Can AI Make Shopping Online Less Annoying?

0:07Hello, and welcome to The Debrief from the Business of Fashion, where each week we delve into our most popular BOF professional stories with the correspondents who created them. I'm executive editor Brian Baskin. And I'm senior correspondent Sheena Butler-Young. When we talk about e-commerce, we often talk about convenience. And it's true that it's easier to buy something with a few mouse clicks than to get in your car and drive to a store and hope they have it in stock. But it's not exactly the case that shopping online is easy or even all that convenient, depending on what you want to buy. Finding exactly what you want among the thousands of items a typical brand sells is definitely not easy, and the tools that retailers provide don't make it all that convenient either.

0:51Nobody likes fiddling with category filters, and product recommendations can seem way off base, and most of the time you can basically just forget about the search bar on a retailer's website. Retailers are experimenting with artificial intelligence to try to solve these problems, some of which have been around as long as the internet. Joining us today is BOF correspondent Mark Bain, who's been reporting on these developments. Hi, Mark, and welcome to the Debrief podcast. Hi, very happy to be here. Happy to have you. So before we dig in, Sheena, Mark, I want to hear where you two fall on the AI hype train generally.

1:27I am personally a fan of AI. I'm fully on the AI hype train. I think it helps us do, if you're really good at something, it can help you do it a lot better. And I'm a fan of it that way, as long as it's not a one-to-one replacement for what humans can do really well. Mark, what about you? So I get to be an annoying tech correspondent guy here. And first thing is like, what kind of AI are you talking about? Because there are a million different types. It's already embedded into our daily lives in so many ways that you probably don't even think about. Every time you search for a product online, whether it's on Google or a retailer site, every time you take a picture with your phone, there's some version of AI already at work in the background.

2:04And what we're doing is we're getting better at these AI techniques, the most recent one being large language models, which people know from chat GPT. So a lot of times when people talk about the AI hype that's happening right now, it tends to be about large language models. And there are certain things that are exceptionally good at, like kind of scary good at. You've probably like had like a chat with chat GPT and it sounds weirdly human. But there are things it's terrible at. Like, I won't use it for anything factual because just the way these things work, they're based on probabilities. They don't actually understand what they're doing.

2:41So for me, it really depends what you're talking about. But I will say that I do think large language models are going to have a really big impact on a bunch of different consumer products and experiences, including how we find things online. Yeah, I fell somewhere in the middle. I feel like a lot of these use cases sound really promising. And sometimes the first time you use them, like the first time you have that chat with ChatGPT, it's quite magical. And then you do it again and you're like, boy, that felt a lot like the first time and it wasn't quite useful enough to actually integrate into my life.

3:14But that's one of the reasons I wanted to have you on is the specific application I was talking about at the beginning, which is the convenience of e-commerce and the way that you find things online. That actually seems like an area that's quite promising and where some of these new forms of AI are actually already being used every day by millions of shoppers. And I was hoping you could tell us about how it works, particularly when it comes to search in e-commerce. So there are a few ways that it's playing a role. One thing that large language models are really good at, you can tell just from having one conversation with it, is that it can understand like user intent really well.

3:55It's not just like a simple keyword search. Like it understands context around what you're saying. You can talk to it in natural language. So that's one hope is that these large language models could change search in a way that you can interact with it more naturally. Just like if you're looking for a product, you can be like, what should I wear to, I don't know, Burning Man next year. And it'll like understand what Burning Man is, what the weather is at that time of year, where it takes place, what the style of Burning Man is, and actually be able to give you some results. Yeah. An example of that that I encountered was I was fiddling around with this on J.Crew's website and they use one of these services called Lily AI.

4:40And I searched for pajamas and I got these kind of standard like blue, like matchy, kind of boring pajamas. And then I typed sleepover party in and I didn't get those pajamas. I got these like crazy, like frilly, like fake feather kind of party pajamas. And that was clearly an example of the search bar actually doing a bit of work for me instead of very literally interpreting what I'd actually put in there. And which one did you buy? I bought the frilly ones with the feathers, obviously. Obviously. I have an example of sort of how that works in a way that's maybe both a boon to the brand and also the consumer, but actually not the intended result.

5:17So I was on Vestiaire Collective about around my birthday earlier this year in March, and I was looking for a Gucci purse or wallet. And I ended that shopping journey with a pair of yellow Bottega Veneta pumps, which could not be further from what I was looking for. And I found that process initially frustrated. I landed on something I purchased, but I'd love to go back and actually get a Gucci purse. And I think Vestiaire is one of the examples that Mark wrote about of a brand or a retail that's actually further ahead in its AI journey now than it was in March and getting me to that Gucci purse, Mark.

