COSMOFACTORY

Deanna: [00:00:00] This episode is about creating space for true creativity and innovation in product development. It's about artificial intelligence and how we can sensibly use this computer science to support cosmetic science. It's about ingredient discovery, product benchmarking, formulation optimization, and much more.
Deanna: Today, on the Cosmo Factory podcast, I'm joined by Yahya Syed, co founder and CEO of Potion AI, and by Hejab Malik, co founder and chief product officer at that same company. Welcome to you [00:01:00] both.
Yahya: Thanks so much for having us. We really appreciate it.
Hejab: Yeah.
Deanna: I'm glad you're here. So before I start asking you questions, I would like to give a little bit more context to our conversation by saying that over the past four or five years, the supply side of the beauty industry has gotten a lot more digital, uh, ingredient makers and manufacturers have updated their websites.
Deanna: They're creating engaging content, not just, uh, The webinars and white papers, but also video and social media content. Many suppliers offer a virtual facility tours. I see some hosting in the lab video consultations with their client partners and prospects. And of course, we also see many more tech service providers in the beauty industry that have developed, um, ingredient discovery platforms, regulatory support software, cloud based formulation tools, most of which are noticeably more sophisticated and user friendly than what has existed in the past.
Deanna: So. We know that one of the most promising digital technologies today is, of course, AI, and that's where our conversation [00:02:00] can get started here. Um, Potion AI is basically helping cosmetic formulators, as I understand it, establish a product benchmark. So, tell us first, if you will, what you've learned from formulators about their conventional benchmarking processes.
Yahya: Yeah, sure thing. So I think, um, perhaps the most interesting learning is that, you know, most product development in the beauty space is iterative in nature. And so, you know, uh, any product brief will come with a benchmark product that's supposed to serve as a point of reference for desired performance, the desired texture that could be a single product that could be multiple products.
Yahya: Um, and the first step is to basically, uh, you know, decompose that benchmark product and understand what is driving the performance, what is driving the texture, what ingredients might be inside, specifically at what concentrations, et cetera. Uh, and, you know, our, our [00:03:00] goal with, with our deformulation tool is to essentially help automate, uh, what is, you know, the, the first step in, in that process of, of, of getting on the bench.
Deanna: Yeah. Yeah. And I think in one of our earlier conversations, because I have gotten to speak with you both, um, previously in person and, and virtually as well, but Hajab, I think you mentioned, um, you're hearing from formulators that you've seen people just pouring through PDFs and, you know, trying to create some sort of documentation.
Deanna: Um, how tedious is that, um, benchmarking process in, in, in what you've learned from, from your clients?
Hejab: Yeah. So, um, what we've learned, you know, after talking to hundreds of formulators and kind of what led to us developing this tool in the first place is that this process of benchmark evaluation is something that's already being done manually every day by formulators when they receive a product brief or, you know, when they're just curious about something new that's [00:04:00] just hit the market.
Deanna: Yes,
Hejab: So what we've built essentially is, um, you can think about it as just an automated way to do what they're already doing. So essentially, you, you know, receive that benchmark, you have an inky list, you can put it into potion and, um. What formulators are doing previously is taking each inky and the benchmark and just, you know, using the tools that they had at their disposal, like Google, various discovery platforms, their own, you know, internal inventory lists and trying to figure out, okay, well, what trade names might be inside and, um, what specialty blends might be inside and what are the ingredients that You know, are just the fluff of this benchmark and what are the ingredients that are really driving the performance, texture, efficacy, et cetera.
Hejab: So with potion. Now, what you can do is, um, basically get those answers faster. So, um, it'll analyze your list and just present to you, you know, here are the different raw materials that might be in here. [00:05:00] Um, here are their potential functions so that you can just quickly figure out, um. Well, it's driving that benchmark and then continue on with your workflow as per usual, right?
Hejab: Um, because obviously you're not trying to just replicate that benchmark. Most of the time you are trying to see how it's working so that you can then decide, well, I want to make X, Y, Z claims. So I'm going to, you know, remove these ingredients, add these ingredients, um, and obviously, like, make the formula your own.
