She Cracked the Code on Why You Buy That Military Jacket. Online Retail Has No Idea Why Either.

Foreign. Talks, the number one podcast for e commerce and marketing applications. Martalks is devoted to covering the latest technology developments that drive the global commerce ecosystem from advertising to last mile. Enjoy all of our content@rosensteingroup.com Martalks podcast. Hey, welcome to Martalks. I'm your host, Darrell Rosenst Stein. And today we're going to be speaking with the CEO of the most exciting startup I've had the pleasure of getting to know over the last few years. I first met her at Shop Talk where I was completely wowed by her presentation given on stage of the startup competition, which she handily won. She's bringing taste into fashion, e commerce and online purchasing. Yes, she's bringing the sexy back to online shopping. With that, let me introduce you to Annabelle Maldonado. Annabelle, thank you so much for joining us. Thanks for having me, Darryl. Excited to be here. Absolutely. Annabelle, you just got back from Shop Talk again. What did you think about the show? Yeah, it was fantastic. There was definitely a lot more candid conversation about AI and the different use cases. Well, at least not on stage, but, you know, on the floor in hushed and hushed, in hushed voices. What did you think about that Deloitte survey that just came out where they said only about a third of merchants are actually engaging in any AI deployments of any type and that the other two thirds are kind of sitting on the sidelines waiting to see what happens? Well, I mean, I think it's still, it's still very nebulous to them where and how they should apply AI. And I think that AI builders are not. A lot of them aren't naturally attuned to the retail world. It's going to take time to find fit. I think that. Do you think there's a part of it that retailers at this point because they are so resource constrained and just with traditional software, they're like, oh my gosh, the idea of implementing anything is just anathema to me. When they think about AI and they think about how complex it is, the idea of adopting something like that is just terrifying. It could be. And I think that as you're saying this, I realize it's like AI has become such a loaded word because it carries with it the sense of, like, transformation, which makes you think of migrations and, you know, they're like, oh, we have to change everything. And I don't, you know, change is stressful. Right. So ironically, though, a lot of AI is kind of the easiest to integrate. Right. And I think that for a long time, I think we're confusing also. I mean, machine learning has been around for, for so long, right? It powers, powers Instagram. It powers, you know, a great number of things that we, we call the algorithm, and that's also AI. And we've come to think of AI as like, you know, gen AI. Right. That's whenever it became so mainstream, which is quite transformative. So I think they're thinking of it in these kind of loaded terms, which, which they shouldn't. Oh, I completely agree. I mean, shoot, I've, I've been working with predictive analytics for close to, you know, 30 years. Operations research is the foundation of AI Richmond Analytics. It's just, you know, it's a label. I think that there, there, it's one term that describes a, not an infinite moat of functionalities, but quite a few. Well, let's switch gears here. Let's talk about psyche. You know, I, I, I know you founded it on a thesis that definitely sounds philosophical. And when I describe it that products have a personality and so do shoppers, obviously. And that you could match the products and the people on a psychological level, you'd be unlocking something that the industry doesn't have. And now it's 26. AI agents are starting to actually do some of the shopping. You know, from where you're sitting, do you feel vindicated? Do you feel scared? Both. I mean, neither. I'll be vindicated when, when agents can, can do the job. And yeah, I'm never scared. It's still the same problem. It's the exact same problem on a bigger scale. Right. Because it is in many ways much like a search term, but the data to create those connections doesn't exist. I mean, if we go back to the thesis, the products have personalities. I mean, they have qualities and people have qualities. Right? And there is a match where function and price is the same. You pick item A over B because of these qualities. And it's, it's not always conscious, but you look at something and something in your brain says, oh, that's really me, or that's not me, based on these qualities. Right. And I think that first I had that realization. I don't know if you know this, but I used to do a lot of product copywriting back in the day. No. When I worked in fashion, yeah, that was my bread and butter because fashion didn't really pay. But everyone wanted their product descriptions and I was in demand because I went deep. Everyone else was writing, you know, comfy and stylish. Comfy and stylish, which is all the agentic Examples right now stylish means absolutely nothing. It's all come point being I started thinking about these things, you know, like romantic and saccharine and playful and quirky. Like all those words you could use to describe people as well. Right. And that's not a coincidence. I mean so in the agentic space it's the same problem because now if you think about the fact that if you pull all of the inventory and it's aggregated, it's like a huge multi brand store and the agent has to make a decision. And the way it works now is they're just basically being pushed by all the platforms, the top products from the top vendors. So they're getting like this part of the tail and there's like millions of other products and they don't know how to decide A over B. And the customer's not going to sit there and write like you know, 15 prompts and try to describe something that they don't even know they're looking for. So it's, it's just a much bigger discovery and search problem. Well, one of the things that got me when I first looked at your deck after looking at hundreds of them for retail solutions is the statement that the E Commerce conversion rate right now is still around 1 1/2% after 10 years solid of personalization technology. Billions, billions have been spent on this endless A B testing. I don't even want to think about how many A B tests are out there. The vast majority of the products that the shopper would actually love to find still never show up in front of them. I mean, why has the industry been solving the wrong problem with this? Well, it's not the wrong problem. It's, it's, it's the only key problem to unlocking everything is relevance and getting the right