Latest / The Interline Podcast / What Fashion Can Learn From Beauty’s AI Feedback Loop

Transcript

What Fashion Can Learn From Beauty’s AI Feedback Loop · The Interline Podcast

• speaker-0: Hey, welcome back to the Enderline Podcast. I've been meaning to do an episode dedicated to beauty for a while now. If you've been a reader or a listener this year, you'll have heard me and me and Grace talk about the unique market opportunity that beauty has to ship consumer devices and the opportunity it has to use hardware to get closer, literally and figuratively, to shoppers. Fashion doesn't have anything close to that outside of wearables. We haven't, at least not head-on. Talked about the fact that beauty also seems to be further ahead than fashion when it comes to AI-led personalization and AI-enabled or AI-adjacent experiences like Virtual Try-On. To be clear, VTO in fashion is having a moment thanks to AI, but broad consumer adoption for virtual try-on and fashion is still very much a work in progress, which is not the case in beauty. And while every fashion brand will tell you that personalization Is the key to increasing basket sizes and customer lifetime values. The actual levers that are available for brands to pull if they're trying to build a personalized look are fairly minimal when you think about it. You can recommend tops, bottoms, shoes, accessories, outerwear in various permutations and in different subjective fits to align with people's tastes as they stand today. But that pales in comparison to how individual everyone's skin is. It pales in comparison to the huge range of features, needs, biomarkers, and other variables that a beauty brand can use to build an extremely detailed profile with their customers. And it pales in comparison to their ability to then tailor products and routines to those customers and then track their response to products and routines over time. And that all gives you a very interesting substrate to start building AI experiences on top. Beauty is also, as you no doubt know, an extremely tightly regulated industry compared to fashion, which adds its own complexities and its own opportunities when it comes to building, refining, and scaling those kinds of AI experiences. In one sense, requiring testing for every claim, implied or explicit, means that the right foundation is there to build trust in AI answers in beauty. And not wanting to harp on too much, but Fashion isn't exactly known for being able to substantiate everything and about its products and everything that it claims. On the other hand, you don't just quickly roll a probabilistic, fuzzy technology out in an industry that has guardrails as strict as the ones that exist in beauty. So as you might expect, I'm interested in how much of all of this is destined to stay unique to the beauty industry. And how much can be ported over to fashion's ambitions for AI, or even just learn from? Fashion brands might not be about to start shipping 3D knitting machines to everyone's homes, or even putting them in every major city just yet. But they are very much trying to apply AI to the challenges of personalization and product innovation. And in both of those counts, I think there's a fair amount they could learn from beauty. So today's guest is someone who's been applying AI to both of those things for a while. As both a scientist and as the CEO of a beauty tech company. Anastasia Georgievskaya is the CEO of HOT AI. If you've been listening to the show long enough, you know I enjoy the opportunity to talk to CEOs from any background. But technical and scientific founders who run companies but stay close to the nuts and bolts are probably my favorite guests. And as you'll hear, Anastasia is both a bona fide scientist and a real operator with some well-formulated perspectives on where AI for beauty is going. And how much of it might end up being applicable to fashion. So let's hear what she had to say. Okay, Anastasia Georgievskaya, welcome to the Interline Podcast. Hello. speaker-1: And it's my pleasure to be on the podcast. Thank you for having me. speaker-0: No, no, I'm looking forward to this one. It's in a space we don't talk about nearly as much as we should. So I think I'm gonna ask you a pretty broad spectrum of questions and I appreciate you being game for me quizzing you about quite a few different things all at once. We always start these shows with a look at what our guests' day-to-day jobs actually entail. yours is an interesting prospect to me because CEO jobs are Pretty variable when it comes to how people balance their understanding and kind of hands-on immersion with the products with the everyday demands of running a company. You're also a technical, or I guess more accurately, scientific founder. So I'm interested to see how close you still are to the science and the technology compared to how much of your time is dedicated to CEO tasks. So walk me through what an average day looks like for you. speaker-1: Sure. I actually really love staying close to the science. I'm maybe not focused as much on their development side, but I really focus on applying it to real business problems. And I think as a CEO, there is a shift for me from going just can we build it? So typically this question will be answered by our R and D and data science team. And I really try to focus more on the questions such as does it solve a real problem? And the second question is, of course, how scalable is it? So how many times can we solve this problem again and again and can we turn it into a repeatable problem solving? I really love staying actively involved in the product and the you know product strategy, meaning how this technology will be rolled out, who it will be marketed to, who is indebted, ICP. Since I'm a scientist by training, I have degree in biophysics, I also like to stay engaged in the validation and actually scientific work. Our team is very actively publishing our research and I like to contribute. And I also still review some of the research papers that we are bringing to life. and I think it's it's quite maybe interesting or important to also tell a little bit of the story because I have been always interested in biology from a very young age. You know, I begged my parents to buy me a