5:53Well, I can't say whether or not Vestiaire was trying to guide you to that completely different product. In theory, a search works really well when it's specifically what you're looking for. But Vestiaire recently came out with a new update to its search engine that, in its own words, translates keyword searches into image pattern recognition. The idea being that it's not just trying to match keywords in a product description with the keywords you're typing in. You can type in keywords. It can understand what those look like visually and then find products that visually match that, whether or not those keywords are actually attached to the product in some way.

6:31So that's a great example of like, again, it depends what type of AI you're talking about, because now we're starting to get into like computer vision forms of AI and understanding visuals and that kind of thing. And did Vestia say how much this is helping their bottom line? I mean, is it actually leading to more efficient sales? They said vaguely, they said that the new release is delivering, quote, significantly more accurate and relevant items and leading to more sales of products that appear in the similar items widget on its site. But they don't generally give out the actual hard numbers on this thing.

7:06So it's up to us to guess how much significant means. Yeah, I feel like the proof in this is just the uptake has been pretty incredible. I mean, it does seem marked like there are a lot of brands that are using some version of this technology at this point. Yeah, it's become a whole little cottage industry, actually. There are a whole bunch of startups that are working on search as a service. And these range from Lily AI, which is working with J.Crew, to companies like Vantage Discovery, which says it's working with a number of large retailers that it can't name yet. And again, they're using different forms of AI and trying to come at the problem in different ways with the same goal, which is to make it easier for shoppers to actually use the search bar to find what they're looking for, because search is sort of notoriously terrible on retail e-commerce sites.

8:01And it's probably partially because if you're a customer like me, you don't really know what you're looking for. There's something I have to admit my own flaw in searching for a Gucci person and buying shoes later. So Lily AI is one of the things it does really well is that it'll take something like I'm looking up red pumps, but J.Crew sells, they call red oxblood or something. There's like a process that they translate things that we know naturally as a customer into what the brand's language is. Can you talk a little bit about that? Yeah. So brands, when they're naming their products and, you know, they're submitting product information to online retailers that the retailer then has to like put into their system, they're not thinking about what the customer is actually searching for.

8:44They're thinking about like, how do we refer to this stuff internally? We don't want to call it red. We want to call it scarlet for some reason, you know, because that's, we're very specific about the shades of red and, or whatever it may be. So if you just put it up on an e-commerce site and you tag it as scarlet or oxblood or whatever, and somebody is searching for red, it may not come up. What Lily AI tries to do is again, using computer vision, it can analyze the product imagery and say, these are the different attributes of the product, and then attach all of that to the product. So when you search for it, the retailer isn't just matching the keywords that you're using.

9:26It's creating a much broader net that's allowing it to pull in things that are very close based on Lily's understanding of how consumers speak versus how merchants speak. So I want to pivot to the search leader that is Google, whose shopping experience is getting a major AI overhaul. So moments before we were recording this episode today, Google unveiled a suite of updates. Mark, tell us about the big news. So it's basically three specific new features that add up to what it's calling basically this big AI upgrade of its shopping experience. The features are, one is like an AI powered guide that gives shoppers relevant info about products.

10:10So if you're like looking up what the best puffer jacket is, I don't know if you've come across those Google like AI overviews that can like summarize a topic, but it's a similar idea. And it would tell you like, here's what to look for in the best puffer jacket. You want something that's, you know, water resistant and maybe down feathers instead of synthetic or whatever it is that Google ends up saying. And this is all based on all the information on the internet. Another feature is a personalized shopping feed that uses your past activity to show you items and information that Google thinks will be of interest.

10:43So, you know, if you tend to shop a lot for a specific brand or you've been looking for one type of jacket, like Google will know that and be able to personalize some recommendations for you. It can also show you stuff like YouTube videos or something like that that are related to what you're looking for. And then the last thing is showing you the best deals on products and brands that you've shown an interest in in the past. So it'll have top deals for you. And based, again, on your browsing activity over time, it understands what you're looking for, what you like, and it'll show you what products are currently on sale.

11:20Google's shopping listings, it's like 45 billion products that are constantly being refreshed. So there are a lot of products that it can draw from and it knows what the prices are and that sort of thing. So it's drawing on that information. Yeah, that feature, the recommendation feature really stood out to me. I mean, that's always been the biggest problem I've had with Google Shopping is that you get this mix of luxury items and more mid-price items, and then a whole bunch of fast fashion knockoffs and they're all kind of just jumbled together. But if Google knows using this new tool that I tend to prefer to shop, you know, sort of mid-priced items, maybe it'll start showing me something more tailored to that.