products in front of the right people. It's been solving it wrongly in the sense that I think everything we've. This is, this needs to be really clear. Everything we've called personalization today is optimization. Right. And I even now, I mean we had a lot of sales calls this week, right? Cause it's two weeks post shop talk and everyone's trying to understand what do you, so what are you actually optimizing for? Is it price, is it purchase history, is it color? It's this. And I'm like what do you mean? We've got like over 700 dimensions. You know, real AI is not optimizing for one thing. It's, it's, it's calibrating among all of them. Right. It's understanding all of the things that matter, don't matter, and so forth. Everything we've done to date has been very rote optimization. Sites have been static. The biggest reason is there's zero concept of product intelligence in the stack. There's no product intelligence in the stack. Let's talk about product intelligence, because that is the secret sauce. That is the contextual data. That's what makes a product special, unique, applicable. What is that data? Well, that data is everything that. Everything about the product. And that's not surface attributes, which is the. This is like the gap, right. Everyone kind of thinks in function. This is a. A military jacket. You know, it's got pockets, olive green, it's oversized silhouette, all these things. That's not why people buy it. Right. Those are bellows Pocke. Okay. I should know that as a former product copywriter, but it's been a minute. Point being, it's not like, I think that for a long time we thought AI specialists kind of have that if you feed it enough of this micro stuff, right. If we get the micro right, the macro will emerge. If we know all the details, the surface attribute details about a product, we'll learn the patterns. But like, that's never happened because those aren't the things that matter. Why do I buy a military jacket? I don't know. It instills grit, aligns with my contrarian personality. It gives me a sense of toughness. It offsets something else I'm wearing that has less of that quality. So overall that feels right. People just aren't conscious of why they make that decision of A over B, where function and price is the same. And how many military jackets are out there? I don't know. Again, like hundreds. Right. It has to be the right silhouette, the right this. And we all react to it. We all will say, oh, like, no, that one or that one. But we're not conscious of it. So that product data is missing. And I think no one has taken responsibility for it because everyone thinks it's the AI guys, it's the brands, it's the retailers, and it's just like it's not really. Most tech people aren't consumer product experts. Right. You can't expect them to be. I see that problem. I mean, there's a lot of, you know, this talk, AI is going to solve all your problems. Well, it can't find something that isn't there. If you don't have that, if you don't have these attributes or this data available for an agent, it's not going to organize or find or optimize anything, it's going to go to the next one that is the same price, the same size, the same color. At another site that has that data. Deloitte mapped out five levels of agentic commerce maturity from assisted discovery to full agent to agent transactions and says that about 63% of global merchants now agree that companies without those agents are going to be left in the dust within the next couple of years. But here's what's interesting. That's a statement about operations, not about taste or that contextual data. When did you first realize that agentic shopping wasn't just a fulfillment problem, it was a discovery problem that nobody had prepared for? Yeah, I mean, I think again, I, I, I didn't have an aha moment around this because to me it just was clear that this is, this is the problem on a bigger scale because we're doing the exact thing. I mean let's, let's just go into like CPG for a moment. Right. Which is fairly functional. Like I'm having a diet Coke right now and it's because it keeps me going, whatever, whatever. And I don't know, I had, I had a call with an agency that was working with well, soda but also a granola bar. And you know, I think basically everyone is just trying to get those, those, those micros. Right. Right. Like it's keto friendly, it's like moderate sugar, it has this kind of taste, you know, or like what is the contextual data around a soda? And it's still like extremely limited. And I think we need to think about, okay, once you get the micro, what are other reasons someone might want a granola bar that aren't someone explicitly searching for a granola bar? So low blood sugar, you know, dieting, extensive traveling, what to put in my backpack, like long flight, like all of those things. And then you can think, okay, an agent, if you actually teach it the context around. Sorry, it wasn't a granola bar, it was a protein bar. Who eats granola? Not that there's anything wrong with that, but you, you basically allow the agent to surface those things on its own and then you don't have to worry as much about like you know, GEO or ads because it's kind of making those connections. Like the LLM isn't going to make those connections. It hasn't lived a life. I mean it's, it's trained on data where it can read about that but, but not to the degree that you'd expect that it can make that kind of connection in a session right now. It's like, it's like a crappy salesman. You know, they are in charge of selling you something, but they really don't know what you like and they really don't know what they have. So good luck with that. Yeah, exactly. I mean, it's like if, if Hermes was, was hiring salespeople who didn't have experience and didn't understand the brand, you know, they, they teach it about the craftsmanship and the actual physical product. They would absolutely teach it the emotionality. But the most important thing are like, you know, the clients and what they're like and the ones they like. And these are the ones that like Kelly's. These are the ones that like Brook and like you would teach them. And right now agents have nowhere to go to learn that. Well, maybe, maybe we have to open up a university for agents. But no, that's another, that's another discussion. No, I, I have something like that. And like in the tech space, there is a lot of talk about agentic land grabs that they need more places to go apart from Reddit to go and hang out and learn about stuff. They need to have somebody they can ask questions too. Yeah. Wow, there's a job description I haven't fulfilled a job for. AI agent educator, AI tutorial. Maybe that's a fail to investigate. Sarah Richter said