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 podcasts. Even digital first companies approach technology implementations with caution.
ensuring they limit their exposure to risk. In today's episode, find out how one e-commerce retailer thinks about implementing and scaling AI. I'm Fiona Tan from Wayfair, and you're listening to Me, Myself, and AI. Welcome to Me, Myself, and AI, a podcast on artificial intelligence and business. Each episode, we introduce you to someone innovating with AI. I'm Sam Ransbotham, professor of analytics at Boston College.
I'm also the AI and business strategy guest editor at MIT Sloan Management Review.
And I'm Shervin Kodabande, Senior Partner with BCG, and I co-lead BCG's AI practice in North America. Together, MIT, SMR, and BCG have been researching and publishing on AI for six years, interviewing hundreds of practitioners and surveying thousands of companies on what it takes to build and to deploy and scale AI capabilities and really transform the way organizations operate. Today, Shervin and I are excited to be joined by Fiona Tan.
Fiona is the CTO at Wayfair. Fiona, thanks for joining us. Welcome. Thank you for having me. Let's get started. We've got listeners throughout the world that may not be as familiar with Wayfair as Shervin and I are. We could probably look around our rooms and find Wayfair items. But can you start by describing Wayfair? What does Wayfair do?
Sure. And first of all, thank you for being customers. Always happy to have customers to talk to. But yeah, basically, we are a digital first retailer in the home goods category. We've also augmented that now with some stores, opening our second store in the Boston, Massachusetts area. And so really excited about that as well as we move to what's being an omni-channel retailer.
That's the opposite direction that most people go. You know, it is, but it's actually kind of neat. It does afford us some interesting ways of approaching it because we are digital first. I think hopefully you'll find that there are some really nice, the way that we are able to tie in the digital aspects. You go in the store, you see what's there, but you can also see the rest of our catalog in a way that's hopefully really useful and a little bit different than the typical brick and mortar shopping experience.
Part of it too that's really interesting about Wayfair and our approach to AI/ML, right? A lot of the, it's under the covers and you don't realize, but what is actually powering the entire experience that you have as a customer and then also for our suppliers, there's a lot of machine learning and AI behind it that's not visible, but it is actually powering everything that we do. For example,
There's a lot around trying to understand the customer's intent. And we do that in as much as what they can tell us in a search strings, et cetera, but also based on where they've looked and how much time they spent looking at something versus another thing. So we try to build up our customer graph.
And then we also look at the products, the items that we're listing on our site. Because of the category that we're in, we don't really have as much branded items. So it's how do we use AI and ML to upload as many items as possible? And we have tens of millions of items on our site.
and be able to get as much product information as possible. Some of that we get from our suppliers, but a lot of the product information we are using AI and ML to actually glean from the photos they give us, from the text that they give us, to be able to form our product understanding.
So we build out customer graph, we build out a product graph, all using IML. And then we do that matchmaking. When you're on our site and you're looking for something, or we can personalize based on what we already know about you. That's the magic. How do we find you that perfect couch when you can't really describe it to me in a very succinct way? Fiona, tell us a bit about your own journey. How did you get into technology and how that evolved?
I went to MIT as an undergrad and I took my first computer science class, 6001, and fell in love with it. And it's one of those things I look back and I'm like, I'm so fortunate to find something that I enjoy doing and I realized they're going to pay me money for it. And this was one of those really fortuitous moments, I think, when I realized, hey, I've always loved solving problems. I always loved optimizing whatever I was doing. And here's a field where I get to do that in practice.
I did my master's as well in computer science. And then I've worked in technology my whole career. Started out building out enterprise software, right? So really looking at how do you build solutions that can be adopted by any and all industries.
I spent some time at Oracle starting out and then at a company called Tipco for a long time, essentially building our enterprise platforms. And then I moved over to Walmart and now Wayfair. The platform mindset that I got from the first two thirds of my career is still very prevalent. Even when you're building a very specific use case, you still want to try to use that platform mindset because it then allows you to build out solutions in a much more scalable, extensible way.
It's still relevant. And then you get to focus on a very specific problem set. You get to be much more business outcome based. And then those are the things that are really different about a very specific retail use case, for example, versus building out an enterprise platform. In one case, you're very, very close to the customer.
And you get to focus on a particular problem set, but still building it, I would say, with the right architecture platform mindset that allows you to then scale, whether it's horizontally, vertically. I think that's one of the things that I've found that has been useful, my background in enterprise platforms. Great. Can you comment a little bit about the overall philosophy of how you're thinking about use cases? Absolutely. And I think that is another key tenant of how we operate at Wayfair.
We actually started out in marketing. This was an area where we felt like AI and ML could really play a big part. We say, hey, look, from a marketing standpoint, the bidding and how much I should bid for and where I should spend the money from a channel perspective, those are things that we feel like we can control and are lower risk.