microscope. then I graduated with a degree in biophysics and bioengineering. first years of my career I worked in drug discovery and when I shifted to skincare, primarily the R and D and validation, and then I found it out with a mission to measure skin. So my typical day to day will be you know, I will wake up and I will start by setting an intention for my day, you know, what I'm trying to achieve in this day. So how it will move, how it's mission to be their frontier skincare measurement technology. When you know, I will do some of the emails using some of the AI tools. then a majority of my day will be calls. I like talking with customers. I think I spend about thirty percent of my time in interaction with customers, understanding where they want to be and how we can help them be there without technology. And then about thirty percent of the time will be you know, focused on the product and the technology, you know, what's new coming and again how we can bring it to market. speaker-0: Okay. That gives me a good idea. The second thing we do with each guest is ask them to define a word. And for you, I've picked personalization because that's one of the two central promises you're making with I've been pronouncing it hout, but I'm you pronounce it how to land on that one. personalization, I think, also means something fundamentally different in beauty to what it means in fashion. Now, in both cases, you obviously need to know as much as you can about the person you're trying to personalize for, but Styling, fit, and price are kind of the only variables you really have to play with when you're personalizing an a fashion offer, something that's made up of garments, footwear, and accessories. Beauty, on the other hand, you've you've got a really complex and completely individualized, idiosyncratic set of dials you can turn to provide a customer with a personalized experience or a custom formulation. You have to track things like efficacy over time and you have to do all of that within regulations that govern both explicit claims and implicit claims. we'll get to some of the regulations around claims and stuff in a bit, but for now I want to know how you define what you refer to as quote unquote hyper personalization and tell me a bit about what it takes to actually deliver that. speaker-1: me, personalization in beauty means putting consumers into increasingly specific demographic bucket. And that goes into understanding what someone needs now and not just what they have bought before. When I think about consumer shopping, I think there is a place of emotional shopping and rational shopping. And an emotional shopping is, you know, you go on a TikTok and you find a product and you engage with it and you think this brand and the product is culturally relevant for you and you want to buy it. But what if you are interested in buying a moisturizer and you have identified, let's say, nine products and brands and you have built, you know, this type of emotional relevance and affinity with nine of them. So how do you narrow it down to one product? And this is exactly where hyperpersonalization steps in. It's the ability to combine not only, you know, the consumer context and you know what brands we consider relevant for them culturally, we also apply product intelligence, you know, what the product is supposed to do, what are the claims, what is the intended use. We also combine the historical behavior, of course, what have you bought in the past. But now they add something that's fundamentally different, and this is the objective measurement. And the idea here is I don't believe that personalization can improve something that's not measured. Because you know there is a saying, you don't know what you don't measure. So for me, personalization is ability to combine the consumer context with the objective measurements that we are using AI to perform. speaker-0: And when you think about putting people in those increasingly granular, small buckets, how far down does that go? Do you get down to true personalization, like true one-to-one? or do you see this more as a case of getting like progressively finer brushes to paint groups of people with? speaker-1: That's a brilliant question because I think quite often personalization is identifying some demographics and then everyone within these demographics will get exactly the same product, right? What we're doing at health is that we are actually way more granular. right now we can identify 30 high-level skin features and 20, 250 skin biomarkers. And by doing that, the product recommendations and routines. are extremely refined to address this 30 high level metrics and two hundred and fifty biomarkers. So I think we go beyond just having buckets where everyone gets exactly the same product recommendations by really doing it like one to one. speaker-0: Okay. And if we think about 30 features, 250 biomarkers, there's an enormous number of permutations of those things. unlike fashion, again, I'm gonna generalize slightly, but there are no one's fit is quite as individual as that, I don't think. if you think about the like garment fit, if you think about length, if you think about compression and things, none of that gets to the same level of individualization that exists, I don't think, in beauty, when you think about all the different combinations of that. So if you combine the things that you just talked about, you you have individualized demand, I guess, is the way I would think about it, in a way that doesn't quite have a direct analog invasion. speaker-1: That's right. There is an individualized demand, but there is an additional layer of personalization measuring the outcomes and measuring the performance of individual products. So we add also the dimension of time. And I think dimension of time is quite often neglected. And the question is I have identified candidate products that will benefit this consumers, but I also need to understand what products. actually will have an impact and how they will change the skin parameters. So I think in beauty we have this additional dimension of measuring over time and knowing what was the outcome over time. speaker-0: And I think I I do want to ask you a couple of questions about the time element later