something pretty profound about this. She said that most brands don't have loyalty, they just have repeat convenience. And we're thinking about Amazon here. Now if, if an AI agent becomes the front door to that discovery, is. Is loyalty going to just disappear entirely? Unless a brand owns taste or adds that contextual data? It depends how good your actual product is. So I mean Amazon is actually a perfect example of a convenience factor. And absolutely nothing special about what doesn't have product just has, you know, obviously millions of different products and to varying degrees of quality. But if you have, let's say I have a pajama brand that I really like and I will go back to them for their pajamas if they making good pajamas. At one point I might ask the agent to find me something similar. I may not be aware of like some other mid range, very nice pajama brand, but I think you like. Loyalty is much more tied to the fact of like how good the actual physical product is. And I think if it's, if you're buying it via an agent, it's not really changing anything. It's just a slightly different interface but you have an opportunity there. As if you like that pajama brand or a smaller brand and they sell it on Saks. Do you like it enough to go buy a D2C or are you going to buy it on Saks? And then like, yeah, the convenience factor matters because a lot of the D2C brands I like, you're gonna take like 11 to 14 days to ship from Amsterdam or something. I'm like, well, I can't as much as I love that physical product. So it's sort of like a, a calibration amongst those things. Well, so again we're coming back to so much of it is about those commoditized factors. Price, availability, convenience. So then we're left with whoever controls the agent that is focused on those characteristics has now has the power. So if an AI agent owns search selection and checkout, what exactly is left for the merchant to own? Products. The products and information about the products. And I think it's just the brands and retailers that are going to win are really going to want to get ahead of this and just give agents enough data to feast on. And so one, one change we've made in our business obviously to satisfy this is, you know, for a long time when we've had our on site application layer, we've been working with embeddings which aren't LLM friendly, right. An LLM can't like, you know, without a few caveats, but can't really read embeddings. It needs to be text. So we are now supplying brands with, you know, they can have six or seven rough attributes with currently with things like you know, side sip and lining and I don't know, dry clean and like things, things that aren't really necessarily going to make you buy it. And we're adding like 60 plus taste tags on top of that which are agent interpretable versions of our embeddings. So like a score of low for, you know, status seeking, a score for a higher score for, you know, romanticism, playfulness, femininity, all, all of these different qualities we're giving different scores and then it can actually make a decision among them based on what's understood about the, about the person, different contexts that they would wear it and different mental states they would begin to want to buy it, which is like so overlooked. So we're teaching them consumerism, teaching the agents to be better consumers. Filling in the blanks, giving them, giving the products more personality again, yeah, all the broader context. So is that, is that data that, let's call it psychographic preference data, is that going to be a defensible moat for, for who? For the merchant. Whoever, whoever does best here, like the whoever has the best data, they're definitely going to, going to win. And I think from there like two retailers have the identical data. They both use us and they both have fantastic data. You know, from there it's the other things, right? Are they shipping quickly? Are there other things, you know that are they're causing one will convert better than the other and then the agent will make sense from that. So the answer here is the, the merchant with the better data is going to win. All of the things being equal price, availability, shipping. The merchant with the better data is going to win. And you, and you give them that. Yes, a hundred percent. Well, McKinsey found 71% of merchants say AI merchandising tools have had no effect on their business. So far, seven in 10 merchants spent money on AI that didn't move the needle. Now in your conversations, especially at Shop Talk this year, what was that the root cause of that? Was it the bad. Was it bad tech or was it bad questions? Terrible attack from what I'm understanding. And it's just completely like a misnomer for all of these things. Like AI merchandising. If we're going to like really pick it apart, should mean that an AI decides where product goes. No one is currently doing that, right, except us in the sense that like category pages, right, where you have the bulk of the product, everyone still sees the same thing. And let's say you psyche, which we do fully per user, fully real time ranking, which is how you know, a lot of big tech works, right? Like your Instagram feed is everything is being ranked. Your, our LinkedIn feeds are being ranked based on what we're doing. So we're bringing that kind of dynamism to the merchandising, which is where all of the products appear. So I mean unless you're meeting that definition, you haven't been AI merchandising. And I think that unfortunately, you know, when you look under, I think the feedback that I've heard from retailers are like, you know, we were sold this, but we're frankly having to manually set it, we're having to configure it, we're having to do all of these things and you know, to be, to be really diplomatic about it. Again, like these companies are trying to solve a need. And they're not product experts either. They're not consumer product experts, they're not fashion or furniture or jewelry or grocery experts. But I think that retailers expect that to expect they Expect that to come from the vendor. The vendor expects that with their tools and the data from the retailer that a pattern will emerge. And it has not emerged. The macro has not emerged because the substrate isn't there. And that's the data around the product. Data. Yeah. Which you can extrapolate back to the consumer in a way that's much more meaningful than, okay, well John's 47 and he's bought this before. What shall we push him? Yeah, just age. It's very one dimensional personalization as it is, as it's been. It's just looking at one particular. Yeah. Segment of that product or that search instead of 50. Yeah, well. Or 700 to a thousand. 