Deanna: Right, right. That's helpful. You both mentioned automation there, Yaya. You have something to add?
Yahya: Yeah, I was just going to say, I mean, to, to, in a more tactical sense, you can imagine to your point about going through a bunch of documents, you know, you, you might have a benchmark that has 50 ingredients inside, um, and you might have raw materials on the market, uh, that may be in that product that have 5, 6, 7, 10 ingredients.
Yahya: And so that, that matching process of figuring out, uh, you know, one, what inkies correspond to which raw materials. [00:06:00] Uh, which of those raw materials might then be inside the formula that that can become a very cumbersome and complex process very quickly. And, uh, you know, technology can help identify potential matches much more quickly than you or I, you know, with, uh, you know, a spreadsheet or what have you could.
Yahya: And I think a lot of folks, uh, traditionally just rely on pattern recognition. So they're used to seeing certain in keys within formulas and, you know, no, okay, well, that might be this plan or that might be that blend. And, you know, automation can essentially, or technology can essentially provide that skill to, you know, just about anyone with an internet connection.
Yahya: And
Deanna: No, it's funny to think about it, but it's almost like, um, the, the task that you're describing helping formulators with is almost some of their administrative work. Um, I don't want to say busy work because it's, it's super critical, uh, to [00:07:00] developing that the end product formulation, but it is sort of that, that admin desk work.
Deanna: Um, I'm curious. Um, I have of course, visited the potion AI website. site. And there I'm, I see that you list some of the users and they clearly include ingredient companies, manufacturers, and formulation consultancies, um, all by company name. Of course, this is my, um, my summary of what they do, but I'm wondering to hear from you, what kind of feedback you're getting from formulators who are using potion AI in their workflow for this sort of benchmark, um, tool.
Yahya: yeah, so I think our core target demographic, the folks that were really focused on pleasing more so than anyone else, uh, are the formulators and product developers. And so that ends up being, you know, really the CMs and the brands. And, uh, I think that the, the first, uh, thing is just, it saves so much time, uh, particularly with blend identification.
Yahya: You know, there are more things you can do with the [00:08:00] tool, but, uh, a lot of the users, uh, Just love being able to quickly thrown an ingredient list and get a sense for, uh, you know, what might be inside. So that's that's probably, uh, gratitude, uh, is the 1st 1st point of feedback. We've received. Um, we've also seen or heard from a bunch of CMs that they are sharing.
Yahya: Uh, with various stakeholders actively that they are using potions. So whether that's with their boards, uh, to show, you know, that, uh, their R and D team with the same footprint can service more customers, create more formulas, et cetera. With their customers showcasing that they're using the latest cutting edge technology that they can help bring products to market faster.
Yahya: Um, and so I think, uh, you know, that that's something that they've been actively sharing with us. Uh, we're collaborating with some of them to, you know, put some numbers behind, uh, you know, what some of these, like, time
Deanna: [00:09:00] Yeah. What the efficiencies are. Yeah. No, I love that. And I think to, um, you know, I appreciate hearing that you're the folks who are using your tools are, um, being candid about it. Um, and, and as you're suggesting it, it certainly helps in terms of, um, demonstrating efficiency, um, and, and process, um, streamlining.
Deanna: Is that a way I could say it? Um, but I think at this stage too, um, You know, for many folks, AI is still a very new technology, and it's true with most new technologies, right, that we, for best practices purposes, like to disclose that we're using them. I think, um, being very transparent about the tools you're using, um, only helps build trust.
Deanna: So that's, that's really nice to hear. I'm curious, um, if you would just remind me when you launched the tool to the industry.
Hejab: so, um, the free platform that you see today that you can, you know, sign up for through our website, um, was launched at in cost this year. [00:10:00] So it's just been, um, a couple of months and it's been, you know, growing very quickly. So hasn't been too long.
Deanna: Yeah, no, that's fantastic. And so I'm curious. I know hijab as chief product officer. One of the things you've told me that you're responsible for is deciding what features get built into the tool. Um, and I'm curious if you've made any changes or adjustments based on feedback from partners or if maybe there's some planned features you can tell us about.