on as well. Now, I've I've zeroed in a bit on the difference between fashion and beauty because my impression is that the beauty industry is further ahead in some important ways. Now, some of those leads that beauty has over fashion are things that fashion can't really replicate. So there's no there's no fashion equivalent to being able to mix foundation in someone's home with a dedicated device. The whole category of devices and hardware is something that fashion can only really kinda touch with wearables, whereas skincare and beauty have a huge spectrum of at home and consumer devices that the the industry can sell. Some of the other stuff though, and I think about virtual try on as the the primary one, is a matter of beauty maybe having an early mover advantage, but beauty also maybe having a consumer that's more accustomed to being an active participant in that process you just described of personalization and then overtime evaluation. And the simplest way I can put it is that virtual try on and beauty seems like a really mature category, whereas fashion still seems to be finding the right combination of technology and user experience. Am I reading that correctly or is there something else going on? speaker-1: that's a brilliant observation. And in my view, consumers in beauty are in fact interacting with products more directly. And to give you an example about you know mixing products in the comfort of your home, a consumer can mix two foundation shades and may and get a perfect one. And they can do it daily and maybe next day I will mix different other products and get a different foundation shade. Or If we're talking about care cosmetics, I can take my foundation and I can mix it with a serum and I will get a new product that now has foundation with skincare benefits. And again, consumer can do it more directly and they can do it at home and they can do it daily. What I also think about differences in fashion and beauty is that skin and hair using devices or just just using your mobile phone, they can be measured and they can be tracked over time. And in that sense, there is a shorter and more straightforward feedback loop. Because if I have a product and I can measure my skin, then I get recommendation, I can use a product and then I measure again to understand what's the benefit of this product. to compare it with the fashion, let's say I have thrown an outfit. Right. And I, you know, go out and maybe some friends or family will give me feedback if you like my outfit or not, but it will not necessarily happen. So I have now with all of the technology and devices in beauty, I have a more straightforward feedback loop about products enhancing or improving the skin conditions. And that's why I think maybe technology is more established in beauty because again, this feedback is more straightforward and more direct and it's measurable. speaker-0: That's a really good way of putting it. I think that captures it neatly, the feedback loop side of things. Now, I I think I mentioned earlier, the the other major difference I think between fashion and beauty is, aside from the feedback loop we just mentioned, is regulation. So in the EU, cosmetic claims need supporting evidence and scientific rigor backing them up. You can't just say things. There's also An unbreakable link between the product promise, as made in marketing copy images and in tools like the ones that you're offering, and the the testable reality. You have to be able to back up what it is that you're claiming you can do. What I can't quite figure out is whether that's a restriction or a boon, a good thing or a bad thing for AI experiences. Because on the one hand, I can kind of see companies being nervous to touch anything that's probabilistic and fuzzy. when they're trading in products that have to either do what they say with a within a well codified framework or they don't and they get removed from the shelves. On the other hand, there's a lot of uncertainty in clothing as well that's still present by the time the product reaches the market. And that's not really the case when it comes to beauty. So for instance, an LLM can retrieve data for a given beauty product. And as long as the citation is there and the data is accurate, the brand, the retailer and the consumer can all be confident that this is an accurate claim because of the regulatory environment it exists within. Not the case for fashion. Outside of the pieces that make it into digital product passports, fashion doesn't have a regulated chain of accountability and those things I've just mentioned, it's kind of an inexact science when it comes to lifecycle impact assessment and and so on in some ways, which means you can't query fashion in the same way because it's not so tightly regulated. I guess this is a very roundabout way of me asking Do you think beauty being a regulated industry helps or hinders the rollout of AI? 'Cause I think that's the other element I see as the difference between the two. speaker-1: Regulation definitely raises the bar for evidence and transparency. If we think about AI in this context, what AI can measure, for example, in clinical will be visual data, how appearances change over time. But we also can use AI to measure differences in skin structure and even chemical reactions happening in skin. And this means that in the context of regulation, we can use AI to understand the changes. But also the mechanisms behind and you know the fundamental changes in skin biology. Then once we have this information, based on the skin profiles of consumers, we can identify what products they could benefit the most, and they can make the claims to match the skin profiles and the product usage. And then what's valuable because the industry has you know, requirements and regulation to test the product. So you can just say that the product does something but and not test that. Based on the clinical data, we can also project the magnitude of effects, because these effects have to be measured before product is put on the shelf and before you can put a claim on the package. And I think that AI here doesn't really replace clinical and you know regulatory reviews, but it can support them in a very meaningful way. AI supports now their product