700 to a thousand. Yeah. Which I think. Well, that's AI, I guess. Yeah, I mean like, yeah, the more multi. Multidimensionality isn't always good. Sometimes things can be noisy. But generally, yes, the more dimensions the better. But I think you nailed it with that word. It's a lack of dimensionality. And I think it's again in our sales calls where people are asking like, what are you optimizing for? It has to be, okay, this is a discount shopper. Okay, great. But you've got like 2,000 products on discount at least. What now? You know, so it's just, it's thinking in a more dimensional way. So when you've got an AI that's looking at 700, 700 different dimensions of a search by a consumer is there, what actually is the agent that's doing this search for the customer going to latch onto? Is it going to be, hey, I found the most data at this merchant, so I want to go back there because I was hungry and this gave me the most to feed on. Or is it going to go well, you know, we don't really need to know that much of what this product was. It was a lot easier at this other merchant. Right, good. Great question. Okay, so here's where we're getting into the nitty gritty of it. Problem one is the data needs to exist. So let's just say, you know, the catalog has all of our enriched data. And you know that, you know, this is for people spending this much that are looking for this for this occasion. They're a little bit anxious and this is for these people. But then you need to really get it right. You still need like a ranking or a decision or ranking or a decision layer. So exactly what we do for on site, that sits like an endpoint for the agent to then be able to rank to choose the best one. So let's just say you search for what's something you would buy a clothing item just to make it easy. I am big into polos that have embroidered features on them. I like em. Yep. Okay. So embroidered polos and a zipper. And I like a zipper. I don't have one today but I like zippers on my polo. And those are all, you know, straightforward language, language friendly things. And you say that and you say you know you might be wearing it like here in the next few months. You give it as much details as possible but then it goes to Nordstrom and there are still like let's just say 210 golf shirts with embroiderers. So psyche sits sort of between the LLM and the retailer to do the ranking for the agent for you, what is the best shirt for you which is the one that you're most likely going to buy. Right. Because at the end of the day it's probability math based on the fact that this has happened and you have said this and this is what you've done in the past. So it's that same ranking and then the agent is just a different interface. So we're doing the same thing for the agentix side that we do on the site. You're giving it what it wants. You're giving the, you're giving it the data that it wants. It's helping make the decision. Yeah, it's the decision layer. It's the brain. It's the thing that does amongst everything that we've considered here. This one or these five. Right. Because you're still even like with G go can get you to like the top hundred. You want, you want to show five. Right. Or three. That is not old school personalization. That's a little bit different. It's a little bit more complex. It's the real deal. It's the real deal. It's the real deal. So you know, as, as a shopaholic, self admitted and somebody who actually enjoys going to stores and seeing things and touching things. Every AI shopping journey now we're optimizing for lowest price CIS delivery. Are we accidentally turning commerce into procurement? I mean where, where, where does the surprise and delight come from in that? Yeah, I mean I'm a big believer that you know, if we look at how big yeah. Shopping was an activity, you go and you discover there's the spontaneity. We definitely like that. And I, I do, I have been really concerned about the fact that no one's thinking about like how do we make agents proactively surface things you're not necessarily searching for. Because one, I think we're subconsciously in the market for like at least 30 things at any given time that we're not actively thinking about. Like, you know, I want to replace my old gy. It's. I want to, I don't know, find like flip flops for next summer that I don't hate when I go to the pool. Whatever we're thinking about, right. But like, they're fleeting thoughts sometimes. So there's that and then there's things that you like, aren't thinking of at all, like that, that just feel very you when you see them. And so with our, our data, our deeper data actually allows you to, to do that because you're not necessarily, like, you can find like a new product comes into, into play, and then you can find a strong match mathematically in the vector space between the user and the consumer. When you say, we think you would like this, it's actually based on very highly dimensional data instead of, you bought something from, I don't know, Brooks Brothers. Here's everything else from Brooks Brothers Optimized for brand, which is how most tools work right now. It's actually like, okay, he really liked that golf shirt that had that big panel. He liked the contrast of it. And here's a shoe that has, like, piping that creates the same contrast. And a soul. And a soul that is spongy. Like, they like a soul. And it's got. Doesn't have too many holes drilled through it because they don't like too many holes and so on and so forth. So you're, what you're talking about here is everybody's been trying to optimize for intent. And what we're talking about here, since it's so multidimensional, is aspiration. What. What would that agent aspire to? What would, what would a. What would appeal to that agent? Not just what are they looking for right now? What would appeal to them? That's, again, that's a completely different kind of AI, Completely different kind of an intelligence to give a merchant. It's like. Well, I mean, that's where thesis. This is where the thesis always started, right? I mean, if you're working in luxury fashion, you're seeing people who, like, are buying, you know, insane things at insane prices. And not all these people necessarily have the means, right? But they're like kind of the average bag, Birkin bags. It's like, I'm gonna. Are you kidding me? And talk about that. But like, the average you know, the most cliche example ever is, you know, no one particularly wealthy sees a crazy pair of shoes as I have to have them. And this has always been explained away by, like, status seeking. And it really isn't. It isn't that thing about the design, you feel in alignment, you feel like it's gonna make you feel more like you. And I thought that was psychologically very