# S1 E14 - Blutag - Shilp & Clive

The Middlemen - Episode 14 - Blutag - Shilp & Clive 

[00:00:00] ​ 

[00:00:05] Elgato Wave Neo-24: welcome to the middlemen. I'm Tom Limongello, and I'm here with Todd Sawicki. We, as middlemen, live at the intersection of media and e commerce, and we would like for you to join us in our discussions where we turn that chaotic intersection into your comfort zone 

[00:00:21] Tom: it's great to be here. This is an episode where we have two guests. The company that we're featuring blue tag is one that we've known for, I don't know, I think about a year now, but they're in the generative AI space. And specifically, what's interesting about them is that they are A retail focused search capability. 

[00:00:42] Tom: So using generative AI to improve search for retailers and with us today Shilp Agarwal, who's the founder as well as Clive Humby. Who is an advisor and also someone who's been, as Todd and I have seen in the retail media world has left [00:01:00] his imprint on many parts of that ecosystem. So I wanted to give a little bit of a quick intro on Shilp. 

[00:01:06] Tom: Actually I think probably best for you to just give us your story of how you got to be in Gen AI search, where you came from in that perspective. 

[00:01:14] Shilp Agarwal: Absolutely. Yeah. Thanks for having me guys. Yes. So my background has been in digital retail pretty much all my career. I've been doing this, e commerce search and e commerce sales for over a couple of decades. I started my career in e commerce selling jewelry online early days of e commerce. 

[00:01:32] Shilp Agarwal: And our focus was primarily on mass market affordable products. And the problem that we faced at that point was that because we were focused on that particular segment, we had thousands of skews that we had to sell online, and it was pretty challenging for people to be able to find the right products because the way people search for jewelry is very different. 

[00:01:53] Shilp Agarwal: They're not looking for a certain color and a size. They're looking for something that has an emotion, let's say, right? Like they're looking [00:02:00] for a Valentine's Day gift or they're looking for something for their mom. How do you make people find the right product was a big focus of ours. We did that a lot by, focusing a lot on the search in early days. 

[00:02:11] Shilp Agarwal: At the same time, there was a lot of impulse buy, right? So fast forward when these Alexa devices started to come in voice assistance, Alexa and Google assistant, that's when the original idea of conversational was hitting our head that these devices are going to be a great way for people to express what they're looking for. 

[00:02:29] Shilp Agarwal: In more than just a keyword format, and it's going to definitely play an impact on how people find and look for products. So that's how we had originally started. The idea of blue tag is building something for these digitalist voice assistants Letting people shop through that. 

[00:02:45] Tom: So you were an Alexa app. 

[00:02:46] squadcaster-ccgd_1_09-24-2024_131355: Yes We were at that point. 

[00:02:48] squadcaster-ccgd_1_09-24-2024_131355: And then, there were limitations of how people could use that because, you still have to interact with the product catalog. And when, when initially GPT2 had come out, we realized that was going to be something that's going to really help enhance that [00:03:00] experience. And that's when we started to train our model specifically for e commerce to let people have that dialogue with product catalogs. 

[00:03:07] squadcaster-ccgd_1_09-24-2024_131355: And our mission always was that let people have. The best possible dialogue they can have with any product catalog. And, fast forward now the Alexa ecosystem had some changes, their priorities changed. But what happened in that situation for us was that we had this really strong model that was now ripe for the new era of people wanting Gen AI on their websites, mobile apps and everywhere. 

[00:03:32] Todd: So to explain a little further, Shilp, one of the things that you all did, and I think it's important to understand, which is, as things like ChatGPP and conversational interfaces to AI emerged, you all ended up building your own proprietary model. And your own conversational search model. I think that's important for people understand because I think everybody thinks that anything they see a is just a wraparound open a I and chat.[00:04:00]  

[00:04:00] Todd: And I think that's one of the things that's interesting. But what you guys have done is building your own proprietary model. And what that means in terms of your ability to deliver a really interesting solution for retail. 

[00:04:12] Shilp Agarwal: Yeah, exactly. Because right now, for the last couple of years, since chat GPT became the new thing, it's easy for people to just say that, okay, we can wrap things around chat GPT. And technically. Deliver a solution for Jenny. I search a lot of issues with that. LLMs by themselves cannot solve for everything, especially when you're looking at these massive generalized Language models that are not just slow, but they're also extremely expensive if you try to do that. 