development on many different levels, such as helping run the studies and do it more scalable way. Let's say you can go from running studies on 40 participants to collect data on thousands of participants. This is also partially what we're doing in HEUT with our clinical software, helping brands scale the studies. But you still have to run the studies, right? You can scale them, but you still have to run them on real participants. There are some techniques where you can use AI to run your studies on virtual participants, but mostly this is rooted in the real clinical data from real participants. And then we're talking about regulations and claims. AI can help you prepare claims that are claims ready, right? And what does this claims ready mean? It doesn't mean that you you have utilized an AI tool to analyze your clinical data and now this claims. will be automatically approved. This is not the case. But what AI can help you do is pr prepare data in a way that's it's ready to be audited and it is ready to be inspected. And then I think the stronger evidence what in beauty industry is moving AI just, you know, beyond being a Nobel new tool. So it's really moving it beyond novelty. By collecting more information about product performance, we also are able to collect more meaningful signals from the clinical studies. And this is I think where the core value of AI in regulation of beauty is that we can have stronger signals and more holistic data for product testings. speaker-0: Yes. And I see the value of that for the the brands themselves. I see the value of that for the labs. I see the value of that from a scientific point of view. I see the value of that for a consumer base that wants to be able to ask fuzzy natural language questions about a very complex scientific category of products and to have confidence in the answers that they get back. speaker-1: Right? it is indeed. because the consumer also wants to understand how was it tested? Was it tested on consumers like myself? what were the claims? What did they measure? and the ability to collect more data also allows the claims to be more holistic for the consumers and variety of consumers. speaker-0: Perfect. So based on the results of the AI survey that we ran this summer, which, full disclosure, fashion survey rather than a beauty survey, around nine in ten professionals tell us that they're using AI day-to-day at work. Only about a quarter of them trust its output enough to support decision making in critical areas. we see the same shortfall of trust in B2C scenarios as well. You don't have to go searching for surveys about how people feel in general about generative content and tools and advertising. And those are now the subject of labeling requirements in the EU. I want to ask you what you see in both audiences when it comes to the level of trust that people put in AI and beauty. So what you've just described is is putting AI very close to the science, as you've just just mentioned. It's not A straight shot from I'm using AI, therefore everything gets instantly approved, but it is, it does exist close to the real work. How do brands feel about AI in your experience? Do they trust it? And how do beauty consumers feel about AI? Same question. Do they trust it? speaker-1: I think their introduction of different LLMs in our daily life genuinely improves the perception of AI by consumers. About forty-five million consumers are monthly asking Chat GPT questions related to healthcare. And when Howdy surveyed consumers in the US, about 70% said that they are already using LLMs to ask for skincare advice, or they consider using these tools for asking for skincare advice. This means that there is already like an established trust in different AI tools. When it comes to their more specifics and the claims related to products, I think the question still goes to the fundamental matter of what has been measured and how was it measured, how these measurements were validated, and again, what type of demographics that has been tested on. I think Consumers have trust in AI, but they are increasingly interested in diversity and inclusivity of AI measurements or generally measurements done by their brands. And also there is a question of not only data validation, but making sure that there is ongoing monitoring of AI systems and you know, this AI systems are not just doing what they want by itself, but there is Specific guardrails applied to these tools. And finally, I think that consumers are becoming more educated in AI and they are also inspecting the claims because they don't want to have the claims that are, you know, over claims. And this is why they want to really see data that is rooted in the real measurements and they want to see the validation of this data. speaker-0: Yeah, I th I think that's right. I and I do think that anecdotally from just observing the way people interact with AI and fashion in particular, I think trust does just grow through emotion and through experience. that that seems to be a pretty universal truth. I think one of the weird parts about the ways that people interact with AI when it comes to specific industries, though, is th the gap between the way that consumers and just everyday people talk about they want from a product and the way that people creating that product talk about it. I had a series of chats about the kind of natural language frontier of e-commerce at the start of this year. And one of the big themes that came out of that was the idea that when you let people talk to software, which whether they're typing or using voice, you've just described them increasingly doing, they use a very different vocabulary than the people who designed the software and a very different one to the people who created the products. So in fashion, somebody might ask for something comfortable. They might ask ChatGPT or Claude to help them find a a comfortable outfit or a boxy fit t-shirt or they might pull in words and concepts from outside fashion, from media and from wider culture. I spoke to the CTO of Daydream and for instance, and she gave me an example of somebody asking for a revenge dress. So revenge, not a concept that exists inside traditional product design and development. No one is sitting down and