Hejab: Totally, yeah, um, and that's correct. I do. I do spend most of my time just, uh, talking to users or talking to formulators. Um, and, you know, anything that you see us build is usually coming from from those users. Um, so in terms of changes we've made to the tool, um. We have, you know, when we first launched the tool, um, we, we immediately started getting a lot of feedback around how useful it was for identifying raw material blends, um, and just figuring out what trade names you could potentially use.
Hejab: So we have made some [00:11:00] adjustments to make it even easier to just see those immediately. So now when you, you know, uh, put a benchmark through Through the tool, uh, you'll just immediately when you click on an ink, you see, like, what are the highest overlap blends right at the top? Um, and then another request that we were getting pretty often, um, was around putting in a benchmark and then.
Hejab: Being presented other similar products in the market that you could consider that, you know, have similar inkey lists, um, similar formats, et cetera, um, which kind of goes a little bit into product development or like product research as well. Um, so then people were asking for, okay, well, what if potion could help me figure out what the benchmark should be in the first place?
Hejab: So we actually combined those last two things into our latest feature release, which is, um, essentially like a. search engine for finished products where you can just type in something like, you know, top rated SLS free shampoos, receive a [00:12:00] list of products from various retailers, and then, um, you know, analyze those products as potential benchmarks and also see, um, other products on the market that are very similar to those.
Hejab: Um, so I'd say since we launched, like, those are kind of the biggest adjustments and, and changes, and then, you know, Um, I think aside from that, we've gotten a lot of feedback, of course, on many things that will take a little bit more time to work on. And I think, yeah, you know, I'll let you kind of point some of those out.
Yahya: Yeah, absolutely. So I think one is, uh, you know, as part of the deformulator, we do provide concentration range estimates, uh, for each ingredient in the formula. Um, we'd like to basically make those ranges more precise. Um, they will probably always remain ranges because to what Hijab was alluding to earlier, we're not trying to help someone, like, dupe.
Yahya: Uh, a benchmark, uh, and it's more to serve for for [00:13:00] inspiration, but I think there's some work we can do on, you know, tighten tightening those ranges also suggesting ingredient alternatives equivalence substitutes in a more intelligent
Deanna: Mm hmm.
Yahya: Probably do this with our ingredient search functionality. Uh, but we would like to bring this to, and folks have been asking, uh, to see that in the context of the formula itself, uh, and then, you know, perhaps not surprisingly, uh, regulatory, uh, a lot of folks have asked, you know, us to incorporate, uh, You know, regulatory features into the tool, uh, we think there are various technologies in particular that are very well equipped to handle this sort of thing.
Deanna: Mm hmm.
Yahya: And so I think it is something we're very interested in. But there are, you know, as you alluded to initially, you know, there are tools and offerings that exist in the market. today that can that can do that sort of thing. So I think when [00:14:00] we think about prioritization, it's probably not the first thing, uh, you know, that that would come to mind.
Yahya: There's also, um, you know, the need for like partnership with like suppliers, right? A lot of the information need to power true regulatory Uh, you know, compliance, uh, requires disclosure of information from ingredient suppliers, which, you know, can oftentimes, uh, only be seen once you've actively purchased the ingredient itself.
Yahya: So there's probably, you know, more things to solve there. But we are very excited to, to, you know, help wherever we can
Deanna: Yeah. Excellent. Thank you for that. And you've mentioned, um, the ingredient search tool as well. There are actually a few tools that Potion has, not just the, the benchmarking one that we started the conversation with. Talk us through the, the three main uses of your technology.
Hejab: Yeah. So,
Hejab: so, yeah, like, like, we already kind of talked about, there's the benchmarking [00:15:00] tool. Um, and then we have essentially two, AI powered search engines, one that operates on raw materials and one that operates on finished products. So, um, the functionality is quite similar in that, um, you know, for raw material search, you essentially see like a, you know, Google ask input box and you can, um, describe what kind of raw material you're looking for in as much detail as you'd like.