[00:04:39] Shilp Agarwal: So yes, our model that is a vertical model that we have a proprietary model on. It's not just LLMs. It's a combination of LLMs, caching, vector search, recommendation engine, all of that combined into one makes it that powerful search tool that is needed for people to cater to this next generation of people looking for products. 

[00:04:59] Todd: To jump [00:05:00] into a little bit on the architecture in terms of, cause I think a lot of people don't necessarily understand that there's more to. Delivering Gen AI or Gen AI search than just an LLM. And right you're, you've built a solution that encompasses a lot of things like the chat interface, the natural language processing, and the ability to like not hallucinate and do other things are really interesting. 

[00:05:21] Todd: I don't think a lot of people understand what it takes to deliver a solution. For retail search, like you can't just use LLMs to do that. I think that's where people have misunderstood where I think JNI is going and some of the things I think you guys have done that make it interesting and deliver a really compelling, fast, quick, and relevant user experience when it comes to search in terms of what we've seen. 

[00:05:44] Shilp Agarwal: Exactly. Especially when you look at that NLU piece in the beginning, that really almost works like a switch to see that. Okay, which part of this model do I need to really hit? Do I really need to go and try to get information from an external LLM? Or do I just need to hit an internal model? 

[00:05:59] Shilp Agarwal: [00:06:00] Because those are the things that Avoid your model from hallucinating and let people actually find what they're looking for. Yeah, our model does not hallucinate. We can say that with a lot of confidence because that's how we set it to make sure that it only results returns results that are actually relevant and otherwise it's not going to return anything. 

[00:06:17] Tom: I think actually this would be a good time to give a little intro to Clive because he can give a retailer perspective in terms of, what are the needs for search now that there's a generative AI capability out there? And what are the gaps between current search and keyword search on every retailer that you see and what could be with BlueTag? 

[00:06:42] Clive Humby: Yeah I've been obviously working with retailers for many years, started in retail location in the eighties moved into loyalty cards primarily with Tesco, Kroger and people like that around the world in the nineties and early two thousands. And I think if you look at the [00:07:00] experience of using data to understand what consumers want there are three elements to that element one is. 

[00:07:08] Clive Humby: How do I make my shopping experience as convenient for me as possible? Big ideas in the early stages were, for example, take what you do, you buy offline and pre populate a favorites list. Very simple idea, but tripled click through rates made actually the economics of. 

[00:07:26] Clive Humby: grocery shopping online come to life, but it becomes very habitual. You're buying the same things every week. You're shopping from a repertoire, perhaps of 200 items against the 20, 000 that are available in the store. Then you move into, actually. The challenge of finding new things. 

[00:07:44] Clive Humby: We touched on a little bit, where does retail media fit in that? How do you tell people about new things, but also actually what consumers really want is a solution to their problem. We should really think about what does the consumer want from us? And, I want to [00:08:00] cater my mom's 50th birthday party for me and the family. 

[00:08:04] Clive Humby: So how do I specify that? How do I say that? And actually, that's where I think a lot of the idea of voice in visual out really is very powerful. If I can say the things I want and then basically get my equivalent of favorites. But for that special event. And see that visually, then to me, that's going to be the next big thing. 

[00:08:26] Clive Humby: I remember in the very early days when we started using data for the first time online, the steps were massive. You, you saw a massive incremental step by using data for the first time. But then over time, the incremental steps gets more and smaller because actually you solve the big problems and the little problems. 

[00:08:43] Clive Humby: You start breaching down and saying, is it worth solving? And then I think this is for me the next big problem. How do I specify? And the reason I'm excited about what blue tag are doing is how do I specify my need? and get a solution. And I [00:09:00] think that's really where it's at. And then that's got to be the next big thing. 

[00:09:04] Clive Humby: It's only as good as the data. And actually, the reality is blue text models are demonstrating the fact and I've seen it in action myself that they solve those problems really quickly and really accurately in a way that would be very expensive in any other technique. 