If we get it wrong, maybe we pay a little bit more for an ad that we needed to. But these were areas that we actually invested in first because we could learn and use AI and ML for that and control the amount of risk that we were following. And then once we figured
figured that out. Once we got more into the use of ML, we then looked at other areas that we could apply it to. So how do we apply similar technologies in terms of our pricing and demand generation? How do we expand that out to the rest of our supply chain, the catalog and understanding of our products into search and our understanding of the customers? And so if you look at where we apply AI and ML now, it's much more prevalent, but we started out with this very specific use case around marketing and customer acquisition and
That was the first place that we started using AI and ML.
You have these two rules, sort of the rules of thumb of where you guys, as well as probably good advice to many, many retailers and in other industries as well. What are the two rules of what use cases lend themselves more to AI ML? Tell us more about that. We use a little bit of a risk framework around what is the reputational risk or other risks to the company if we get it wrong.
Back to marketing, a lot of it's going to be if we get it wrong, we pay a little bit more. Other areas where we don't go fully automated because we're a little bit more concerned from a risk perspective could be, for example, product information or product quality. I think we try to do that as much as possible, but then to some degree, this is also something where we would include the humans in the loop to do that extra level of checks. So we don't fully automate because if we get that wrong, that is problematic. Right.
We use that as a way for us to first figure out where we lean into first. If we can automate fully and controllable risk, that's where we feel like we can go a little faster. And then other areas we might go in, but then also involve the humans in the loop, the controls to make sure that we have the extra level of checks. So that's one way that we look at it. The other is around data. And this is obviously something that I think a lot of
Other technology organizations are also thinking about when they think about ML. Are we ready from a quantity and availability of data as well as the usability of the data? I think that's something, frankly, that a lot of companies struggle with, making sure that there's one source of truth versus one
Now there's five people who've used the source of truth, done some adjustments to it. And now I got five things that are sort of similar to that first source of truth and the manageability of it becomes a bit of an issue. We look at where I have good, stable sources to truth. You talked about human in the loop and this idea as...
fundamental or simple as it sounds, I think that it's still a misconception for many because many still think it's not AI if there's a human involved or
It must be that it does everything all by itself. Otherwise, it's really, really not full AI, which, you know, Sam and I have looked a lot at this across companies. And what we've seen is there's a whole bunch of use cases that you're either not going to approach at all if you expect fully automated or you would be suboptimal or substandard or, as you said, highly risky.
What are some examples of human in the loop? I can imagine there might be ideas where human and AI work together and AI has some ideas and human says, oh, maybe not quite this one. Let's try this other idea. Is that also something prevalent in your organization?
Yeah, it is. It is actually quite prevalent. And this is the part where it's really much more business-driven and pragmatic in terms of our application. And so to some degree, there's probably someone out there who might say, hey, look, that's not pure AI or pure ML because you are involving this human or that human. But in our case, it doesn't really matter. We're trying to achieve a strong outcome from a business perspective.
The way we've thought about it is sometimes we use the automation and AI ML to narrow down the choices, right? So we do some of the work initially, and then we narrow it down to say maybe five, six, whatever it is, a smaller number that we can bring the experts, the humans in to make that final decision. So quality is one of them. The other is style.
That's something that's always a little tricky to be able to get right. We can narrow it down. It just makes the human part a lot easier as well, but then also very valuable because some of those are sometimes pretty nuanced. I'm actually not a style expert, so I probably couldn't tell you the difference between like
two or three styles and there's a lot of places where they cross over, et cetera. And so that's when you have an expert. And we do have design experts on staff that can help us with some of those definitions. And then in our space, things change, right? Styles change, what was in, what's not and all that. So again, having the ability to bring in humans in the loop is super interesting and helpful to us. People, as you mentioned at the very beginning, they may not even know that you're using AI or using machine learning.
We are kind of fascinated with this at the moment because it seems like there's a whole lot of uses that people once it's out there and you can actually do it practically. It couldn't possibly be artificial intelligence because that is a mythical being. But once you can do it, well, then it seems normal. You mentioned how widespread use of it is throughout your organization. And how many people in the organization would say they're using AI?
Yeah, I think that's the part around we try to build it into the fabric of the technology organization, right? So we have a data science team, but they work very closely with the software engineers. We want to, again, even within the technology organization, make sure that the scientists who are building the models, that it's something that's actually production worthy.
You want to make sure that the teams are sort of well integrated and we know. So even if you have a software engineer and maybe they're not a data scientist, but they work very closely with them and they understand what the needs are. And that's what we found to be successful. So enough of this pragmatic stuff, though. I mean, you've got ML and AI throughout the organization. You're using it lots of places. What's next? What are you excited about? What's the fun thing that's coming up next?
Yeah, there's a bunch of things that we try to do as well. We're also looking at innovations in terms of incorporating other tactile type capabilities. We have a small group that plays around with looking out for technology, you know, whether it is advancements from the
mobile apps and the native capabilities of the devices that will allow us to do more. So it's looking forwards towards embedding more augmented reality into our shopping experience, for example. One of the things that we looked at also was, you know, there was some technology out there that was allowing us to get you to almost quote unquote feel the thing that you are trying to buy.
