COSMOFACTORY

63 - Virtual Skin Modeling, featuring Haut AI CEO and Co-Founder Anastasia Georgievskaya
Deanna: [00:00:00] This episode is about machine vision and anonymized data. It's about skin analysis software, about virtual tryon technology, AI supported R and D, and about how tech providers are establishing accuracy, accountability, and gaining consumer trust. Today on the CosmoFactory Podcast, I'm joined by Anastasia Georgievskaya, CEO, and co-founder of Haut AI. Anastasia, welcome to CosmoFactory [00:01:00] .
Anastasia: Thank you so much for having me, Deanna, and I'm excited to be here.
Deanna: Of course, I'm, I'm happy to be speaking with you how, uh, AI has been in business since I believe 2018, and currently offers two tech tools for beauty makers. The first you launched is AI skin analysis, and this gathers information from an image and then gives product recommendations. Beauty, this type of tech is not new, right?
There are many companies around the world offering similar solutions, uh, to beauty brands of every size. I'm wondering what the end consumer might notice. Um. Or benefit from if the brand they're shopping with uses a how AI solution. And I guess what I'm really asking here is, is what makes tech from one provider different from tech, from another provider? Um, I'm not asking you to say anything nasty about your competitors, but I, I really wanna understand how nuanced and tailored skin analysis tech can be these days.
Anastasia: [00:02:00] Yeah, I think that's an excellent question and I totally agree with that. When Skin Nails had just appeared, it was just enough to have it because it's exciting, it's new, it has a wild effect. Let me look at this nice new shiny thing. Since skin L, this is everywhere right now. It's not novel anymore. And you can also go to charge GPT and without even.
Using tool of any brand, get your skin analysis or, you know, color analysis and style analysis. I think what makes a good tech is first credibility, uh, the authority of the technology and the brand who is implementing this technology in delivering. Uh. Information, insights and data that has been validated.
So I think this component of having a human in the loop and having technology, which essentially is rooted in science and has biological relevance, this is what feels [00:03:00] different for consumer. Because if I'm trying attack. It just doesn't really hit because there's some inconsistencies or insights are very generic or recommendations feel very off.
You know, your skin type is predicted wrong. Your skin cells predicted wrong. You instantly, you're gonna feel okay. Why should I be using the skin analysis? It essentially is not just better when I look in the mirror myself. It doesn't really give me any insight. Doesn't give me really any benefits. So I think what makes a technologies different is how techno, how active technology is.
Second thing is not only the technology itself, but all of the consumer experience. Because a great technology in the time that everyone has ai, essentially no one has competitive advantage because everyone uses ai. So what makes a good product and what makes a good provider, in my opinion, is ability to integrate technology in a very flexible way.
So you can build. [00:04:00] Very seamless, very smooth, very consistent because a modern beauty brand has multiple media channels, multiple sales channels. So if your use of AI is very constrained, you know, you can only do specific things, one specific flow. It'll not work for a brand because every brand has different strategy and everyone has different media, uh, presence and presence in different sales channels.
Deanna: Excellent. Thank you. I wanna think about, um, the second technology. That I know about from your company, which is what we might think about, um, as really virtual tryon for skincare or skincare treatments in particular. And if, if I'm get learning my ai right, I think this sort of tech actually relies on what we think of as predictive ai. Can you tell us about the solution that you call skin, GPT?
Anastasia: Yes, absolutely. So skin, GPT is house AI generated AI engine, which. Is capable of simulating or projecting [00:05:00] effects of different interventions, and the type of interventions that skin GPT supports include raw ingredients or finished products or specific effects, let's say sun damage or particle matter pollution or aging.
In addition to that, you can also combine all of this effects to make multimodal effects. But technology was actually inspired by clinical trials, uh, and analysis of consumer studies of products where we saw that first of all, products work. The question is, are you using them consistently and are you using them long enough?
