying artificial intelligence to three areas within the company. One is smarter creation and it's amazing what you can do, not replacing designers but augmenting their work. Um, so this year, in order to make sure that when we got more skilled people, not just 12 people, but more skilled people in the company who knew the industry, but also we're going to pick up machine learning skills. We created a machine learning boot camp, we would take people who had been in the company um, and had always wanted to pick up some analytical skills but had not had that chance. We extracted them from their day job for eight weeks, we put them full time in a very intense bootcamp. We call it a boot camp for a reason, it is not a summer camp. And they did pick up a lot of skills computer link, they large computer vision, they learned, they, uh, so we have beginner and advanced classrooms and the beginner classrooms picked up python data exploration, modeling the advanced classroom picked up neural networks, computer vision and natural language processing. And so with some of those graduates, one of them is in design. He's actually applying computer vision today to create new Levi's trucker jacket. So imagine a neural network that he created a style transfer algorithms. He's feeding thousands of art images into the iconic Levi's trucker jacket. This is a kind of it and outcomes a van Gogh, starry night Levi's trucker jacket or a David Hockney Levi's trucker jacket or a Jasper, johns Levi's trucker jacket, which we're going to actually manufacture that is mind blowing and it's completely new and it's done by someone who just pick up machine learning skills on that's a use case of training and actual applicant application of training and workforce. How about yourself, what are you using some of these technologies for? You know, it is, these technologies are exciting and can solve a lot of complex problems. We did a press release just today morning and what we're saying is we have now figured out a way to measure your own space with your own iphones, um you know, Jonathan, you and I were talking and I was just saying home improvement can be a complex space. For example, if you want to change the carpet of this room, Somebody has to come in, measure, look at the corners, are there staircases and then give you an estimate as to how much material, how much labor. But just imagine if you, as a homeowner could just hold your iPhone 12 pro or 13, which has lighter technology and just go around the room. The app, Lowe's app will give you exact instructions as to how you're walking around And that measures the room. It tells you how much material you need, how much it will cost you. And lets you buy it all on the app just by holding your iPhone 12 pro or 13 in your hands. And I'm really excited about it because it is a combination of vision and a our technology. But the most important thing is this, I think the best products we can ever make is to put a I into product and not make it complex for our customers when the customers are using it. They don't know they are using the Ai powered product, but they find it really easy to do their most complex task and that's what we are achieving simplicity and design. I'm gonna remind everyone if anyone has questions now, is the time to raise your hand. Um Let's get to a topic, I think everyone's discussing here, um which is getting a return of investment on some of these projects. Um you know, because these are all really cool, interesting stuff. The van Gogh jacket is all interesting. How do you get by into executives to be like, that sounds really cool, is going to generate profit. Do we need to be putting money in this? I mean, I think that's a lot of the challenge a lot of technologists have is getting that executive buy in. Yeah. So my job is actually a strategy and artificial intelligence. It is a brand new horizontal function that levi's pioneers across any industry. It was one of the first, I believe it will not be the last roll. And that is because This wonderful technology that we are now fortunate to live with is actually a means to an end and the end is the strategy of the company. What is the vision? What do we want this company to stand for in the next 168 years of its existence. So the function that we have started in the last two years is a pro former P N. L. It has its budget and it has to commit to a specific um profit that it creates. Now we cannot do this alone, all of us in this room know that the best algorithms are meaningless if they're not implemented. So I worked very closely with commercial markets with e commerce, with retail stores, with marketing, brand inventory management and optimization throughout the entire company. But we have a very strict BNL and we have delivered X 10 times a return already even within our first year because if anything this pandemic that we have had to live with has accelerated this like it has with everything. What I'm what I'm hearing is that essentially you're taking all these other teams marketing and everything in to make sure this project goes to partition. So it's not just the data analytics team getting sort of saddled well if it fails, then it's a failure on your team itself, sort of like everyone sort of banding together. We are banding together but we band together around specific products that have a specific P N L go it could be a higher margin is um, it could be revenues, it could be cost savings and we deliver it together. So we have a shared goal and we deliver on it. The question over there, could you please introduce yourself and your affiliation? Yeah, rick JV my octopus interested in the lowest perspective and taking the next level of that personalization you're talking about for measuring the carpet, but also the whole special order process itself, which is always very cumbersome and all the way into the implementation because you have so many parties involved and then, you know, perhaps also on the levi side too as it comes to, you know, special ordering garments as well. I love your question because it's so close to my heart because like