[00:09:21] Todd: About Clive and you and your background, and I don't want to, you completely undersell your background and program, right? And to our listeners. You've heard of Dunhumby, the organization, and it really pioneered the loyalty programs and using them for data marketing. 

[00:09:38] Todd: And in many ways, Clive is the godfather of the modern loyalty marketing program, which is why he was super successful, which is why, Tesco bought Dunhumby and Kroger bought Dunhumby America. , it really showcased the way to use data. To target and market consumers. And I don't want to for the people out there, if you have a loyalty card and the [00:10:00] retailers are demanding, you give phone numbers at the checkout counter, if a cashier it's Clive's fault and anyway, I say that in a loving way, but it, you really are one of the huge and meaningful pioneers in terms of the way we manage and work with customers and work with data in terms of retail. 

[00:10:15] Todd: And I think that's one of the interesting things here in terms of your background You know how data is being used in terms of in store marketing, how it's being used by retailers in general. And we talk about retail media consumer data profiles from loyalty programs are the basis of this entire retail media push. 

[00:10:34] Todd: So in many ways, you're not just the godfather of loyalty marketing, you're the godfather of what we're now calling retail media and retail media networks. And what's interesting to me is, even though you are one of the foundational, Lynch pins to what we're doing today that you're a bigger believer in the impact of, let's say, search and Gen AI powered search for retailers versus this push into retail [00:11:00] media and, is the future improving the user experience around things like search or is the future really about better monetizing users? 

[00:11:08] Clive Humby: Yeah, I think, for me, we had some very golden rules when we were working with both Tesco and Kroger and other retailers around the world. We would never try and switch a buyer who was loyal to a brand to another brand. You didn't try and sell Pepsi to Coke buyers. The consumer has demonstrated what they want. 

[00:11:22] Clive Humby: That's not to say that, there are enough repertoire shoppers who will buy both. That you can market to the repertoire shopper. And there are loyalists that you basically want to buy more. So there's, I think the whole retail media space is. It's an exciting space and actually because margins are so tight in grocery retail media feels a very exciting space to be in. 

[00:11:43] Clive Humby: But at the same time retail media has to still meet the needs of the consumer. Ultimately, the consumer doesn't want to be monetized unless it's convenient for them. If it gets in the way, if there are too many pop ups, too many suggestions that aren't what I want, actually that customer is going to shop somewhere [00:12:00] else. 

[00:12:00] Clive Humby: That's what I would do. And, if you're really honest with yourself, would you do it? That's the thing you've got to ask yourself. So I think you've got to, it's getting the balance right. There's nothing wrong with recent media. I'm very excited by it. It was the biggest part of Dunhumby towards the end. 

[00:12:13] Clive Humby: How we were monetizing the data. But actually there are lots of ways to monetize data. 

[00:12:18] Tom: yeah, I'm laughing about this because just this week AdExchanger wrote about how retail media and programmatic are starting to buttheads and some of the outputs of that are Allowing Walmart to do conquesting. So like you said, you don't sell a Coke drinker, Pepsi products or whatever. 

[00:12:38] Tom: That's what's starting to happen. And even places like Walmart where they didn't want to do that, it's starting to happen. And so is that too much retail media? From one, one perspective. 

[00:12:47] Clive Humby: sounds like another podcast guys. 

[00:12:48] Tom: Yeah. And I think, on the other angle of it what's interesting to me was, yeah, when I was at quotient, our playbook was pretty much. 

[00:12:55] Tom: An extension of what you were working on, at Dunhumby but to get [00:13:00] back to Shilp and Bluetag let's get a level deeper on, I think you're calling it retail language model or, what are the guts of that system? Why is it able to actually take the intent? 

[00:13:11] Tom: And what are the outputs look like? And we can show a demo of this as well. , once we finish the conversation. But I would really like to hear , how is it that this is going to be better? 

[00:13:21] Shilp Agarwal: A few different things, right? So once I'll quickly touch on the retail media versus search, I think, being in the space for a couple of decades and focusing on search, one thing we have to realize is that more than half of the users on an e commerce site, they go directly to the search box, right? 