using that as a tag. It might exist in marketing, but there's no kind of real crossover between that. And so what you have is people using fuzzy language to describe something that has a bunch of objective criteria. That feels like an even bigger challenge in beauty. Because you've have people's preconceived idea of what ingredients do. So I might believe something about what retinol does, for instance, or peptides. And then you have the actual literature behind it and the actual science behind it. And you also have people describing their skin in very personal ways. So those 30 features that you describe, those 250 biomarkers, I don't know if I have the right words to describe even half of those. And if I do, the way that I think about my skin, does that align with the way that that science thinks about my skin? So you have people using a lexicon that is not the same as the people developing the products. I want to know as an AI company, how do you approach that kind of translation? How do you approach that vocabulary shift between what the consumer thinks they're talking about and asking for and what the objective science says about it without accidentally pushing them in a different direction and certainly without overclaiming. speaker-1: I think that generally consumer we need to understand that consumer concern, again, in their own words, doesn't mean this is what was the clinical endpoint when the product was evaluated. And the way consumers talk about their concerns doesn't mean that the product necessarily has a claim for this concern, even though it may improve the appearance of this concern, but you know, it's not something that was claimed. In my view, it's AI task to be able to translate between those three worlds. And these three worlds are consumer and what has been clinically measured and what has been claimed. I think genuinely there is a problem in their beauty and aesthetics industry, but it has never agreed on its own vocabulary, so it's not really a problem of the fashion. and it's not just me speculating. Allergan Aesthetics published earlier this year research paper in a journal of dermatology surgery. And they have analyzed more than 900 papers and about last 20 years of being published. And they surveyed about 200 beauty providers, aesthetic providers, and about 1,000 patients. And what was quite interesting is when consumers were talking about their concerns, they were quite consistent with each other. But when they started talking to providers, the language fell apart, meaning, you know, a provider wouldn't understand what the consumer would say. And of course, AI takes its knowledge from you know, the open context, and also we know that LLMs will be searching information on sources like credits where consumers will speak about their concerns in their own words. So what Aragon did is that they built a shared vocabulary for 15 different agreed terms and they called it Skin Quality Index. And the point I want to make is that this is exactly the place where we need to have an ongoing work. To build specialized artificial intelligence. So specialized artificial intelligence for beauty or specialized artificial intelligence for dermatology and aesthetics. And we can't just get away with general purpose LLMs because we need to infuse this domain knowledge into the specific use case. For example, in HAUT we have a product called SkinChat, and the goal of SkinChat is to exactly help consumers navigate product choice and match it to the concerns in the natural language. And then It comes to, you know, identifying how consumers shop. We now have a powerful weapon, which is AI skin analysis, because by using the camera and the photograph, we can't capture what's visible. So we can capture this additional context, this visual biomarkers. And I think that's quite important because consumers don't always know what the camera knows. And there are two use cases. Sometimes consumers see some concerns on their face, but they don't know these concerns are. And maybe they go around and use the terminology that doesn't really reflect the skin concern, it causes confusion with the product recommendation. Or consumer may not even see some of the biomarkers in their face because they're so subtle or because we're just not trained to do it. So I think in the beauty industry, we can now utilize not only the context and the consumer's own words, but we can also enhance and enrich this data with measuring objectively the biomarkers. to form the skin profile and you know refine the product recommendations and choices. And as I mentioned, there is an ongoing work to build a specialized AI to translate consumers' onwards into how it connects to the clinical endpoints and product claims. speaker-0: Perfect. That's a fascinating answer. That gives me a ton of insight there. we did talk about time earlier, but I think this feels like a good time to get back into it slightly because it's not something that fashion has to deal with really outside of color fastness and product durability. You you buy clothes and you wear them and hopefully they last. It seems like one of the biggest pushes with generative AI beauty is showing people those predicted results over time, as you said. And then the Closing the feedback loop through the objective measurements, capturing real results, and then using any delta between predicted results and real results to get deeper into personalization. This feels like like the highest possible stakes for AI to me. because you're talking about predictions made based on tested claims, then being demonstrated and attested to, or not over time for within an individual, or within potentially the same personalization engine. if the product performs as expected or above. then that's an incredible flywheel for securing customer loyalty, right? If you if you show me my skin pre-product, you show me a predicted results, and then we see observed results over time and they are as good or better than I expected, I'm gonna be very loyal to that experience and I'm very loyal to that product. If it doesn't, if it underperforms, then what you end up with is undermining confidence in either the product or the AI, depending on how we frame this. Tell me how you see this playing