Bonus Episode: How Is Artificial Intelligence Transforming Retail Organizations?
25:10 Share

Today, we're airing an episode produced by our friends at the Modern CTO Podcast, who were kind enough to have me on recently as a guest. We talked about the rise of generative AI, what it means to be successful with technology, and some considerations for leaders to think about as they shepherd technology implementation efforts. Find the Modern CTO Podcast on Apple Podcast, Spotify, or wherever you get your podcast.

Hi everyone, Allison here. Our show is on winter break until March 19th, but that means we have the opportunity to share some fun bonus episodes with you. The first, which you'll hear today, takes place at the World Bank event at Georgetown University back in December of 2023. Sam and Shervin attended this event and moderated and joined panels to talk about how artificial intelligence is transforming organizations.

On this first episode, you will hear an abridged version of a panel interview conducted by Shervin that features Fabio Luzzi, Chief Data and Analytics Officer of Tapestry, the brand that runs Coach, Kate Spade, and other fashion brands, as well as Prakhar Mahotra, who you might remember from the very first episode of Me, Myself, and AI. Prakhar Mahotra is the Vice President of Applied AI at Walmart. Tune in to hear from these two retail leaders about how their industry is being shaped by artificial intelligence. Let's get started.

I'm Shervin Kodabande, senior partner at Boston Consulting Group. I'm excited to be kicking off this panel on AI in retail. Very pleased to be joined by Mr. Fabio Luzzi from Tapestry and Mr. Prakhar Mahrotra from Walmart. Maybe I'll just turn to you, Fabio, for a quick intro and Prakhar, you do the same, and then we get the panel started.

Thank you, thank you. So, very happy to be here. So my name is Fabio Luzzi. My background is in statistics, economics and computer science. So I always worked with data and analytics in different shapes and forms across different industries. I started with IBM, so technology and consulting, then moved to American Express where I spent a long time doing analytics and advanced analytics for business travel, risk management, business insights.

And then after that, entertainment with Paramount and most recently tapestry. So retail and fashion. I'll tell you more about the industry. But yeah, so I think it's interesting to mention about my experience because it gives you an idea of how data analytics can really be applied across different industries. Yeah. Hi, I'm Prakhar. I have been with Walmart for

five and a half years now. Prior to Walmart, I was at Uber. One of the early data scientists at Uber worked on dynamic pricing there. Prior to Uber, I was at Twitter. And then I did my PhD in aeronautics, actually, from Caltech. So I'm an aerospace engineer who became a data scientist. So anybody can become a data scientist. If an aerospace engineer can do that. AI is like...

we are part of a revolution right now. This is like back in '40s when airplanes were getting developed. I think it's the same parallel right now in AI. People are paranoid, people are figuring out whether how to build airplanes back in '40s, World War was going on and same in AI right now. It's the same parallel. So very exciting to be here and share my thoughts with all of you.

Thank you for that. Did you say Aeronautics from Caltech? So I've known you for like four years and I had no idea. I went to Caltech as well, so it's good that we meet.

So let me build on what you said about the transformational power and I'll ask you both the same question. So Fabio first, where do you think the biggest sort of value pools are with AI and how is it transforming retail? Before I answer the question, maybe I can give a quick overview of the industry I'm in.

Many of you may not be familiar with Tapestry, but I'm sure all of you are familiar with the brands we own. So we are a house of fashion brands. We own Coach, Kate Spade and C.R. Weitzman. So three fashion brands, all based in New York. So we are in the business of designing and building fashion.

beautiful product. So it's a very complex industry, as you can imagine, and it's a data-rich industry. For the most part, my company specifically is direct to consumers, so we capture and own a lot of transaction-level data. So it's a data-rich company and it's a very complex business.

You can probably imagine the value chain is very long. It takes around one year from when you start planning and start designing a product to when that product is available on the shelves or online for the customers to buy. So the lead times are very long and there is a lot that happens in between.

So it's a very intertwined steps across the value chain from planning to merchandising when you start building the assortment to buying when you start investing on the assortment and then supply chain, logistics, marketing, and all these steps are connected. There's a lot of feedback loops. So it's a very complex business, but it's data rich. So it's a great candidate to

to really leverage data, AI and ML. So we want to grow, but in a healthy way, with healthy margins. And there are two macro areas where we've been applying AI, and we're looking into AI to make this happen. One is enhancing customer experience, which you can do in different ways, and we can drill down. Enhancing customer experience, and then enhancing operational performance.