But even within this clinical study. We saw that consumers are not benefiting equally from the treatments. It's not an about concept. I think it's very well studied in pharmaceutical field where we know that some patients or you know, some respondents just will be more, uh, accepted to the [00:06:00] treatment. You know, they will just benefit more with a similar.
Think in skincare and we thought, why can't we then use AI to predict the outcomes of their specific study based on the clinical trial. We can learn about how the intervention impacts skin and how the cohort was benefiting from this treatment. And then we can apply this effect to essentially and you face.
Uh, for me personally, skin GPT is an amazing tool because I think that still. In the beauty space, there's not enough education about what products do, what ingredients do, how they, how they are supposed to work. You know, is it, is it fair to expect that I will be using this treatment and in two weeks I will look five years younger?
My spots count will be like maybe a half, nine spots right now. Can I expect them to be free? Um, and all of this, uh. Ways of consumer education, whether [00:07:00] through, you know, guidelines, textual information. Now we finally have this visual component. Exactly. Skin can tell you based on the clinical trial data, this is what results you can expect.
This is the visual representation.
Deanna: Yeah. Yeah. No, that's wonderful. Uh, I wanna think about these technologies sort of in, in terms of how practical they are for the consumer at this point. I know, you know, it was maybe 10 years ago we heard a lot about this idea of gamification or game design as a way to attract and engage consumers. So I, I guess I'm wondering if the technologies you've described to us. Feel like something that consumers play with, um, or, or what the level of consumer trust is at this point? What, what is that balance between entertainment and usefulness?
Anastasia: Um, before I answer about consumers, I wanna say that in clinical, actually skin, GPT is used by some of our clients to. Simulate effects and, uh, conduct [00:08:00] so-called consumer perception studies. We can even answer the question, what effect is enough so that the consumer has the conviction to buy this product?
So that goes beyond just what products do because skincare not only about the effects, it's about consumer self perception on what they think others think about
Deanna: and what you are describing then is almost, and tell me if I'm wrong here, but like percent in formulation, right? How much of an ingredient maybe would belong in a product to then generate a result that the consumer is like, oh, that product works.
Anastasia: Exactly, exactly, because, um, brands, uh, I think solve it. Consumers want effective products. So I think it's also basic, but more than 80% of Gen Z will choose performance products over just the marketing message. So this changes the beauty industry and the give r and d teams with tools to keep up to the pace of [00:09:00] consumer.
But getting back to gamification, I think gamification is essentially a very good thing. If you look at a lot of the educational platforms, you know, the dwelling Vogue green os an O, it's like we, it's like funny bird. Uh, you know, when you get your words wrong, it like cries and like, uh, it rolls on the floor and it can eventually just simplify.
So I think it, uh, giving consumers tools that don't have to feel so serious, right? So just try on some ingredients, you know, and this can spark some. Ideation. Like if it, if it's a game, it's, it's, it's even okay. Like I shouldn't be even, uh, you know, feeling uncomfortable. I didn't know about how this ingredients work, right?
It's a game and I think, uh, this takes out the arrogance out of the conversation. Uh, it's okay not to be a skincare guru. It's okay not to be a ski influencer. You can just educate yourself at your own pace [00:10:00] in the format of. Conversation or, you know, previewing a couple of ingredients on your own face. So I think it's, it's, it's good to give more options for customers to discover the skincare world.
Deanna: Yeah. No, I like that because as, as you're describing it, it, it sort of takes that shame aspect out for the consumer, but I think it's a, a helpful reminder for some folks who might be highly educated in the science space and, and want to share that information with consumers that we can share it effectively.
Um, but we don't all have to sort of. Do it in the same way, right? We don't have to learn in the same way as the person sitting next to us or, or the consumer might. That's so interesting to think about, and I like that you, you know, were helping us think about, um, product development there with this sort of technology.
I, I'd love to hear more about that. How product research and development happens with a tool like skin, GPT or even, um, is it, you know, is it being used for ingredient r and d as well?