out and what you think is next. Cause to me that seems like, as I said, just just the highest possible stakes in that if your product is great and the experience is great, then this is a a a miraculous experience for customers. If either of those pieces is missing, then it's potentially a very detrimental experience for customers. speaker-1: First of all, beauty is inherently longitudinal. If you look at a typical study for skincare products, is that it's usually at least eight weeks. And the reason behind is that skin cycle is between forty-five to sixty days. So you will see first results about after this period. And on top of that, we also have like this well-established category in beauty that's like anti-aging. Or recently it's also been called prejuvenation or, you know, longevity, but it's all around you know, maintaining your skin health and improving its resilience and you know reverting the science of skin quality deterioration. And because of this longitudinal nature of beauty, of course everyone wants to build you know towards their results. And again their the flow here is that we start from the baseline, we recommend the product, we use it, and then they want to follow up and compare what were the predictions. However, a prediction Is usually derived from a clinical study, which is an objective measurement. But it's quite important to understand that a predicted result is not a guaranteed result. And there are different reasons why it happens this way. So, first of all, skin changes are quite individual. And you also will see in the clinical studies that not every cohort will be benefiting from the product in the same way. There's always a group of good responders who really benefit from their Product and some there's a group of even non-responders who will not respond to the product. And by utilizing clinical studies, we can measure objectively on a population level, how the product performs, and then we can make a visualization of its product effect, but we're always making the disclaimer that the simulation is a projection and it's not a promise. So in that sense, I think we need to trust the measurement and objective data from the study over the prediction, because objective measurement for a population is something that's historical and it's real, and the prediction is in a way approximation of what effect can be achieved. speaker-0: Yes. Okay. And and the the thing that we are talking about that we can do additionally nowadays is that objective measurement on an individualized basis rather than and and in an unstructured ad hoc kind of population or or cohort which is just based on products sold and experiences had versus a tightly controlled study. So i rather than sort of saying, okay, I can see there is a an end-to-end study here that measured efficacy response and so on. then you could do it on a personal basis and have them say, Okay, this was my skin pre, this is my skin after. and that to me seems like a big part of the personalization feedback loop that you talked about. speaker-1: That's correct. We can now on an individual level make a projection of how face could change after utilizing a product. And I think another benefit that we need to talk about with using AI is before we would have the claims in the format of numbers. For example, I would say, Ben, if you use this product, your wrinkles will improve by twenty one percent. But you may ask me back, how would twenty five percent be different from twenty-one? It's very hard for consumer to envision this change. However, if we're utilizing generative AI and skin GPT technology that has been developed by HAUT, we can visualize it and this is much more comprehensive for consumer, and it sets up the right expectation. And expectation is what builds the motivation, and it also sets up the expectation management because you can project what's realistic result, not to feel frustrated and disappointed if you, you know, don't see a dramatic change because it doesn't mean that your skin is not improving. And again, beauty is longitudinal. So you need to use a product long enough. And I think skin GPT sets up this expectation what results you could achieve based on the clinical data. speaker-0: Now, if we think slightly longer term with some of the personalization that we've just talked about, I'm I'm keen to see how that plays out and who the beneficiary is. Now, I'm paraphrasing slightly, but the in fashion at least, the current growth environment, according to business of fashion research and data is very close to being a zero-sum game, in that you can grow your market share as a brand by securing it for someone else, but you can't really grow the overall pie. There's there there's not absolute market growth happening. At the same time, fashion shoppers are more fickle than ever when it comes to brand affinity. People judge individual products on their own merits and have relatively little brand loyalty over time. If we think about everything we've just described, the short feedback loop, hyperpersonalization, let's assume we build that kind of trust in what AI can do there. Is the long-term beneficiary the brand or the consumer independent of the brand ecosystem? And I'll explain what I mean because I can picture a world where Every beauty consumer has their own shopper passport or ID with a detailed skin profile, all those features and biomarkers you talked about, the time dimension you talked about. And that that passport is portable in the sense that they can take it and use it across different brands storefronts or different retailer selections and not lose that history, that time element we've talked about. Maybe that passport is also portable. you know, informed by intent and memory from their LLM of choice. Maybe it's provided by a third party. I don't know. But then I can see a very different world where hyper-personalization exists within the walls of a single brand. So I buy my skincare or my cosmetics from a particular brand and I get personalization within their walls, but not beyond. And if I move away, I lose that personalization. I lose that visibility and that distinction between 21% and 25%. I guess I can see lock in here. And the idea of lock in in beauty is extremely strong because