Hejab: And then you'll get a result with the best matches on the market. And then the formula search. Similarly, you just. Describe like, what kind of finished product, um, you're looking for. You can even say things like, you know, make sure it includes or excludes certain in keys. Make sure it, um, you know, meets X, Y, Z criteria, and then you'll see products on the market that that match.
Hejab: Um, so those 2 search engines and the benchmarking tool are like our 3 core offerings at the moment.
Deanna: Yeah. Excellent. Um, and I believe you've started offering, uh, what can be described as insight and [00:16:00] analytic reports for supplier partners. Can you tell me what's going on there?
Yahya: Yeah. So, you know, as I was referring in the, you know, discussion on regulatory, you know, we, we'd love to basically increase the amount of information we're able to provide, uh, for any given ingredient to increase the utility of the tool to our user base. And so essentially to incentivize. The raw material suppliers to to provide that to us.
Yahya: Um, we are providing them or offering them to partner. And, you know, basically the exchanges. Uh, we will collect whatever information they're willing to share with us. We will incorporate that into our tool. We will verify their ingredients. Um, you know, their ingredients are more likely to show up, uh, for relevant search queries and the like, as a result, uh, and an exchange will give them insights on, you know, trending products that we're seeing on the platform, like, what are people trying to [00:17:00] deformulate, uh, at a high level basis.
Yahya: Uh, you know, what ingredients are trending, what claims are trending, so on and so forth. All of this is on a, on a company and user agnostic basis. So this, they will not be able to tell that, you know, XYZ formulator at, you know, Acme company is, is about to make, you know, such and such product, but they're getting a pulse for, you know, what's going on in the market.
Yahya: What are people working on so that they can, you know, better adapt their, you know, strategies. in terms of marketing and product development?
Deanna: No, that makes very good sense. Thank you. So let's talk more specifically now, if we can, about the AI portion of Potion AI. Tell us how this technology works.
Yahya: Yeah. So that is a good question. And I think the way that, I mean, the, the, the very quick answer is we use an ensemble of models, uh, you know, to do various things. And I think, uh, [00:18:00] we're very oriented around thinking about the job that needs to be done and then determining based off of that, Uh, what is like the best technological approach, uh, to, you know, essentially address that problem.
Yahya: So as an example, um, you know, when you think of the D formulator, uh, you know, that uses basically graph neural networks. Uh, to ascertain, you know, what the concentration of each ingredient might be, uh, within that formula. Um, when it comes to digitization, which is something that we do both for some of our customers and for the purposes of the platform as a whole, you know, LLMs are a great, uh, application, or I should say, like, a great paradigm to use to basically do that very quickly and efficiently.
Deanna: And just help me. It's language learning model,
Yahya: LLM is large language
Deanna: large language model. Thank you.
Yahya: of course. And so, you know, they're like, [00:19:00] just depending on, you know, what, what you may need to do, like if you're trying to predict the stability profile of a product, you know, that uses a different sort of model. So we kind of, uh, have built probably too many models at this point, uh, to do various things over the course of, of the, you know, last several years.
Yahya: And, um, you know, we're pretty ambivalent. Uh, on on what we use when as long as it solves the users need that is definitely, you know, guiding North Star.
Deanna: Thank you. Yeah, I think here maybe we should mention that Potion AI does have three co founders. Um, Alex Stack, I believe, uh, is your company's chief technology officer. He's not on the podcast here today. Um, as I understand his job, and please correct me if I'm wrong here, he's in charge of, of topics like deep learning and data science, that side of the business.
Deanna: Is that accurate?
Yahya: That's right. So Alex is actually a childhood friend of mine. I've known Alex for almost 18 years now, I think, um, and he has been nerding [00:20:00] out about artificial intelligence, uh, for over 10 years now. So he has a PhD in applied math as specific research focus was in deep learning and. This is, uh, you know, he, he basically enrolled in that program right after this, like, infamous ImageNet competition that took place in 2012, where, uh, you know, people say that that's almost like the, the rebirth, uh, of, of the AI era.