The other thing too is that because we are heavily invested in imagery, imagery is a big part of what sells in the home category. And we have a lot of 3D models, et cetera, for a lot of the items that we sell. It's like, how do we then potentially create a digital twin of your home, for example, in the cloud so that you can almost furnish your home virtually to match what you have in real life. And you can use that to influence what you're buying in real life. Or maybe that's your home in the metaverse and you're going to
Finish it a completely different way. So there's a lot of really interesting technology and concepts out there that we are trying to keep abreast of while we're continuing to be practical and pragmatic. But yes. Good portfolio of high risk and high reward and practical stuff. Very good.
The haptic things you mentioned seem particularly interesting. We do focus a lot on visual. We've made so many advances on visual and sound. Yeah, but not so much feel. Yeah, what's the Pantone color set equivalent of haptic or of touch? You know, it seems like if we had some of those sorts of things where I could have an array at home and I could touch these four things and this is what this couch feels like, I feel like that's kind of interesting.
I'm not sure how I want to go there with smell because I'm not sure if I want that Pantone array of smells in my home. But it's exciting to see that you're thinking about these, let's say, non-traditional or non, you know, first two primary senses that we tend to focus on. Yeah. Yeah. We have a segment where we ask you a series of rapid fire questions. You're just supposed to say the first thing that comes to your mind. Are you ready? Yeah. Okay. We'll try. Okay.
What's your proudest AI moment? I think one of the ones that I'm most proud of is as we've built out the AI capabilities across different functions, we have one particular capability that we're now building, which is what we call GeoSort. But basically, it allows us to take advantage of
the capabilities that we have that are foundational, right? The understanding of a product, the understanding of the customer, and then being able to take that. And then we also factor in where products are located. Basically, we have a sort order based on my understanding of your intent, my best understanding of all the products that we have. And then we look at where the product is located, what it costs to ship for you. And then we do another round of optimization around that. In a way, the reason why I'm proud of it is
Because we built the foundational capabilities, we can now deliver second order solutions on top of that. And that's very specific to us, but I'm sure a lot of companies are at that point, hopefully, too, where they have foundational capabilities and they now figure out, oh, there's a second order solution I can now devise because I've laid the groundwork.
Yeah, you spent a lot of time and effort on those foundations and getting to use that foundation seems fun because some of the foundation may be in the suffering category of getting your data house in order and getting things ready for those next thing. Yeah. Okay. What worries you about AI?
In our application, I would say it's part of back to the whole risk category that we talked about, right? We feel good about the way that we are using AI. I don't think we are anywhere close to the boundaries of where we start to worry. And part of it is just around it. We'll be looking at how people are using it, but it's all anonymized. We're trying to figure out trends. For example, when we do marketing, we look at what channels are successful, but it's not going into the details of who bought where.
But something that I think in general, people do need to think about in terms of how you're using the data that you have and making sure that it's at the aggregate. And how do you make sure that it continues to be so? What's your favorite activity that does not involve technology?
I have two current favorite activities. I'm learning how to golf and I think that's going to be a lifelong endeavor because it seems like it's very hard and I enjoy cooking. And it's funny because I approach cooking, I think the same way I do with technology. I'm always optimizing. So I never follow one recipe. I pick out the best parts of like six different recipes. And it's funny because people ask me, oh, which one did you follow? I'm like, ah,
It's actually very complicated. Let me explain. You do this and you do that and you trade off. And it's how I think. So that's how I cook and how I bake as well. It's an ensemble model approach, right? It's just like a random forest. You're just reaching in the bag. You're grabbing another selection out and building an ensemble recipe. Exactly. What's the first career that you wanted? What did you want to be when you grew up?
My first career, I wanted to be a vet. Isn't that like what most children want to be, right, initially? Yeah, and then you took your first computer science class and everything changed. Yeah, exactly. And then I was hooked. Okay. What's your greatest wish for the future for AI? What do you hope we're going to gain from artificial intelligence?
I hope that we can continue to use it and make just really good practical applications of it. I think there's so many, right? And obviously we're using it in a commerce and retail arena. There's a lot of use cases where we can help with understanding healthcare, et cetera, right? There are just so many applications of it. I'd love for it to just be prevalent and for folks to continue to practice it. And again, it's...
looking at the data and helping us understand things that we might not have understood just from an analytical perspective, right? I mean, I think that's the part around the AI part of it is like, may not be things that we might think of ourselves, but looking for solutions in a very novel way. Well, Siyona, thank you for taking the time to talk with us. I think a lot of the things you said about pragmatic approaches and careful about risk, I think those are things that apply in lots of different places, even if you're not a digital first and a
physical second. I think those things apply to lots of people. And I think people learn from that. Thank you for taking the time to talk with us today. Yeah, thank you for having me. I enjoyed it. That's a wrap on season five. Thanks for listening. We'll be back early next year with new episodes. In the meantime, please follow me, myself and AI on LinkedIn to stay up to date and to be the first to hear about bonus episodes and other content.
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.