Anastasia: Uh, yeah, like, um, so the [00:11:00] skin G PT is a part of our clinical RD suite of our products. Uh, what the first step is we help our customers. Again, it could be ingredient manufacturer. Sure or could be finished. Product manufacturer, collect information about, uh, actual product performance. If this can be done using professional clinical images or over selfies.
We collect this data, it's very standardized, and then we can build prompts based on the clinical trial of, uh, this ingredient or product. After that, we can better apply this prompt on the new thesis to generate more materials, whether we want to build marketing materials, let's say royalty free materials, or you know, some materials that are like allowed for publication on our website.
And uh, the second part is, as mentioned before, is studying a perception of appearance. So. In some of our studies, we have seen that when you [00:12:00] combine effects and products actually target several key concerns, it produces more impact and more effect on consumer compared to just, let's say, treating your pigmentation and it really goes down.
It may have lower perception of, you know, the product work compared to improving pigmentation, maybe by less. Degree, but also improving correctness and like overall skin fairness. So I think that's also important because in the end, marketing is not only not promoting the product, but meeting the demand of the customer.
Right? So maybe the customer doesn't really care about so much percentages, rep pigmentation change, but they like where they, how they look. Then let's say they combine two effects. So we essentially help formulators tap into this. Psychology of perception. So how consumers want to look essentially, uh, can, they, can again, can they again give it an illustration?
Because before, of course, [00:13:00] everyone wants to like look beautiful, like, uh, probably younger, like have a better skin tone or you know, more plump skin. But can they give it to visualization?
Deanna: Yeah. Yeah, that's great. We're we're learning what people think. Beautiful is in a whole new way. Um, you,
Anastasia: Yeah.
Deanna: you mentioned creating images. just just to help me get clear on this, the, the sorts of images that your technology creates are for marketing purposes. Can they be used as clinical photos or that's probably a legal issue.
Help me think about how these sorts of images make sense.
Anastasia: Yeah, I think that everything related to AI is still kind of yet to be defined. What is specific framework? Um, o obviously all the images should be labeled as, uh, generated by artificial intelligence and that they have been modified. These are not real pictures, uh, but the images are, uh, we are now piloting their use of, uh, gene braided images [00:14:00] for marketing to illustrate.
Facts of products, and I think it's, it, it's quite different, uh, from what you usually would hear about use of Gen AI in marketing. When it comes to faces, for example, do Real Beauty, made a statement that they will visit introduction of Gen ai. They still will not use any type of genetic FA technologies to modify the image and faces and the looks of real humans.
But skin g in, in that sense is different because it allows you to demonstrate effects, uh, of the product and consumer can do it in their old photo. Um, and I think it's more. Rooted into education rather than marketing. Um, because, you know, sometimes the effects, what you see in the clinical trials were not as impressive as your average before and after when you walk into retailer store.
And why is that happening? [00:15:00] That's because a lot of brands would pick up the best. It's like one in other cohort has this amazing effect, but everyone didn't, everyone else did not. So skin, GPD is actually, can make it more fair because they can take an average effect and they can show this is what you can expect based on the average.
So, uh, that's much more real than one cherry pick photo, but had amazing results, but everyone did not have this amazing results.
Deanna: Mm-hmm. Know it's interesting 'cause the technology is really changing from things like Photoshop and Instagram filters into something that is much more. Rooted in reality, um, and, and that sort of trust is building. Before we finish up, I wanna talk about this idea of facial image anonymization technology, um, which I think is super interesting.
And I'll just say what I think it is, but I really wanna hear what, what you know, that it is and, and how you're using it. Um, but for me, [00:16:00] this is basically keeping the. Not necessarily the data of my face, but the identity of my face private, um, from the tech provider, maybe private from the brand. Um, in the same way that I might not, you know, want to share my email address or password with, you know, a tech provider.
Um, it's encrypted in some way. Uh, so, so help me think about facial image anonymization, which is fun for me to try to say, um, what is this? How does it work? Why are you using it? Is this a data privacy thing? What? What's going on here?