Yahya: And so he's kind of been, uh, in the throes of this for a while, and he's always been very, uh, Eager to and inspired to, uh, apply A. I. In an industry based context. And so, uh, yeah, he's the guy, you know, that kind of builds a lot of the stuff, uh, the plumbing, if you of our applications and how they work. And, uh, you know, he's, he's, he's busy chipping away now, which is why, you know, the 2 of us are talking to each other.
Yahya: yeah,
Deanna: yeah. No, I love it. I love it. It's it's nice to know who the people are [00:21:00] between project behind projects like this. Rather, I'd like to take a minute and think in in broader terms, if we can about what the technology might enable. And I have one idea. I'm going to run by you and you let me know what you think here.
Deanna: Um, but one of the challenges that I see ingredient makers in particular facing is what I would describe, um, Um, As new ingredient adoption. So when something, um, exists in the market, right, an ingredient is time tested or good enough in, you know, it's already available in the ingredient marketplace, brands and manufacturers can be reluctant, right.
Deanna: To invest the time and money that it takes to start working with novel ingredients or prospective alternatives. Um, So I'm kind of wondering if tools like Potion AI actually make space for further creativity or product differentiation, and we might actually see an increased demand from beauty makers for some of these more unique or, um, new ingredient technologies, maybe even, um, for novel ingredient delivery mechanisms, which people like to share with me.
Deanna: [00:22:00] Um, And, and I'm thinking maybe even uncommon manufacturing technologies that emerge from time to time. So I guess the question for you here is what is possible if we get to spend more time and effort on the formulation optimization piece of this?
Yahya: Yeah. So that's a great question. I think so. A bunch of thoughts come to mind. Um, one is a platform like ours should theoretically democratize access to information about all of these ingredients and assuming we have sufficient information and we can share that information with our users. Uh, you would imagine that some of these up and coming and fledgling startups or companies with new ingredients will have more of an opportunity to showcase their unique value proposition to the world, you know, and get their product out there.
Yahya: So that is something that a lot of ingredient suppliers have a voice to us. It's something that we're, uh, you know, very, uh, aware of, and we want to be thinking [00:23:00] about. You know, because we're in some ways we're kind of playing, not playing, but, uh, servicing both sides of the industry, right, both the formulators and so it's really important to us, especially as, uh, folks that are so inspired to build with the industry, the future formulation, like, how can we make sure that, like the, the upstarts, uh, have an appropriate, uh, voice, uh, in the market.
Yahya: So that's just one thing I wanted to mention. I do think what you're going to see naturally from tools like potion. is a concurrent, uh, you know, increased, well, I guess, uh, faster speed to market lower. It'll be easier to bring a product to market.
Deanna: I think I missed part of that lower, which,
Yahya: basically just, just faster development times. So just being able to bring products to market much faster.
Yahya: And I also think we're, because we're democratizing access to formulation and information, I think the bar to clear in order to stand [00:24:00] out and differentiate yourself in the market. will be raised. So as a result, when we are able to take some of those administrative tasks that we were discussing earlier and, and compress the time you spend on those, you can then spend as a product developer or formulator that time discovering, you know, that new active that can make all the difference in your marketing story or something, you know, where perhaps the value proposition, you know, resonates more with you.
Yahya: Or your organization. So, um, I think people will, in fact, be able to spend this time creating what are hopefully better products for the market. Uh, and then the customers.
Hejab: Yeah, I wanted to chime in, um, with, you know, a tidbit from what you might expect from potion. Um, you know, Deanna, you would ask what, what else is on our roadmap? What might be coming next? And, you know, something we're currently working on is essentially virtual formulation where, you know, You could [00:25:00] input elements of a product brief into potion and we would suggest to you a series of starting point formulas with raw materials that you could use.
Hejab: And so, you know, this example you gave of how can you leverage, um, raw materials that already exist, but perhaps, you know, no one's figured out how exactly to incorporate them. I think that could be an interesting, you know, experimental playground where, um, you know, you could, you know, Use potion to generate various starter formulas with these like new novel ingredients and then take those to the lab and you know, um, conduct the process of figuring out, you know, what works and what doesn't.