Give us an example in either one so that it brings it to life for everybody. Yeah, enhancing operational processes. There are a lot of things that we do across the value chain and a lot of them are time consuming and very number heavy. So one example is allocation, right? So allocating products from distribution centers to the stores, what we call in seasonal location. So the product is built.

It's ready to be sold. It's sitting in the distribution center somewhere. We are a multi-DC, multi-distribution center company, so we have many distribution centers across the world. What, when, and how much of that specific product to send to which store? So should I send how many of these bags to the store in Manhattan or the store in Ohio? You name it.

And the KPI there, the goal, we want to minimize lost sales, we want to minimize stock out. So we don't want to run out of a specific product in a store or online.

at the time where potentially could be sold to a customer. So in that case, we use sophisticated forecasting models. Some of them use AI in the form of different neural networks to forecast a customer demand at a product level, at a store level.

or online, and that helps us optimize allocation from distribution center to stores. The impact is measurable through different levels of experimentations. But yeah, so by doing that, we achieve two things, right? One is automation. So as I said, it's a very number heavy, number intensive that often cannot be performed by human because there are so many, you know, these hundreds of thousands of products that you need to look at any minute.

So we help automating most of the allocation and then the outcome, it's improved, right? So we reduce stockouts. So, you know, there is a lift in revenue. So that's one example where we use it. Wonderful, wonderful. So you gave the audience an overview of retail. I think this was a great example where you got to forecast demand, you got to have a good sense of elasticity and then...

an optimization problem of how much of what goes to what store when. Prakash, same question to you about the value pool. Walmart's a massive company with 2 million plus employees. You guys are everywhere and sell everything. So where do you see are the biggest transformational opportunities with AI? Yeah, pretty much everywhere. I think AI is in the fabric of everything we do at Walmart. You could start from

the most fundamental problems in retail. As a retailer, your objective is what items you want to sell, how much you want to sell, and at what time you want to sell, right? To that, you need to anticipate customer demand. So it looks very simple, but at scale is what makes it

very challenging and that's where AI comes in. And Walmart has been using statistics in the olden days before AI became very popular. It was one of the first companies, I think, pioneers in that area. To now, in the modern day era, like you can think of

The difference between today's customer and 10 years back customer is that customers have a lot of options today. They control the market. I want to buy these chips by walking into the store, but I might want to get the dip by ordering online.

In the previous 10 years back, you would pick it up together. And so I exactly know to keep chips and salsa together. But in today's world, you can order it on your phone. And so I have to make sure that chips and salsa are placed everywhere in the NFC, in the store, and it has been delivered to you using last mile delivery, right? And so this idea of omni-channel retail has came into existence, especially during COVID times. And that's where AI plays a major role.

On the customer side, we use AI to like how customers interact with Walmart, right? Like conversational AI, chatbots, like you can go on walmart.com and talk to one of our chatbots and say, look, what is this? I need an iPad with XYZ specifications. Or hey, I'm going to a birthday party and I need some recommendation for the gifts. That's where Generative AI comes in today. If you're an associate in Walmart, AI helps you with the

with being more productive right like it can tell you okay what you should do next and so it's in fabric of everything we do at walmart at the end of the day walmart's mantra is everyday low prices

So the goal is to keep the prices down. And if you can marry that with the amazing customer experience, the way you interact with customers, that's a win-win situation. And that's how Walmart uses AI. It's pretty much in everything. Anything that you are going on Walmart.com, on Walmart stores, there is AI in it. This is actually quite fascinating from both of your comments, is that retail has been around for decades.

since there's been civilization, but what's happening is everything is getting so much more competitive that the quality of a decision matters so much and the precision of that decision matters and the timeliness of that decision matters. So you each gave some examples of that. Let me switch

to a different question, which is, okay, so there's all this opportunity talked about merchandising and inventory optimization and customer servicing and all these things that could be dramatically improved with better data, better algorithms, et cetera. What's in the way? What's making it difficult? What are the challenges of taking a retail organization with this technology, this opportunity, this data,

uh to getting value i'll ask the question from both of you but prakar maybe you start and then fabio you build on that yeah i think what gets in the way is first of all data right you need to as catherine talked about right the sparsity of data and the fragmentation of data it's a real thing that gets into the way uh more for

If you don't have the right data that you have, I mean, omni-channel retail hasn't been around for a long time. So if somebody says, I need to do one hour delivery on this item, like we don't have data, so you need to do a best guess about it. So that's probably the first challenge. The second challenge is how do you integrate AI in your existing processes? Like how do you weave it into humans' daily jobs, right? And so that's why at Walmart, we are very clear that we are people-led and tech-empowered.