interesting. But no one talks about it. Right. You know, they're not buying it because of the craftsmanship and the petite and whatever other stuff like the heritage brands are touting. Right. There's just something about the design that you feel excited about. It's the same thing as when you see a city. You know, not everyone responds differently to different travel destinations. You see a city, you're like, wow, that's somewhere I'd like to go. And to some people, that's like some desolate mountain in the middle of nowhere. To some people, it's like a bustling city. There's just different reasons why you see, like, you know, a picture in Asia with like, many, many people versus, I don't know, a monastery, and it's going to. You're going to respond differently. It's the same thing for design. It's the same thing for anything. Sensory input, reaction. Where was I going with that? Yeah, but that's what it's trying to capture, right? And it's like, I think that is a very hard problem to solve unless you actually understand consumer psychology and product psychology, because it's much easier to just work with input, output. Okay, they're looking for cat food. Okay, here it is. Right? Like, yeah, it isn't just a search and. Search and retrieve mission. It's. It is. Consumers are consumer. They're people. They're incredibly. We're all unique. We all have our own wants, needs, desires, and what we purchase is, is a reflection of that. It's not necessity. A lot of it is what we aspire to, what we're in the mood for. And, you know, as someone who's never made an a, an intent, an unintended purchase in my life, I don't know much about this, but I'm just going to go on a limb here and say being able to give a merchant or a purchaser at a retailer the data that Psyche has about. Look, you're looking at building out your line for next year. We actually have the data to tell you what's going to appeal to our customers and what's going to appeal to the market. And that is revolutionary. That's Amazing. I have never heard of that before. That's incredible. You can actually say, okay, you know, this is what our. This is what people aspire to when they come to our brand, when they come to our shop. Yeah. And it's, it's reminding me of a quote that I once came across. It said, I can't remember who said it must have been designer, but like, style is your style or your taste is basically a reflection of your most deeply held values. A similar one is your style and your taste is, is basically everything that ever happened to you. Right. Or your inside, on your outside. And I think if you start thinking about it that way, then you're not thinking so much about color and trends and surface. But it's like, well, what happened to. I mean, even. Cause I, I tend to not talk about trends much. I think there's just too much focus on it because we all respond to different trends. But if you think about what creates trends, there's something that happened in the world that has shifted our values and that is the reason that trend exists. And if you can understand that, then you're not just reacting to like Brown, you're reacting to equality. Then you can then extrapolate for much better future prediction than just looking at a color. Well, you know, you've developed the Ocean framework, which is really cool. And essentially, you know, it is, it's a personality. It personality traits that are then mapped to a product and essentially give you the. I don't know if I. If I'm going to use this term. Right. The copy to expand or to create the data that build the story, the context around a product. Tell that product story. Walk me through how, how Ocean works. Yeah. So Ocean, or the big five or the five factor model is the most robust predictive scientific model in personality psychology. And that just means that, you know, personality scientists and psychologists that have tried to reduce us since the 50s of our like, most basic traits that they can actually measure. And like, you know, across populations and across a lifespan, this has been always kind of the most scientifically sound. And their openness. So that these are scales, by the way. It's not like you are, you aren't, but you sit like when you do the test, you, you end up from 1 to 100 on each of these scales for openness, conscientiousness, extroversion, agreeableness and neuroticism. And so if you were to do the test, you're. You end up with a really granular score, like a fingerprint, if you will. Right. If you're like, you know, if we make it from one to five, just, just for ease. If you're like a 44424. Anyway, there are like over 3,000 combinations if you look at it that way. So it's not as simple as saying he's an extrovert, he likes bright colors. But how, how they interact, like how, how they interact together within that fingerprint is really kind of like a very specific mix of those qualities. Right. And yeah, and in, in the world, I mean in academia we know that there's really strong correlations between where you people sit on these traits and all kinds of interesting outcomes. Right. From political and health and relationship driven outcomes down to like food and music preferences. Right. So here I was like, okay, these, these are actually like the ground truth. The ground truth, real world drivers of this level of taste. There's obviously everything else, income levels and occasion and lifestyle and everything else we focused on that's more obvious. But this is like that core layer that makes you pick the, the A over B to a great degree. Okay. So one thing almost no one else in the market has cracked is you track guest users. So you're able to see the folks that are not cookied, that are not logged in, that have not made purchases and that's actually the majority of E commerce in for any store is those folks that just show up. How do you build a taste profile from, from some, from a cold start when you don't know anything about somebody before they've shown up and how many times do they have to actually come back to that merchant's page before the signal they give is meaningful enough for you to optimize the site for them? Yeah. So continuing on for, from explaining Ocean, the way we get it on the product is, is the answer to this question. Right. So we spend about, I mean it's ongoing. Right. Like R and D is a big part of what we do to create this, you know, what I'm calling a world model of taste because it's capturing what's actually happening between consumers and products as they interact. But yeah, years of research has gone into the aesthetic and surface attributes that map out to the qualities. And so when we ingest a catalog, all of that lives in the product. So we bypass needing to understand who the consumer