Hejab: So hopefully that's something that, um, can tie those things together in the near future.
Deanna: no, that's really interesting in one of our earlier conversations, you told me that UX, um, user experience and UI, user interface, will be absolutely critical in the age of AI. I'm wondering if [00:26:00] you can help me understand what you mean by that.
Hejab: Yeah, definitely. So I love to use the example of, um, chat GPT here because I think that, um, you know, we really believe that powerful product is what drives adoption. And it's not necessarily, you know, how complicated or powerful the AI models or the technology, you know, behind the product is, of course that's important, but at the end of the day, The user needs to be able to interface with and use these tools, um, in a way that's intuitive.
Hejab: And, you know, they would, they would want to do every day. So I think what's interesting about chat GPT is, um, you know, the, the frameworks that they're using have actually existed in the tech space for quite some time. It's just that they're winning innovation. I, how I see it is. The interface, so they product ties those tools in a way that made it so that anybody of any age group, you know, from anywhere can just log in.
Hejab: [00:27:00] And see this really simple, easy to use interface and immediately know what to do. Um, and I think that's so powerful, and I think that'll become more and more important because I think the tools that, um, gain adoption and that, you know, people continue using will need to have, um. This sort of user love
Yahya: And I would just add to that. I mean, people are very practical, right? Like when, when you and I use our cell phones, we're not thinking about what sort of batteries inside or, you know, what the lumens are of the screen, you know, we just pick up our phone and text our friend or email our colleague and put it down.
Yahya: Right. And I think this is very much the same thing. Obviously we need to take into consideration that like, Okay. You know, like, the software is safe to use, like, the data is being treated appropriately, but beyond that, like, people have problems that need to be solved, and they want to use tools to help solve them.
Yahya: Uh, and that's kind of what determines success and failure, uh, you know, in [00:28:00] most of life, I'd argue.
Deanna: Yeah. Yeah. Great. So, um, before we wrap up here, you mentioned safety. I do want to hear a little bit about how you're safeguarding individual and enterprise users, as well as, you know, the data that you clearly have access to, um, from, you know, perhaps the, the search, uh, that folks are using and, you know, any other, um, sort of notes on safeguarding here.
Yahya: Yeah, absolutely. So I think it's really important for people to consider our incentives, right? Which is, if we are trying to build a tool for formulators, if we do anything that can put that, uh, trust at risk, you know, our entire business will fall apart. So we are very, very You Focused on making sure we don't do anything we shouldn't and that any data that we have or is being generated on our tool is being treated with the utmost, uh, respect, uh, and, and being, you know, treated appropriately.
Yahya: So we are not collecting anything on an [00:29:00] individual basis. We are not recording that the sort of information you will see us record is, for example. You know, if a bunch of people are looking for actives to address a particular claim, you know, we are basically capturing those macro insights in the deformulation tool, input a percentage into our table, uh, we will observe whether or not that percentage is within the range we provided outside of the range we provided, et cetera.
Yahya: Again, in the spirit of improving our models, right? And et cetera. That's kind of where things stand right now. Uh, there is a chance perhaps that we will have dedicated enterprise environments with some of the larger companies at some point.
Deanna: Sure.
Yahya: Where, you know, they just don't want their data even in the same, uh, you know, literal physical metal hardware that, you know, someone else's might be and we completely respect that and, and want to be able to service those [00:30:00] folks as well and provide them the benefits of our tools.
Yahya: But as, as of right now, that is, that is not, uh, you know, that's not how the application is set up, at least the free version of it.
Deanna: Excellent. Excellent. I appreciate that. It's been, um, really helpful to understand. Well, um, Hajab, Yahya, thank you both for such a fascinating conversation here today. I'm very glad you could join me on the Cosmo Factory podcast.
Yahya: Thanks so much for having us. It was a blast.
Deanna: You're welcome. [00:31:00]