12:00And it'll be a little easier to actually shop using Google, which I'm sure Google would love. Yeah, that's a big part of how they sell advertising is getting people to come and to shop on their site. So yes, this is directly related to their bottom line to get shopping right. Although what do we think of that first feature where it's recommending things actively using Gemini? I mean, you said right at the beginning, Mark, you know, you're wary of using chat GBT for anything factual. So I mean, would you actually trust Google's AI to recommend a puffer coat to you? I would probably look at what it says and then still go do my own research because I don't fully trust it.

12:40I mean, it can be helpful, like something like a puffer coat, that could be a pretty significant purchase. So I would definitely look into it more myself. And one of their other examples was like a tea kettle for matcha. And it's like, ideally, it should have like a gooseneck spout. That's something I probably wouldn't have thought of. I'll take your word for it. Yeah, exactly. But that would be like surfaced by the AI. And then that's the other thing that, you know, these large language models are actually, in theory, really good at is sifting through these really large amounts of information and being able to like summarize it and edit it down and tell you what's relevant based on what you're looking for.

13:19But again, we've seen plenty of examples of that not going so great just because of the way these models work. It doesn't really understand what's factual and what it's just pulling in off the internet. And I know we're minutes away from this launch of this suite of tools, but Mark, do you have a sense of how brands might engage with this? Or does this come with, if you're already using Google, this is just an add on to your existing package? Yeah. I mean, Google, I asked them specifically this question because I figured every brand would want to know, like, what do I need to do? And their answer was basically nothing.

13:49You don't really need to do anything. Like if you're already using that Google's merchant center, keep using it, make sure your product's information is up to date. The one thing that they did tell me is that user reviews could play a much bigger role with this AI. Because again, it's looking for ways to recommend products. If there are a lot of user reviews of your product out there for better or worse, that can play into whether or not it gets recommended. Are they using the reviews in other ways? Are they going to synthesize what people actually said about the product and put that into their summaries, for example?

14:25They haven't said that yet. I mean, that wouldn't totally surprise me. That's actually one of the other ways that a lot of companies are looking to use large language models to help with e-commerce is like summarizing user reviews. You don't want to read, like Amazon already does this, you don't want to read a thousand different reviews to get a sense of like, here, the pros and cons of a product. It's like you use the large language model and it will basically give you like a little summary of that just at the top. Then there's some other retailers that are exploring doing the same thing. I'm very curious of, does this leave more or less room for brands to game the system?

15:03If this is being pulled in through reviews, is there anything brands can do to ensure that their products get to the top of the funnel or to discourage negative reviews? How do you see that playing out? Search is like a constant cat and mouse sort of game where Google is constantly updating how it surfaces information. And then people out there are constantly trying to game the system to find ways to get their information to the top. And I mean, that's what SEO is. And that's not to say that it's a problem. Like it's, you are trying to optimize, but some people are, you know, literally trying to like find the loopholes here.

15:41And there are actually ways that researchers have found they can game the system. This group of researchers realized that they can add something called a strategic text sequence to like a coffee maker's product page. And it increased the likelihood of it being listed as a top recommendation by a large language model pretty significantly. But that's a pretty technical thing. Like if you look at what the strategic text sequence is, it's like gibberish. It's not like a string of like relevant keywords, like red scarlet oxblood pumps, high heels, like whatever. It's not that sort of thing. It's like talking to a computer.

16:21So there are ways you can do that. I suspect companies like Google will figure that out and try to crack down on it. And then people will come up with other ways of trying to game the system. So I think it's this sort of constant arms race almost. The ethical question I'm more concerned about, it's less how companies are going to game Google's system and more how Google is going to game its users. Maybe that's not the right word, But I guess I was talking to someone who works in web design about personalization, which is essentially what this is. And they said, you have to be really careful to fall on the right side of the line between concierge and creepy.

17:00And what you've described Google doing sounds like concierge to me. Like it sounds like a very convenient service that I would like. I'm curious whether everyone's going to feel that way, whether they want their search engine giving them recommendations like this. There is actually some research out there like IBM has done surveys and people do feel like one thing that AI could be helpful with is like helping them find products and stuff. That's all very abstract. It becomes a lot more concrete and maybe veers into creepy when it's, you know, I feel like we've all had this experience where like, we're talking about something, and then you open up your phone and there's an ad for it.

17:41And you're like, how, where did that even come from? Like, is this thing listening to me? Like, there's no indication that that's what's actually going on, but it can feel kind of creepy. The thing that I'm always curious about is like, yeah, maybe it's creepy, but if it's the right product, are people still going to respond and like click that link and maybe go buy it? Like, it's probably more effective to be creepy than to be totally off base. So I don't know. That's true. I guess creepy never really stopped. anyone on the internet. And I think that product category probably plays a huge role here.