is and can personalize right away for guest users through what they're interacting with. And you know, we track granular live session interactions. Right. Like hovers, ignores which a lot of people miss. Right. So you're starting to get a sense of Aversions. What's sort of interesting, what's so even from like one going down, one pop, or like interacting with the homepage, we very quickly get like enough signal that by the time they go on page two, and that can just be in 10 seconds, we re rank the entire catalog based on those signals. Like I always compare it to a really good salesperson on the shop floor at like Bergdorf's, basically observing someone on the shop floor and they're like, are they looking at the price tag? Do they seem like there's intent? Or they just disagree discovering they can see them, they get a sense of who that person is right away, intuitively. So what you're talking about is real time personal. It's real time and personalization. A lot of search engines claim to do that and they don't. So how do you tell the difference? Well, it's, it's not up. Yeah, again, it is personalization because we are doing it per person. Right. Like it's a, it's a pretty personality person personalization. It's not a coincidence. Right. Like you. I mean the, the nice thing about it is a lot of retailers have come to the conclusion that they've been sold this, they've done length integrations, they've configured it more than they thought they ever needed to. And it all looks the same, right. So they sit with their co workers and they're like, what do you see? What do you see? And nothing's changing, nothing's moving, right. So like, you know what a lot of. Yeah. Like it irks me that a lot of these companies are selling dynamic merchandising when it's not. They're like one product moved in the last like two months from like position from one row to another row. Right. Whereas like we re rank in real time from the second you go on the site. That's dynamic merchandising. And I think, I don't know, attribution. Obviously there has to be a clever way to, or a robust, clear way to prove attribution. So I think a lot of other companies as well will just kind of be everywhere. Then they say they've made this company a ton of money, but there's no, there's no control group. There's no way to see how the retailer would have done without their technology. Could have been better. So we always like route traffics where we show and prove that everything is the same except the rank order. So RAI sold the product because it got it, it moved it from page 71 to page 2 in that session. What you're saying is if I have two people sitting looking at the same catalog and they both browse different items on the first page of that catalog and then they both go to page two, if page two is the same for both of them, that is not real time optimized. No. Personalized. Well, neither, frankly. Yes. No, it's not. How can it be if it's the same? Right? Yeah. If I go on and I'm ignoring like all the floral prints and I hover on a few black and white things, open a few black white things, hover in a few gray things, that's enough to tell you like, that you should push away the florals and like move the neutrals up and vice versa. So if you're a merchant and you're evaluating your next search tool, that's an experiment for you to do with your peer, is both you visit a site that's using whatever tool you're evaluating purchasing and both have a different interaction on the landing page and then see what happens on page two. And if nothing remarkable happens on page two, after you've looked at different products on page one, you know, you're looking at a product that does not actually real time optimize. And I'll add that one of the reasons that this hasn't happened is if you only have surface attributes like color and brand. I think that, you know, retailers don't want their merchandising to be heavy handed. You don't want, because you click on one black thing, everything on page two to be black. But if that's all the AI knows, it doesn't know what else to show. So it's like with multidimensional embeddings, it can understand that more deeply. Right. So if you clicked on something that's. I know we've. In another conversation you mentioned you like printed pants. If you like printed pants and you found one on the first page, you're like, oh, neat. You know, you opened a tablet. It's okay for people to know that about me. It's all right. Sorry, you told me, you told me that in Instagram. It's okay, it's okay. It wasn't incompetence, it's okay. Okay. But you don't necessarily now want all of page three to be crazy prints. You'd be like, whoa, okay, this is a lot like, you know, so you. But we'd understand the quality of that item that you liked and we would diversify it. So it could be like shoe, a sock, not necessarily prints, but Things that have a touch of exuberance to reflect who you are. So because you have deeper data, you don't need to bombard people with just like one of the dimensions, if you will. You know, it's kind of like on Facebook you read one article about the Middle east and then it's just like it's over or so. If you're going with site. If I'm going to a site with my, my favorite target of, you know, aviation themed polos with zippers. As we discussed, how many attributes are on the average merchant's page describing that, that an AI agent would find? Well, I mean I think there's the image, the name and the description is sort of like where they go and then like the details. And so a lot of, a lot is gleaned from the, from obviously the name and the image. But then descriptions aren't like people just aren't investing in descriptions. They're actually auto generated. They're, they're generating them and they've just gotten really bad. Again, everything is comfy and stylish, so it doesn't actually differentiate one thing from another at all. Everything is like comfy and stylish for a smart casual event. But this is basically the description isn't really, it's noise is the point. And then the details are just really kind of, you know, I think if you're looking for something functionally, like it ticks a box, right. If you definitely want something that is machine washable or you know, hypoallergenic, that's fine, that's in the details. But that doesn't really help solve the taste gap and the decision making the decision layer when there are hundreds of things that fit the function tick. Right. So yeah, there aren't many actual attributes. It's just sort of using all of these multimodal. Right. It's just looking at the description and it's saying this is the description. And then I've got size, color, you know, those things. Now let's compare that to, to what you're