of that longitudinal element you talked about. People want to the results, the positive results they've had to continue. What's your bet here? Where do you see this playing out? Is the consumer the primary beneficiary here? Is it the brand? Is it somewhere in the middle? speaker-1: First I'd like to talk a little bit about the modern consumer and how they shop for skincare. I think the modern consumer wants to see claims, but they also want to see the results. And their sentiment is that consumers actually do not trust brands and they don't trust the brand marketing. And I will give you statistics from Capstone that only about two percent of consumers see brand marketing as their most trustworthy source of information. And I think by utilizing AI and skin measurement and doing it on ongoing longitudinal basis, the consumer is benefiting by getting an increasingly useful understanding of themselves. This means what are their skin perimeters, how their skin responds to different products, how does it respond to different you know lifestyle changes? And they can understand indeed what ingredients will respond the best and they can understand better you know what type of product categories we should be targeting what we see the most effect. And then it comes to the brands. I think that for brands there will be winning by understanding what channels they need to be in. Because AI technology can be deployed in multiple channels. Consumers are increasingly shopping in TikTok. We're utilizing websites as source of information and source of additional insights about the products, but they then will go to YouTube to also maybe look for derumatologist reviews for these products. Meaning that brands need to be in multiple channels. And I think a way to capture information about all of these different channels where their consumers are is by utilizing AI. So I would say it's somewhere in the middle where consumers are benefiting by getting an increasingly better understanding about themselves and what works for them. And for brands they get more insight and information about where they need to be to meet their consumers. speaker-0: And if we think about AI as an intermediary there, so you mentioned earlier the people in general spend more time interacting with ChatGPT and so on. that seems like an additional channel. so what do you think it takes as a brand to show up in an AI conversation here? So let's say for instance, I'm sitting down and completely de novo, I'm asking ChatGPT about my skin concerns. How much of that gets eventually handed over as intent to a brand in the through that sort of channel? How much of that is a brand marketing opportunity? Because this is definitely an area that fashion is spending a lot of time on at the moment is what you would refer to as like generative engine optimization or answer engine optimization. And I'm keen to see your experience of how beauty is approaching that. speaker-1: I think at the moment, generative engine optimization and LOM C beauty is still a prisonable topic. And in a way, a lot of what we will be discussing and their general conversation about LLM C beauty will be type of hypothesis and maybe to an extent speculations, and we'll the time will show us what actually was right and wrong. I think the practical takeaway for brands will be that. You can't just do nothing, right? There is you don't know exactly what to do. There is no playbook and there are no rules. But not doing anything, I think, is the worst strategy at the moment. So and we and we see different ways to approach LLM partnerships. there was a big announcement between L'Oreal and OpenAI in June this year about the strategic partnership between these two companies. And also, I think an interesting news recently is partnership between Asteloader and Profound. I think brands definitely are treating LLMs as the new channel and quite a strong one. And perhaps maybe Agentic shopping is not as popular right now. but definitely this is where the future is. And consumers will utilize LLMs to shop on increasingly Bigger volumes. speaker-0: Now, so far we've talked about AI interactions and AI use cases with a couple of parties. We talked about the brand, we've talked about the consumer, we've talked about the AI itself. I think there's this there's an additional person here. I know you still see the role for human advisors and clinicians as important in these interactions in the future. Tell me what you think the human advisor's role is now and tell me how you think that's going to change over the next few years. speaker-1: I think humans' role in beauty, aesthetics, and wellness is extremely important. They bring the context, judgment, and you know, empathy and relationship with the customers. And specifically, consumers trust professionals. They trust professionals over LLM, they will trust professionals over AI. We don't know how l how long it will last. I think it will stay because we want this emotional connection. And I see AI as giving the scale. To the beauty professionals, because for example, you can't deploy dermatology expertise digitally because there's just a limited number of dermatologists. But what you can do is provide and democratize access to skincare to those who maybe cannot say a dermatologist, or you know, maybe they don't have time or they don't have resources to do that. So I think in a way AI will augment this human expertise. It will take away some of the routine work as it does a lot in other industries like legal or let's say logistics. and then it will leave more time for this human connection and interaction, which is consumers not necessarily only shopping functional, right? They they may want to shop for beauty for and win their mind and give a moment of peace. And this is exactly where human relationship will step in. And I think that beauty professionals are adopting AI pretty fast. You will if you go to a clinic you quite often will find some type of face scan, you know, to help explain to consumers their skin concerns. but is also ongoing demand for tools to upskill and educate the beauty associates, for example, in retail stores or in different spas and boutiques. And I think that essentially AI will also give better information for