[00:13:38] Shilp Agarwal: And that's why the search volume is always there. We try to realize that the margins are slim. So let's try to generate revenue through retail media, which is great. I think it's a great way to generate additional revenue. I think what gets missed in that is that why are your, search conversions only, let's say 9%, right? 

[00:13:57] Shilp Agarwal: People who are typing something into the search box, [00:14:00] If they're converting low, that's the real problem that I think people need to first fix, improve that experience, realizing that more than half of your audience is going directly to search. And I think that's the problem that I think needs to be fixed first. 

[00:14:17] Shilp Agarwal: And then yes, you continue to combine retail media, but Clive said, 

[00:14:20] Tom: I would go a step 

[00:14:21] Shilp Agarwal: people what they're looking for, 

[00:14:23] Tom: I would go a step further and say, did search ever actually work for retailers? , people aren't buying M and M's. They're not doing the impulse buys because you don't search for that. You walk by that. And so those are the types of things where that model hasn't really worked. 

[00:14:38] Todd: I think that's an important one there, Tom, right? In terms of is it, have we actually solved The consumer experience for retail online in some ways. Yes, obviously you can buy things using your credit card online. Having things show up at your door four hours to 48 hours later is obviously working, but [00:15:00] to your point of there's still fundamental experiences that we haven't seen translated impulse buying, you point out as a great one and complex related purchases like Clive talked about. 

[00:15:13] Todd: Meal planning is still not something that's easy to do. Recipe shopping is still not easy to do. I can't really do a click a button and have a basket created very easily. And I think that's one of the things. That is worth exploring here. Maybe search isn't as solved as we thought yet for the web for Google. 

[00:15:31] Todd: It is perhaps, but in terms of inside a storefront, maybe not. And I think that's interesting one to dive into. And I think, Tom, your point about it as a former grocery guy yourself. It isn't any better. It isn't still a great experience. And so maybe that's the thing is that we should, we can talk to. 

[00:15:49] Todd: And I think that's one of the things why Clive, maybe you're excited about Blutag which is, you can see this. You talked about your catering. If I want to cater for a party search, isn't a great retail experience right now. And [00:16:00] maybe the future is something where, Shilp and what you guys do at Blutag it is a better experience and you can deliver oh, I, yeah, we can solve the basket problem, the related purchase problem in a way that hasn't been done. 

[00:16:13] Todd: Maybe that's something we should be more honest about,  

[00:16:15] Shilp Agarwal: Yeah. What's been, what's happened is that over the last 25 years, we've gotten trained. By these websites to go into the search box and put in a keyword that we feel is going to help us find what we're looking for, as opposed to telling that search box what you actually want, if that makes sense, right? 

[00:16:35] Shilp Agarwal: So we've been trained to say that. Okay, what is this keyword? That's going to find me what I'm looking for, as opposed to asking for what you need. And I think those are the kind of things that are changing. Now is that as opposed to trying to say coffee. What if I just want to say coffee without caffeine, right? 

[00:16:50] Shilp Agarwal: Like it's it seems pretty basic, but it's not that simple when you look at keyword search, right? So there's a lot involved. So those are the kind of things that are changing, and those are the things that we've been working on and are [00:17:00] making a big impact and how it changes the search conversion. 

[00:17:03] Shilp Agarwal: Because what's happened over the last couple of years is that even user expectation on what they expect back from a search box has changed because of things like chat GPT. People are now not just putting keywords and they're asking questions and they expect answers the same way they're doing it on e commerce and it's translating in the same manner. 

[00:17:23] Tom: how can you provide those answers? What's the content that's coming back? 

[00:17:28] Shilp Agarwal: Exactly what Clive said and these are the conversations that we, when we chat we really enjoy is that, if you think about what's the fastest way to communicate with information and, that's something that Clive is a big fan of. And, we always talk about that as a voice input and a visual output is the fastest way to communicate with information. 

[00:17:45] Shilp Agarwal: Anything else you do? Slows things down, right? So whether it's using your Alexa device to speak. Yes, it's fast. But when you start to hear it back, it slows it down, right? So that's why, happy to show some examples. I'll see if [00:18:00] Clive has something to say regarding that, because I know, Clive, you talk about the visual aspect a good amount. 

[00:18:05] Shilp Agarwal: And that's made us really focus a little bit more on that since we've been hearing that from Clive as well.  