18:16I don't mind Google saying of the puffer jackets you've been searching for in the last three months, here's a few options versus something like, I don't know, mental health tools or something else will feel a little bit like Big Brothers watching kind of thing. I think fashion seems to lend itself a little more to this than maybe some other categories. Yeah, I think that's a really fair point. Yeah. Yeah. I think people like when they, I just compare it to going into a store and people do like when a knowledgeable sales associate sort of guides them through the selection and make some recommendations if they're good recommendations.

18:50And I think the search tools can do the exact same thing. And just like you can tell the sales associate, Hey, I'm just browsing, you know, go away politely. You can, I'm assuming do the same thing with Google, can't you? I mean, you sort of can, but only to a degree. It's just the way information exists online. You're leaving traces everywhere you go. You can try to minimize that as much as you want, but you're never invisible basically. But again, it might depend on what information is being used. Brian, your point about the like walking into a store and having a sales associate like recognize you and know like what your past purchases have been and be able to recommend products that they think you'll like.

19:37Anytime you hear these companies talk about like what AI could do, that's like always the example is like their ideal is like having an online sales associate who kind of knows you and not in a creepy way, like in a personal way, where it doesn't feel like you're just some random individual and it's just throwing products at you just to see what sticks. It's like, no, I have a sense of what you like. The idea is if you have never looked for office clothing, it shouldn't be showing you office clothing. If you're always looking for active wear and gym stuff, it's like if you look for a blue shirt, it should ideally, hopefully understand that you're probably not looking for a blue button up to wear to the office.

20:21You are probably looking for, you know, like a blue sweat-wicking shirt to wear to the gym or something like that. And it's interesting because the example of the store associate that can remember your past purchases or know you well, or this concept we call clienteling even that brands use, that's a product of an ability to invest, right? So you've got brands that do that at the store level, but you also have the ones that are doing it with AI. How much harder does it make it for an independent, smaller brand to compete as bigger brands make the investment in AI search tools? So there are a lot of startups out there that are trying to turn these capabilities into services.

20:58You know, like the search as a service that I mentioned earlier. So they could work with one of those. But the reality is like the bigger a brand is and the more resources it has, it's just going to be better equipped to use these things. basically because even if you have a tool that doesn't require that much work from a brand, like a lot of small brands don't have tech teams. It's a handful of people in a studio. They don't have like their own engineers and that sort of thing. And so they're probably using some off-the-shelf e-commerce software. And so there is a bit of a challenge for smaller brands, but just in the same way that they always have challenges compared to bigger brands with more resources.

21:41Yeah, I think when it comes to brands own sites for a small independent brand, this search problem is less of a factor. They don't have thousands of items for sale. It's a little easier to figure out what you want when you're on those sites. What I worry about is actually the Google product and the effect that will have on independent brands. Because one thing we all know about ChatGPT and similar models is they kind of lack imagination. They always go for what they predict you're going to want. And that means I imagine they're going to default to the default when it comes to the puffer code, or they're going to recommend Levi's.

22:15They're not going to recommend some obscure Japanese denim brand unless someone really is precise about how they're searching for it. Yeah, I think that's a really fair point. I mean, the way large language models work is, again, they're looking at probabilities. And the probability that Levi's has been mentioned in connection with denim, as opposed to like a small Japanese maker like Oni or something like that is probably just way, way higher. So yeah, I think that's exactly right. Like if you're using Google shopping, unless again, like you said, you're very specific with your search, it probably is going to surface the more sort of standard stuff.

22:52But we'll see because again, it does also kind of depend what information Google is pulling in to make its recommendations. If you have a lot of blogs out there and user reviews that are suddenly talking up a new brand, like who knows, perhaps it could show up in the recommendations. And I think that's a nice way to sort of pull this all together. It depends on how brands engage, but also how Google and other AI tools pull in their information. Mark, this was such a helpful, informative episode. Thank you so much for joining us and sharing your thoughts on how AI is transforming e-commerce. Thanks so much for having me.

23:28Listeners, be sure to check out Mark's articles, The E-Commerce Search Bar Gets an AI Makeover, and How Google Aims to Transform Shopping with AI. These and other stories can be found at businessoffashion.com and are available to BOF professional subscribers only. You can find the links in the episode notes. You've been listening to The Debrief, produced and edited by Olivia Davies, mixed and mastered by Eric Bria. I'm Brian Baskin. And I'm Sheena Butler-Young. And we'll be back next week with another episode. Be sure to follow us wherever you get your podcasts.