advisors to help them act on this information. And if we think about research, I think the bigger val one of the biggest benefits is again objective measurement and the scale that AI can give to the professionals. speaker-0: Mm-hmm. That's a really good answer. I think there's a very justified place for people in there. Sometimes those answers are a little unclear about the role of people, but I think then beauty has a very clear one. Final question. So when we run down the idea of personalization, when you run it sufficiently far, you can there's two pieces to it. So I think fashion's learned this the hard way. Is you can optimize selling things by personalizing, right? You can optimize conversions without. really meaningfully doing anything for the final product experience or in fashion's case for returns. And that comes about because there's a difference between creating new experiences and new ways to be sold things downstream and then actually using that data to change the next cycle of product development, to actually make better products that fit consumers individualized or or broader bucket profiles. Your other pillar. Aside from hyperpersonalization, is the use of AI for product innovation. So walk me through an ideal example of where a brand has learned something from these kinds of consumer interactions and personalization steps, and has then taken those findings into either optimizing existing product mixes or researching and introducing new ones, because that seems like the bit we haven't touched on yet. And it seems like where things head afterwards. speaker-1: I think with how it we have with two loops, consumer loop and the RD loop, and they are in fact interconnected and our customers typically will utilize learnings from consumers in the RD and vice versa. So in the consumer loop, we will measure, scan, understand, recommend, and then if there is follow-up measurement, we will also learn what was the outcome. But just by making one single measurement, we're also learning about where Audience, the skin profiles, and what is the demand on the market for which type of products. The RD loop is also measuring the population, analyzing, you know, learning what's the product effect. When you also can use this data to improve the product. But by combining the clinical data about product performance and demand from consumers based on what are their skin profiles, you can also identify what are the new niches that you need to target with your products. So, what are these unique skin profiles that have not been addressed by your consumers? What you then can do is run a new study that is targeted for this new consumer profile. And you can generate a data. And what's valuable at clinical data is that you can also use it to target new channels, such as LLMs. A good example is that for beauty, it looks like LLMs are favoring. Brands that have clinical evidence and they have a clinical backing. So by being able to provide the clinical data about the product performance, you potentially can show up better in their LLMs. And this means that you not only using AI to, you know, improve your existing product catalogs, but you in fact can identify what new products need to enter into new product development and probably your NPT department will be very happy to get this intel. So in a way we are learning both from consumer and by R D by hopefully creating better products for the audience that has a demand for this product. speaker-0: Perfect. Very final question. so we'll the Internal be doing an AI report again in well, spring summer of twenty twenty seven, let's say, and without putting too fine a timeline on it. If I wanted to pay attention to something that's happening with AI for beauty that maybe isn't currently earning the headlines that it should, or that hasn't quite percolated to the top of people's attention, what should I be paying attention to? What do you think is gonna be a significant force in AI for beauty by, let's say, next summer. speaker-1: I think in the consumer where longitudinal measurements is still undervalued. If you think about it, there is nothing like overall for skin, where you can track your skin changes daily, weekly and annually. And I think we will see a breakthrough in beauty where we'll start collecting much more evidence or which will be longitudinal in its nature. And we will utilize AI to change to track these changes over time. speaker-0: Okay. And from the brand side as well, 'cause there's a good that's a good answer for the consumer. How about from the brand perspective? Is there anything that is deeply happening with AI that we should be paying attention to? speaker-1: I think AI is driving the new ingredients development and there will be ongoing release of new ingredients with new claims and with new performance. And I think the whole industry is moving towards not just products that have emotional connection with consumers but have very strong clinicals. So I think this will be one of the bigger transformation that consumers will increasingly expect. more performance, more claims in new ingredients, and AI will play a big role in that by first helping discover these ingredients and then also measure their performance of different audiences and having better marketing who to target with these new products. speaker-0: Okay. So there's AI embedded in a much bigger loop there, which is gonna be fascinating to see how that plays out. Anastasia, thanks for letting me quizzing on so many different things today. it's been a pleasure talking to you. speaker-1: Thank you so much, Ben. speaker-0: And that's the end of my conversation with Anastasia. Like I promised up front, she has a real scientist's perspective on things. I don't just mean where AI is being incorporated into the scientific discovery and development process. But where it crosses over with some of the biggest things the beauty industry wants to accomplish in its relationship with consumers, many of which are things that fashion wants to accomplish in its relationship with consumers as well. I hope you've enjoyed this one. I certainly did. We have a very different topic next week, and Grace and I will be back on Tuesday with the edit as well. So I'll talk to you again really soon.