[00:18:10] Clive Humby: Yeah. I think, thanks. Thank you. I think the reality is that, consumers have problems and they want to solve them. And, there's no question in my mind, the big revolution in grocery shopping, particularly was offline to online. That was the big, that was the big revolution. So I took what I bought in store with my loyalty card and I populated a shopping list online. 

[00:18:31] Clive Humby: Which basically reduced that repertoire of 20, 000 SKUs down to the 200 or so that really mattered to my life. The negative part of that is it's really hard to find things that you're not used to in that situation. And there's no question in my mind, it doesn't matter where you look. The interaction that we have, For the most part, the interactions that consumers enjoy most are visual. 

[00:18:54] Clive Humby: We get addicted to Instagram, TikTok, those things, visual stimulation. And [00:19:00] I want my problems solved visually. I want to say, I literally am having a small party this evening. And we have got basically 10 guests. And we decided to put on an Arabian supper. 

[00:19:13] Clive Humby: Now, I would have loved to just simply say, how do I cater an Arabian supper for 10 people, but actually I had to do my homework and say I need this, that, and the other, et cetera. And, part of the problem, and I think Todd mentioned the idea of recipes, one of the great difficulties with recipes is, you click on the recipe for, Nigella's lamb shanks or whatever, and you get the shopping list and there's 12 items in it, it's great, but actually you've already got four of them in your cupboard because there's You know, salt and so basically what are the essentials and what are the things that really make it up and that to me is, that's why it needs to be visual because if you can show those things visually, then I can go, yeah, I want that one, and that one. 

[00:19:53] Clive Humby: I've already got that one. I've already got that one. It's so quick. Whereas if it's like a great big long list and then you've got to go [00:20:00] down in ticket and it's just tedious So for me the future is voice to visual that's where it's at That's going to be the great big next step in making retail successful And I think of all the people playing at the moment blue tag are ahead of the game and there's a big difference I think the other thing you have to remember is You can't get it wrong. So you can get incremental benefit from data in store, for example, on pricing assortment those wins that retailers can gradually grind out a little bit more, a little bit more, a little bit more by using the data well, to merchandise well, to price well, it doesn't matter if you get it slightly wrong in the algorithm in the early days, but you'd say cater in an Arabian meal and you get a pile of stuff with barbecue ribs and stuff like that. 

[00:20:50] Clive Humby: It's a disaster, so you can't get it wrong. And that's, again, a great reason why you need something that's already got a lot of learning built into it.[00:21:00]  

[00:21:00] Tom: Yeah, actually there's one other trend there that you're starting to hit on that I'd love to hear your thoughts on, both of you guys, and that is It sounds like you're going to be returning recipes. So that means that there's content on the retailer's site and the trend in the retail media space is the outside publishers getting paired up, maybe even acquired by the retailer. 

[00:21:24] Tom: And so what do you see as the future of the retailer actually owning some of this content so that you don't have to go to Instagram to find Nigella Lawson's, Arabian pudding or whatever it is that you're, that you're having tonight. I'm actually be interested in hearing what the menu is once you figure it out. 

[00:21:41] Tom: But yeah, generally like I'm interested in seeing it sounds like you're already starting that move with this system to bring that content into the retailer's walls.  

[00:21:51] Shilp Agarwal: Yeah. If you think about, currently, if you look at a lot of the grocers, most of them have a bunch of recipes that either they've acquired or they have because they [00:22:00] have products mapped to that makes their shopping life easier, right? But now what happens in case of let's say you're looking for Lasagna, right? 

[00:22:07] Shilp Agarwal: And they have a recipe for meatball lasagna. That's great. But now what if you want like a vegan version of that, right? So those are the kind of things where you can take their existing content. But then when you start looking at, generally how you plan for meals, there's a lot of this open source data that does not require you to have your own data. 

[00:22:24] Shilp Agarwal: And that's the beauty, right? You necessarily don't need To own this data because it is very limiting off. And it's not scalable, honestly, in that kind of a manner where you do want to make people you want to give them a blank slate where they should be able to ask anything like they should be able to ask. 