Anastasia: The technology that you mentioned, skin ization, we incode in how AI call it skin atlas. It's our patented technology that allows to take any face image, whether it is high resolution lab image or it's a photograph that you have captured with your phone, like selfie and produce an anonymized derivative, which is compatible with our image analysis system.[00:17:00]
We can extract still all of their facial skin metrics, but we will not use the original photo, but rather this anonymized derivative. I think that, uh, there was an article in Harvard Business Review saying, you know, a lot of innovation happens when you, not in a way you expect it's not planned and it's not.
The roadmap and skin atmosphere is an example of this type of. Spontaneous innovation be because before we decided to use Skin Atlas for organization, we essentially were working on methods to optimize speed and performance of AI systems. So we wanted to build a format that doesn't make, uh, mistakes when they look at the eyebrows and the hair color, or like, essentially discriminate uh, people by, you know, the, the way they sell their hair.
And as an outcome, uh, we realized, oh, wow, it looks like now the system is very accurate, but also the image is no longer personal images. Like completely. It's like [00:18:00] transformers doesn't really look like a face anymore, so. And I thought, okay, uh, why don't we use it for anonymization? Because, you know, algorithms just work better with it and we don't really need original face and there is no, for us as the company, there is no reason why they would hold original face image because a, a lot of the privacy laws, a lot of the risks that this imposes.
So then we decided to, you know, talk to our customers and say, Hey, would you like to ship to anonymized processing? And I would say. 99%, I think like 100% said yes. Just give it to us. And, um, I think that what the idea is for businesses, of course, it's mitigating all the privacy risks and all of the complexity associated with that.
It's also a question of trust, uh, consumers knowing that you really here to. Provide me with service and not to use my data for some type of other side applications that we, you know, in, in our [00:19:00] contract with via terms and conditions or in our social contract is me being your loyal customers. We didn't agree to.
it's quite funny, but we still have to, you know, talk about these things in 21st century.
But, you know, essentially if you are leveraging someone's resources, you know, it has to be on the fair terms. So I think that, um. Skin Atlas technology is fair for consumers because it preserves their privacy. It's also better for companies because navigating privacy laws, it's extremely challenging and it's also better for how, because our system actually work better with anonymized format because we have specifically designed this skin atlas to improve the performance of our AI systems and skin analysis algorithms.
Deanna: Yes, yes. Oh no, that's fantastic. I, I appreciate all of that information. And I, I just wanna ask one quick follow up question. Is there any noticeable change in terms of that personalization aspect? Um, you know, you, you, you know, mentioned that it doesn't necessarily look like [00:20:00] my face anymore. So how does the customer notice anything?
Do they feel like maybe this isn't as personal? Is this a more generalized recommendation or information that that comes out of anonymized tech?
Anastasia: Yeah, here's the thing, uh, while we no longer have the original photo of the prediction of skin ana skin features, essentially more accurate. As an outcome, your recommendations and the conversations that we will have for our LLM systems will be more personal. Uh, when we have, uh, in this new model that we call Skin Atlas and also our new duration of skin analysis metrics, we can differentiate different types of lines, you know.
Variant lines, fine lines, deep lines, nest label falls. So we couldn't do that before, but we didn't have this anonymized format. So essentially now we can even get very targeted recommendations. For example, for pigmentation, if you have sun spots without specifically the product that works in the stubborn sun spots, or opposed to having melos, and probably you will need different [00:21:00] products.
So in a way, this analysis is better on. Not only the level, but it's anonymized, but also like just from the quality, um, level of recommendations and insights you can get from the system when you are using this new module, this anonymization.
Deanna: Yeah. That's fantastic. Well, Anastasia, I have to thank you for sharing your time, your knowledge with us here today and of course for being a guest on the Cosmo Factory Podcast.
Anastasia: Thank you so much Janet. It was an amazing conversation and thank you for all the listeners for staying with us. I.
[00:22:00]