I just said, home improvement is really a complex business and to your point, if I want to redo my whole kitchen, right, It can be quite intimidating and a long process and that's exactly what we're trying to target and make sure that it becomes easier and easier. So those today has a kitchen planner which is all virtual, you could completely and again, kind of combination of vision and they are that we were just talking about, you could completely design your kitchen and then what we're doing is we are supplementing that because that's why the data ai engineer, software engineering, uX all needs to work together like Jonathan, you were saying it has to all hang together and what we're doing is your lows app is going to be your project management tool to kind of track everything that is going on with your kitchen project, track everything that is going on with your floating so all these things coming together can make the special project um special order project really simpler. Did you want to go him? Yes. So the apparel industry is in the it's at the precipice of something revolutionary that what's happening to it now is the biggest thing that will have happened to it in the last 100 years and all of us who are wearing something for the most part here, someone manufactured that government and someone sold it after it was manufactured. What's actually going to happen now because of the convergence of various technologies including Ai but not only is that we are now going to first sell and then manufacture so that you can order exactly what your size is exactly the pattern that you want, whether it's Van Gogh inspired or not, you can pick your favorite other artists, the style, you will actually create your dream government and we will make it for you at some point and imagine that's where that data integration? I know it's a fancy word, but still it's got to be so tied together because if you're ordering something that's so personalized you better make sure your supply chain can pull that off. Yes. But if it sounds like science fiction, I also want to bring it a little bit closer to what's happening today. We already have something called future finish and it's the combination of machine learning and lasers. So this industry apparel has not been a good citizen of our planet. Traditionally, it has actually used a lot of chemicals, a lot of water uh, through the combination of machine learning and lasers, we are now doing the finishes and the patterns with no chemicals with no water, but also customizing it to specific patterns that a consumer wants anyway in the world, through the lasers. So that's something right here right now Jonathan to your point. I think it's not just your right data integration but overall process prioritization. Everything needs to be integrated to Well how big of the culture changes that just come into a retailer that might not be accustomed to that. Looking at the world in that prison. I love that side. And the google leader said that, hey, this is a, I really is about change management and it really is a huge change management exercise and you have to do it organization. Why do you have to get you know, we have been talking a lot about skilling up across the organization and all that will have to happen together. Are you doing any interesting skilling programs or anything like that? How are you dealing with the talent shortage? You know, here's the thing, we are talking a lot about talent. We seem to be in the season of scarcity and everything seems to be scarce including talent and I think we will be in this season of scarcity of talent for quite some time. I don't think we should bank upon it getting better. I don't think we should take talent for granted. I feel it makes us all a much better leader because retention is so important to us and so he's hiding new talent. I just think it makes us better managers. I feel like killing up programs have been um you know scaled up at all cos I really love that because I think people have more potential than sometimes we give them credit for. And lastly, what I would say is um you know, we are coming up with lots of creative ways of how people work and contribute and by unleashing all that by skilling up by um you know giving cross opportunities to people by giving more flexibility. I feel like we'll actually open up this space for talent and I'm really excited about that. Yeah, on that note actually the war for aI talent is a two front war on the one front we must hire people from outside, especially in an industry that does not have this inside it yet. On the other front we must also grow our own data scientists, our own engineers and as we speak we devise our graduating our 103rd homegrown data scientists this week uh through the boot camp that I mentioned earlier and we also have an enterprise wide digital fluency for all program, which is just the beginning of this kind of culture. And I imagine you got to keep doing those programs because they're in the business of retention as well. I mean you're you're training a lot of people who are gonna be very valuable in the workforce and what we're also seeing is the data scientists we hired from outside. They actually stay with us more when we up skill people from inside because they now have counterparts who speak their language and wants to help them. We're coming up on one final note and so I want you to give just a tip to get an AI project from testing into production. What's one tip that both of you can share. You know, I would say put you know, don't put a I sometimes people used to proudly tell me that there is this floor where all the data scientists are sitting there and I never used to understand that. Like how can they just sit by themselves and you know, I I never got it. What we have done is we have put our data scientists ai experts in teams along with software engineers, along with us along with product managers. And I think together as an integrated team, they are able to make magic happen. Okay, so mine is very short, two seconds. Think Big Start. Small scale fast. Very good. Thank you so much.