[00:22:41] Shilp Agarwal: Hey, I'm looking for some, ideas for some Italian or Arabian. Entree is, can you suggest something and, gives you some options. Visually, you look at them, they look good. Maybe you can narrow it down and then you can buy some ingredients. Those kind of things how much of that do you want to really own and what's, and why do [00:23:00] you really want to own that, right? 

[00:23:01] Shilp Agarwal: Like the question is is there a particular reason why you would like to own all of that content when it's not really needed? I don't know. Is that could be a question for maybe if you've experienced some reasoning why people like to own that as opposed to letting people have access to that. 

[00:23:17] Tom: Think it depends on if there's influence, right? So the influencer, Nigella Lawson connected New York times. You can push those recipes to Instacart. That's a workflow. But it sounds like in your case with blue tag, you could potentially have somebody who goes to their grocer site on a weekly basis, find content just based on the intent of what they wanted to do. 

[00:23:39] Tom: And you don't need an influencer in that case, or maybe that just enhances it. 

[00:23:43] Shilp Agarwal: You could still have an influencer, but at that point, you can actually make it a little bit more proactive as saying that, okay, There's a Halloween is coming up. So let's just try to promote some recipes with this particular new recipe that they've created. And you [00:24:00] can still generate that recipe right on the fly. 

[00:24:03] Shilp Agarwal: And you could still have that influencer coming to you, but they're not limited now to the content. They actually can use their influence to bring people regardless of the limitations of that content, if that makes sense. 

[00:24:15] Todd: Shilp, I think you actually are talking about an interesting coming at it from different perspective, right? Retail media folk are looking at content as a vehicle to get traffic as a vehicle to monetize, and you're looking at it from the perspective of, I'm trying to give a customer the information they need to drive a purchase, going back to that purchasing problem. 

[00:24:36] Todd: And Clive and Shilp, you both talked about one of the things that's interesting about BlueTag is the ability to have a conversational interface. And if you're in a store, it allows you to get out your phone and speak to the device, like typing while you're pushing a card is a pain in the behind. 

[00:24:52] Todd: And the ability to get what you're looking for. And I think the other thing, Shilp, you're talking about here is that, As a conversational [00:25:00] search platform, you can pull in, you call these open source recipes and other information and present it back into a retailer. So it solves the retailers problem. 

[00:25:09] Todd: How do I give information to a customer so they can make a better decision? I think, cloud. I want to go back to something you were talking about earlier, which speaks to one of the reasons that loyalty marketing took off, it was incredibly low hanging fruit to drive improved business and margins and results for retailers, especially grocers. 

[00:25:26] Todd: And you made a comment about how they've basically taken the whole hanging fruit out of loyalty and data marketing. They've squeezed almost every last ounce of margin out of it. And retail media is in some ways, another effort to squeeze more margin out of it. If we do a better job converting people, especially as people move to devices and online. You can radically improve your company's performance through solving a problem that people thought was maybe solved 20 years ago, search, which really isn't necessarily solved when you get into it. 

[00:25:55] Todd: And that's an interesting thing to look at from a retailer standpoint, which is search could be a way bigger bang for [00:26:00] your buck than people perhaps are thinking. 

[00:26:03] Clive Humby: I think you're absolutely right, Todd. For me, what we're talking about here, I've talked about this idea of taking your offline experience and making it online and populating your favorites. What really we're talking about is using my voice to say what my problem is, And populating a list of favorites that would solve that problem. 

[00:26:21] Clive Humby: It's exactly the same challenge. What you're looking at is basically saying, this is what I want to do. I want to entertain. I want to I want to serve up. I want to do this and making it really easy for me to buy the eight or nine things that I need to do that. That's what we're doing. And that's why I think this is the next big step because actually You're absolutely right. 

[00:26:49] Clive Humby: Searching on, typing in mayonnaise doesn't really help you very much, because actually there are 52 varieties of it. There are different pack sizes, this, that and the other. That's not the [00:27:00] problem. The problem is solving a meal event problem. I've got 20 kids coming around for my son's birthday party. 

[00:27:07] Clive Humby: What can I give them? I'm worried about nut allergies. That, you know, that, that's a very specific problem that probably if you just said those basic thing, party for 20 kids but I'm worried about nuts. I'm worried about this. Oof. Here is a list of 40 things you could buy visually. Now just tick the ones you want, or even this is where you'll find them in the store. 