doing and how many more, how much more data are you adding to that product? In the package we sold recently at least 60 taste tags that we're adding to, I don't know, like two. So a lot more. 60, 60, 60, 60. And some of these things would be like drape or nubbly or the kind of, I don't know, you know, we definitely clean up everything functional to make sure it's there. Like if you know a brand, brands tend to call their products really weird things. We'll make sure it has, like, all the possible colors, a color family. So, yeah, we go to town on the physical, but it's more the contextual and the psychographic that people find enriching. Right. Like, so it's nice, like a very conservative polo jacket that comes to mind that we did. Not only are you putting, like, you're putting high and low scores for all the different qualities that I mentioned. Right? Like, is it statement, is it status seeking, is it romantic, is it feminine, is it playful, is it modest? It's, you know, what does it focus on? So you're giving, like, scores for all of that. You're. You're seeing a personality profile. So you're saying, you know, it was a very simple jacket. So you're saying, like, grounded, low drama. Like how you would describe the person who wears it. Right. Like grounded, low drama, dependable, wholesome, conscientious, all of those things. Right. And that helps. It's not necessarily someone has to type that out, and that's not visible to the consumer, but it's how the agent understands who this is for. And that's the other piece. It needs to know that you're going to like that one because. Because you are dutiful and wholesome and conscientious and all these things. And then we give a bunch of other context things, like anywhere you'd want to wear it, right? Like, so very specific things. Right. Like on the way to the airport for spring break. Not just like, brunch with friends is like, endless example. You'll put things like Carolyn Bessette, Kennedy, Gwyneth paltrow in the 90s, like, we'll put everything that might be related to that product. And then think of how much that surface area now expands. Now you have some hard numbers, obviously, from some of your deployments. You've deployed with Kima Zabeti, with Toys R Us, with Anne Klein. Can you walk me through one case study where the data really shocked you, not just the revenue, but what the behavioral signals revealed about how the shoppers were actually behaving when you showed them the right stuff? Yeah, absolutely. I mean, I think that with our mono brands, we were always the most surprised because a tool like ours works well with large catalogs with lots of variants. Intuitively, people know that with brands, we always thought maybe it's pretty singular. They have their customer already and it's not a huge variety. So we thought going into some of these pilots, we're probably going to do a floor of 2 to 3% and ended up being 22% with some of them, we're like, why is it so high when the selection isn't, you know, numerically speaking, as large as some of the bigger retailers? And I think it's because a lot of brands have this dichotomy. They have a classic customer that likes the heritage stuff. Brands innovate and push out, you know, new directional things to the younger customer. And I think it's, it's dangerous for brands to alienate one over the other for different reasons. And there's always this tension, right? And with our tool, you don't have to sacrifice that tension. And so we, we created that lift because we were able to show, you know, brands have a lower tolerance for finding the right thing on a D2C site. But you know, they landed and they were able to, to see a coherent selection. Very quickly. You're showing a $16 million revenue unlock without new traffic or new SKUs or more ad spend. Why are retailers still chasing that cost of customer acquisition efficiency before they're fixing their discovery efficiency? It's just with, with acquiring new customers, it's just felt more tangible and you know, they can put, they can put their arms around it, okay, we spend this much, we get this much customers. But the math is easy. And you know, we've. No one has crack. Personalization, it's still that problem of like we've hired these data scientists, we've hired ML guy from mit, they're supposed to fix that. That's a tick. They're on it. They've assured us that now the models are going to learn the patterns and you know, it never happens or we're using, you know, we signed up with Incumbent 1 for the next two years and they've assured us that this is going to happen. But it's still that that problem of everyone expects someone else to solve the thing that's missing, the product intelligence that's required for the models to understand the patterns for personalization to happen, for revenue to increase. And that just hasn't happened because there's that gap, that gap that we're bridging. Deloitte's Global Consumer Products Outlook found that only 31%. Geez, that's pretty low. Of merchants are actively addressing how to influence those AI assisted shoppers. So 2/3 of retailers are still sitting on their hands watching agentic commerce happen and sitting in the bleachers. You know, what, what should $100 million retailer be doing in the next 90 days to avoid becoming invisible? Gentex has become as much or if not more loaded than AI. And I think that that's created a little bit of fear and fatigue or gosh, I have to think about like another channel and another format and I'm still wrapping my head around it. It's just another interface. And I think the things that you can do to improve conversion on site, which will always be important, are the same things that you can, you can do to set yourself up for success. For agents, you know, you need to improve product intelligence. If you were to work with us, we ingest the catalog, we make sure you have all the contextual data that an agent would need or that your system needs for, for the right searches to come up, for the right PLPs to rank. It's just, it's just preparing yourself and creating a core foundation for this AI commerce, whatever form it may take. The relevance problem is the relevance problem. You need to give it the right data to make the right decisions. Bain noted that third party AI agents increase market transparency and are going to give those big EDLP low cost, high speed players an advantage. Meaning retailers who can't compete with them on price or assortment or shipping or speed alone are going to need a differentiator that's attractive to those agents. You know, that rich data that they can feed on. There are recommendation engines and discovery platforms and dynamic pricing tools. Now every major cloud vendor has something with