And that clarity, when you go in front of 2 million people and say that, look, we are people-led and tech-empowered, that helps adopt AI. So if I'm an associate, I can easily embrace AI. On the customer side, I think everybody's waiting for AI, so that's not a problem. Internal processes, how do you think about it? And I think...

To some extent, we are also in some AI hype, right? There's a lot more promise about AI that has been talked about. But when it goes to implementation, you like as a data scientist, you encounter some real world challenges, but there is executives are expecting something different. And so I think as a data science leader, one of the most challenging job is how do you set that expectation?

where the person who signs a check is expecting a lot, while you know you can't get there. It's not because you don't know what your stuff is, but because the data is not there or something has broken. And I think that has been my biggest learning. So what I took away from this is there is a technical challenge on the data and stitching and some of the inferences you have to make about the complexity of today's customers' preferences and also the change...

management aspect of it that is people's judgments and managers and merchandisers and merchants and pricing experts that have had

maybe years or decades of experiential knowledge and preferences or ways of doing things are realizing AI is allowing them to make better decisions, but they have to change things. So Fabio, same question to you. What are you seeing as some of the biggest challenges and maybe an example of how you've overcome these at Tapestry?

Great question. So I think it comes down to three things. It's like in many other frameworks: people, processes and technology. So people, obviously, because it's about changing the way of working. So this is a traditional business where people have been doing things in a certain way for a long time. It's not easy to transform. So it's very important

to work with them. So design led thinking, so embed the adopters, the potential adopters, the users in the development cycle from the beginning to the end. So you know, work together with the with the adopter of the technology with a user from the ideation to the design to the deployment to the test,

That will help with the adoption eventually. So that's the people, right? And processes is similar, right? So people and processes, they go often together. And also I think another thing that is important to mention is that depends on the use case. Some use cases of transformation by leveraging AI and ML are easier than others, right? Sometimes

you just identify a point in the process that needs a prediction. We understand now that AI is our prediction machines,

and you just replace that point in the process, you'd replace the human doing the prediction with the machine doing the prediction. Those use cases tend to be easier because you're not really changing the way of working, you're helping actually the person that is in the process by giving them a tool to do their job better. So, you know, some use cases are easier but then on the other side of the spectrum some use cases are more difficult because you are

Then you talk more about system transformation, that then requires coordination and collaboration across different parts in the company. One of the great examples that Goldfarb made, he was a speaker note at one of these events here.

He gave a great speech and he made the example of the America Race Cup, right? So there's been a lot of innovation in racing, even if we don't use sailing as a form of transformation. And they really accelerate the transformation when they realize that it was not only about designing the boat, but designing the way the sailor was sailing the boat. So that coordination that needs to happen to change the system, not just the prediction.

So, you know, some use cases are more complicated because of that. So always work with the right people along the process. And then last but not least is the technology. We know Walmart is very advanced in that space, but one of the challenges in companies that are more traditional and non-technology companies is that they need to understand that AI, analytics, ML, it's a technology problem at the end of the day.

You cannot just focus on the use case. You need to build the technology framework that can help you to scale. So, yeah. Wonderful. Wonderful. You know, the people-led tech-powered, I take it from that...

AI is not replacing human, but empowering and augmenting. And maybe to build on what Fabio was saying, Prakhar, maybe you give an example of how AI has made a individual or a group in the organization sort of more powerful and whether that's been effective in sort of creating change.