[00:27:29] Clive Humby: It doesn't have to be online. It could be, as you say, on the cart I could talk to my cart and get a list of, you want to go to aisle six, aisle five, aisle 12. These are the things you want to put. And I've still got to pick, I'm still going to select, but I've got a list of favorites for that problem. 

[00:27:47] Tom: great. I wanted to thank you guys. Shilp what's the best way that people can find you to learn more about BlueTag? 

[00:27:54] Shilp Agarwal: We could go to our website blue dot a I and we are about to release a whole new set of product offerings that are all [00:28:00] got to have to do with you know how Jenny I can influence, whether it's your search results, whether it's your meal planning or it could just be as simple as product recommendations. 

[00:28:10] Shilp Agarwal: We just recently launched predictive card building. Essentially, for grocers, people spend so much time shopping for groceries every week. And the reality is that, as a retailer, you have enough data that you probably know more about what the customer wants next week as opposed to the customer himself or herself. 

[00:28:28] Shilp Agarwal: That's what we're doing right now is being able to save these customers a lot of time and a lot of frustration, providing them with a lot of delight. By just predicting their shopping carts for the next week, and we're getting to a pretty strong accuracy, and I think that's where that's those are the kind of things that we're really working on, and all that is available on our site. 

[00:28:47] Shilp Agarwal: BLU dot AI. So yeah, 

[00:28:49] Tom: great. Yeah, please go to BLU. ai and I want to thank you Shilp and Clive for your insights and yeah, it was a great conversation. So thanks again. 

[00:28:58] Shilp Agarwal: thank you. Thanks. 

[00:28:59] Clive Humby: I [00:29:00] see you. Take care. 

[00:29:00] MacBook Pro Microphone: So for the next two minutes, we're going to show a quick demo. It's pretty easy to fall along, but if you can't see it on YouTube or Spotify, you can fast forward.  

[00:29:09] shilp-agarwal_2_09-24-2024_135101: All right, cool. So here's just like a demo site so you would have your grocery site. You would have a search box and then we add this shopping assistant. So, you know, I'm clicking on this thing. It pops up in a shopping assistant. 

[00:29:20] shilp-agarwal_2_09-24-2024_135101: Um, you know, I could do simple recipes. I can see things like find me a recipe for chicken tacos. It's going to find your recipe and from that particular catalog, find you all the ingredients that you need. to make those chicken tacos, right? You can swap those products and you can just add them to your cart. 

[00:29:38] shilp-agarwal_2_09-24-2024_135101: Now going to Clive's example of kind of how, you know, his problem of having some meals. Let's try that and see if something works.  

[00:29:46] shilp-agarwal_2_09-24-2024_135101: Give me some ideas for Arabian supper meals. 

[00:29:50] shilp-agarwal_2_09-24-2024_135101: Show me some with just chicken. 

[00:29:56] shilp-agarwal_2_09-24-2024_135101: There we go. So it narrows it down, right? And then you can, you [00:30:00] know, click on it. You can pick what you want. You can look at different things and it basically finds you all the ingredients that you need for that. It tells you how much time it takes to make it. All of that stuff, you can keep adding, adding them to the cart. 

[00:30:11] shilp-agarwal_2_09-24-2024_135101: So definitely cuts down your shopping time quite a bit. Uh, you know, something different that if you want to look at just on the search side, we have been able to replace search for some people. Again, this is just a demo page so that it can show you what it can do. So I can say, Snacks that are good to take a long hike, for example, right here is going to find you, you know, through the catalog. 

[00:30:37] shilp-agarwal_2_09-24-2024_135101: It should be able to find you everything that you need. Along with that, it can also give you a little bit of, uh, you know, a snippet, which explains. Why it's finding these products, you know, drastically helps conversion on those because it gives you products that are relevant for that particular need. 

[00:30:55] shilp-agarwal_2_09-24-2024_135101: And, , at the same time, in that same query, it can not just find you those product, but [00:31:00] also give you, uh, things if you would like to make on your own. So, again, showing the power of how you can use, , this model of blue tag, which, , Has a combination of that NLU recommendation engine, all of that combined into one 

[00:31:14] shilp-agarwal_2_09-24-2024_135101: making it a super powerful tool for people to be able to quickly find products. 