personalization pasted across their label. So when you walk into a room and a merchant says we already have a personalization vendor, what's your answer? You don't? Well, I try to understand what they mean when they say personalization because you know, it's famously been kind of everyone has their own definition of it. I like to look at what a true definition would mean and I think that personalization would mean not a set of rules. You know, I think most people just think if you do A now B happens as personalization. If you clicked on discount, you see more discount. You know, often what retailers say is with X tool we can optimize for price or color or one factor. But the way we're thinking about it, personalization only can be that you have as a person a completely unique journey that factors in all of the factors that matter. So it's not one thing or another thing that that is personalization. If it's not, it's optimization. Well, you know, we're getting going towards a world. You know, I think you mentioned once to me that Jeff Bezos said, you know, in an ideal world you're going to have one thing presented to you. And that's always going to be the right thing. But we're a long way away from there today. Where we're going towards is a world with no browsing, no carts, no checkout pages. So what is actually going to survive that brand relationship? Who's going to win? The entities that win are the ones that capture the best latent representations of people and products. And latent is like the things that you can't see. See. Right. Because the things that you can see is what we've been building on to date. Surface attributes, demographics. Jane45, burgundy weather resistant jacket. Some CEO will say this in a conference. Shopping is emotional. And everyone's like, wow, okay, yes. What does that mean? What are the implications of that? And it's just, it's emotional meaning you've got lots of things that fulfill the request from a, from a price and function standpoint. And you've got too many things. It's, it's, it's that next taste level where function and price is the same. Why you pick blah over blah. And those things are latent. You can't see them. So representation to an AI is, is how it can actually reason and decide what's right for you based on what it knows about you and what it knows about the product. And that in that gut way. What is the retail category that benefits most from psychographic merchandising? That would surprise people. I'd say grocery. You know, because it's food is. Is. We tend to think of it as very functional and biological in regards to, you know, why you like what you like there. But I think what we're forgetting is there is a lot of. We're always grocery shopping. We have our orders and then there's a vast, vast assortment. There's always new products on the market. So actually increasing latent discovery there and making connections to show people things that are hidden in siloed categories that they would discover if they were in a physical store. Because they're walking up and down the aisle and they're like, oh, keto crackers, that's cool. Knowing what to surface there proactively would bring so much added revenue to grocers. I want a new taste. I want a new sensation. Where is that? I'm going to the grocery store. What's one thing the big platforms that are building agentic checkout are still missing entirely a taste or decisioning layer to actually say, like, why this over that? Out of the 200 that are viable, why these three? Well, five years from now, does a human ever Browse a category page again? Yes. I think we'll always need to visually calibrate and those products need a home. We can do that in an agentic interface where they can show us five or 10 things, but at one point you might want to see all. And that, that's a category page. We're visual creatures. That's just the way it is. Now, if you could get one data point from every shopper that no agent currently collects, what would it be? Personality scores. Because that's the they compound best. You know, you can understand them across domains. If you know someone's personality scores, you can understand what they're more likely to buy within grocery, within furniture, cpg, clothing, all of it. Dunhambi's head of data called agentic AI systems. Toddlers learning to walk. We're a psyche on that developmental arc. I would say that's inaccurate. Agents have huge brains, but they're not trained to do specific things well. So they're untrained when it comes to commerce. They're untrained salespeople. They need the context, they need the training. Psyche provides that training, as we've said, it provides that education. You want to get good at selling furniture, you want to be the best furniture salesman. We can help. Here's everything you've ever wanted to know about furniture, people who buy it and why they buy it. Boom. Annabelle, is there anything you want to leave us with today? Yeah, and it's a bit of a philosophical one, but I think some of the more meaningful conversations I've had at, you know, at Shop Talk recently and at other events, you know, with other professionals often always goes back to look, we're all consumers. We all have individual tastes and you know, it's, it's obviously I've spent so much time, you know, in the space and neuropsychology and in product and E commerce to, to make these connections and I think just take time observing for yourself to understand your choices better. Right, like why you're picking A over B, how your, your desire to buy something, what it's influenced by, how it relates to who you are, you know, beyond the usual things of, you know, affordability and occasion and function. And once you understand it, once you start to actually see the connections, I think it can really help help your perspective in terms of understanding why this data is so important, how you evaluate you technology opportunities and vendors and you know, how to move your product going forward. The promise of client telling come to life another term from the ages to bring us to our close because it has been, as always, a fascinating and educational conversation with you. Annabelle. Thank you so much for joining us. Lots of fun. Darryl, thanks for having me. And thank you all for spending the last hour with us. We'll see you again on Martalks. Thanks for listening to Martalks, the number one podcast for e commerce and marketing applications. Be sure to subscribe wherever you listen to podcasts. And while you're at it, leave a rating and review. To find out more about how the Rosenstein Group can help you find the right right leaders for your client development teams in MarTech&E Commerce, please visit our website at Rosenstein Group.com.