Yeah, I think a good example could think about is a classical job of merchandising. If you're a merchant, that job is to get the items and negotiate the lowest price possible and then decide what those, where would those, how would you place those item in 4700 plus stores that we have. Now for a human to understand

the neighborhood around 4700 stores is very daunting and then multiplied by 300 or so items that are there that I am responsible for which toy goes where like which which which which apparel goes where so it's pretty much impossible for humans to and so a classical way of retail was to cluster the stores and they say okay I'm going to have 10 clusters and I'm going to just send these items and the process worked which is why

Basically averaging it up. Averaging it up. And that's why if you go to, like five years back, go to any retailer, large-scale retailer, the stores would look similar everywhere, right? Like same items. But I think Walmart pioneered this idea of personalized store. So not personalized, like so I can...

because we understand the customer demographics and the store of the community, store of the neighborhood, we can precise that, look, this store in San Jose, there's more, um, certain type of population, say more people of Indian origin lives there. So maybe I should put Basmati rice there, right? Like, uh, uh,

and it will sell, right? And so you can build a supply chain for that if you're very good on your demand, right? And that was pretty much impossible. Not this breakthrough cream because you have more compute today, but also you had powerful algorithms that could do that and you could learn from your mistakes quickly. So you had the ecosystem to get this going. So that's one of the very good examples. So you organized the stores around, rather than around

some average of break 10,000 stores into 10 clusters into each store is unique because the demographics unique and you manage the merchandising and allocation that way, which also explains why when I go from one place to the other, I can't find anything because I'm not from that place. I over-forecasted or under-predicted.

But one day you'll figure out how to do that. I want to switch gears and talk a little bit about Gen AI. You know, from what I've seen in industry since a year ago when Gen AI made a big splash, of course, large language models have been around for, you know, quite four or five years. But since everybody started being more exposed to it, many...

companies are either rethinking or doubling down on their AI strategy. So I will ask this question to both of you in terms of your views on Gen AI as a transformational power and how it relates to the more predictive, I would say, traditional AI. And I start with you, Fabio. One difference of GAI compared to more traditional AI methodology is that

it can be put in the hands of non-experts, right? So anybody can use an LLM model in the form of a conversation and make decisions based on the output. It's a double-edged sword, right? So it comes with a lot of power but also risks. So because of that... Like giving a race car to a regular driver. Yeah, to a five-year-old...

Having that in mind, I think it's very important not to forget the human at the end of the equation and making sure that you take into consideration that there could be biases, there could be mistakes, and also the way we validate these models, it's very different than the traditional way we validate, you know, more traditional ML algorithms. Okay, very good. Prokhar? Yeah, I think...

Generative AI is like, I think it's that quantum jump in AI because it's probably, as a data scientist, I understand what generalization is. But I think what large language models proved was that, look, yeah, I can, at scale, that I can have one model do multiple tasks. And so elegantly, the other person will not even realize that.

And that was so powerful, like just on the technical side of it, that it brought generalization to light. It also showed something about the human nature, that humans are actually very forgiving. Like hallucination was not a thing before this year. And now people are forgiving to JTPT or to Palm 2 or to Bard, right? Like, okay, I gave a wrong answer, okay, no problem. And the Twitter and all the social media feeds are filled with like all sorts of things that can go wrong.

So that's very powerful. I think it advanced adoption of AI by light years. It almost is so contagious that it puts enormous pressure on predictive AI to not perform. But predictive AI is at a saturation point right now. Most of the models are data hungry, they require labels of data.

And I think generative AI on the other hand is like self-supervised, so I don't need it. I can train models and scaling laws are still at work, right? Like, I mean, it's not yet proven that look, where does it end? There's more parameter. Are there more, if I were to add maybe a billion more parameters to GPT-4, do I get a much more advanced model? Like it's like, so there's no stopping right now. And that puts this crazy asymmetry in place

If you don't understand all this and you are a business user and you're like everything is AI and you don't understand the difference between two, this is crazy for you because now you're like, okay, if ChatGPT can answer, write me something, why not can you predict me something? Fabio and Prakhar, thank you so much for your insights.

Thanks for listening to Me, Myself, and AI. We believe, like you, that the conversation about AI implementation doesn't start and stop with this podcast. That's why we've created a group on LinkedIn specifically for listeners like you. It's called AI for Leaders. And if you join us, you can chat with show creators and hosts, ask your own questions, share your insights,

and gain access to valuable resources about AI implementation from MIT SMR and BCG, you can access it by visiting mitsmr.com forward slash AI for Leaders. We'll put that link in the show notes and we hope to see you there.