[00:31:17] shilp-agarwal_2_09-24-2024_135101: So yeah, I hope that's helpful. Helpful to see. 

[00:31:20] Tom: it's very helpful. I mean, it looks like it would be great for retailers to have. And so I think when you talked about this previously, The, , items that you're seeing in the grids that, , for listeners who can't see. 

[00:31:32] Tom: it, you get a grid of products once you've, , decided on what you wanted, what, where's that coming from?  

[00:31:39] shilp-agarwal_2_09-24-2024_135101: So these are so we would connect to the retailer product catalog, right? So we essentially once we connect to the product catalog, our system is then able to, ingest all the metadata needed and then it trains our model specifically for that particular retailer. , I'll give an example. Let's say I'm looking for, 

[00:31:57] shilp-agarwal_2_09-24-2024_135101: recipe for a [00:32:00] strawberry cheesecake, right? And now, yeah, it's going to give you a recipe that, , takes fresh strawberries, take cream cheese, and these are the steps. But now what happens in case there's a retailer that does not carry fresh strawberries, only carries frozen strawberries. , we will now present you with a recipe of how to make it using frozen strawberries. And not fresh strawberries because the goal of this model is to sell from that particular product catalog, right? So, yes, we're connecting directly with that product catalog. Everything is designed to sell for that particular product catalog. 

[00:32:34] shilp-agarwal_2_09-24-2024_135101: And how do you do that is by making sure that the model itself gets trained each individual specific catalog. Um, that helps. 

[00:32:43] Todd: No, I think that's uh, interesting and a point to hit on for our listeners on the retail side, which is this is not a generic LLM. We use the term like a chip likes to use retail language model, RLM, and even more specific when he works with a particular retailer, they then [00:33:00] train a variant of that on that retailer. 

[00:33:02] Todd: So this is not a generic LLM interface that has crawled the retailer some way, somehow, this is a specific variant trained specifically for a basket of user data, a basket of product data. And so that it's, it really has a very unique understanding of that specific retailer, which I think is really interesting and valuable in terms of helping drive better results. 

[00:33:25] Todd: Like how do you drive better conversions for that retailer? You train a model specific to that. I think the other thing that's interesting here in going back to your story about beginnings with Alexa, one of the interesting things with Alexa is you ask questions and using natural language. And one of the big things that came along with chat TPT was they basically took an Alexa like interface and applied it to a. AI model, but that's, what's interesting about blue tag is it's been working with a conversational interface for five or six years. This isn't a new thing. This is a, I have a lot of [00:34:00] experience working with conversational, um, interactions and then applying that now to a broader set of use cases outside of just say Alexa now to devices now to the web. 

[00:34:10] Todd: And so I think that's an interesting aspect for blue tag, which is it has actual real life experience. With this type of use case, that's pretty unique. As you start looking around, a lot of these A. I. Starts running around are just a couple of folks who got together and started building a wrapper around open A. 

[00:34:31] Todd: I. Versus someone like blue tag was actually built something built it on real user interactions and now building a business around that.  

[00:34:39] shilp-agarwal_2_09-24-2024_135101: Yeah, exactly. A model is only as good as his training data, right? So we know that and we have had the luxury of having the most amount of conversational commerce real data sets over anybody else out there. Maybe Amazon, right? We have that luxury and we've been able to train it on real conversational commerce data sets. 

[00:34:59] shilp-agarwal_2_09-24-2024_135101: We've trained it on [00:35:00] over $6 Billion of GMV data by now. So how do you get there as a retailer today? Yes, you could probably build this in house, maybe. Technically, yes, you can build this in house. But the problem is going to be that how do you go from having that product to actually something that works well, right? 

[00:35:16] shilp-agarwal_2_09-24-2024_135101: And that's where I think it doesn't even come down to cost. Yes, it's going to cost a lot more. It's going to do all that stuff, but it initially is just not going to work. And you're always going to be playing catch up. And that's the beauty of having a train model like this is that it's going It's economical and it's ready to go. 

[00:35:32]