“We wield AI the same way you’d wield a hammer — it’s a tool,” said Alex Wiltschko, the Google alum at the helm of AI-powered fragrance house, Generation by Osmo. 

The 2023-founded start-up, which coined the term “OI,” or “olfactive intelligence,” and is soon to graduate from its 10,000 square-foot New York City lab to a 60,000 square-foot office and manufacturing facility in New Jersey, is on a mission to “digitize scent,” as Wiltschko puts it. 

For Osmo, that doesn’t necessarily mean using AI to refine and improve fragrance note compositions — in fact, the company’s in-house master perfumer Christophe Laudamiel (of Tom Ford Amber Absolute, Ralph Lauren Polo Blue fame) — often does that. Rather, the business-to-business company — which counts Google Ventures among its investors and Geoffrey Hinton, the 77-year-old computer scientist widely referred to as the “godfather of AI,” on its scientific advisory board — is using OI to “build fragrances from the molecule up,” Wiltschko explained. 

On one hand, Osmo uses OI to develop “source codes” for naturally occurring olfactive ingredients that, once generated, make the scent of these ingredients infinitely replicable without subsequently requiring access to the ingredients themselves. 

In other words, “we’re reverse-engineering smells,” said Wiltschko.

The first ingredient to successfully undergo this process at Osmo was a plum, one slice of which was placed into a dual-phase device called a gas chromatograph mass spectrometer (GCMS), which is able to separate, identify and measure — in that order — the chemicals and compounds in a given product or mixture.

“We took the slice of plum, put it into the GCMS, digitally encoded and reprinted the smell, and put it into this bottle,” said Wiltschko, holding a glass vial carrying a translucent fluid — one of hundreds in Osmo’s lab, each containing and labeled with a formula for a different, reverse-engineered olfactive ingredient — from sandalwood to cedarwood oil. “You have to use [generative] AI in order to digitally encode, and then decode, these ingredients — the data that comes off a GCMS is just too complicated otherwise.” 

Osmo is also introducing new olfactive molecules altogether, the first three of which have been unveiled as glossine, fractaline and quasarine. 

“We use AI to screen through billions of possible molecules that have never been made before and to digitally ‘sniff’ them — or, predict the smell of these molecules. From there, we can make a very small number of them — we’re still just humans, with hands, who can’t do everything — test them for what they smell like and for their safety, and turn the best ones into products,” said Wiltschko. “Glossine, fractaline and quasarine — these are born from OI but ultimately touched and tested by people.”

The possibilities these nascent technologies enable, the founder and chief executive officer said, are vast. 

“You can create a fragrance that nobody else can, because nobody else has that ingredient,” said Wiltschko — a likely enticing proposition for denizens of an industry where coveted fragrances are notoriously duped, and even more notoriously difficult to protect from a copyright perspective. 

And because fragrances can be brought to market by brands via a maker like Osmo more quickly than has historically been the case, “it means things like drops are possible — products that capture a very brief, ephemeral trend are possible,” Wiltschko said. “If you can now navigate the fragrance development process in two, four, or maybe six months depending on the level of customization — you can react at the speed of culture.” 

Osmo is one of many companies — emerging and existing — using AI-driven formulation to push the boundaries of what beauty and personal care products are and can be. But the role that AI plays in product development, and thus, its end impact, varies greatly depending on who’s using it and how. 

“There are largely three classes of AI: generative AI, which has been all the rage of late; predictive AI, which was all the rage before that, and automation AI — all three types have a place under the sun in the world of cosmetic creation,” said Iva Teixeira, cofounder and CEO of 2018-founded The Good Face Project, a fast-growing subscription software she refers to as “Canva for the cosmetic chemist.” 

Beauty’s Best-kept Software

According to Teixeira, Good Face is used by more than 30 percent of all product manufacturers in North America and has roughly doubled its subscriber count each year since 2022. Its hundreds of beauty clients, who “pay a hefty monthly fee” to access the platform’s AI-powered features which can streamline market research, claim validation and regulatory compliance for products, include L’Oréal Groupe, E.l.f. Beauty, The Estée Lauder Cos., Milk Makeup, Saie Beauty, Amika and more. 

“We’re a solution for the pros — our software does not author the formulas; it supports the chemists and R&D teams that author formulas,” said Teixeira, adding that clients across skin, hair, makeup and fragrance, are “equi-prominent in our customer base today.”

The software has been trained since 2018 via “huge scrapers that download all ingredient lists of all products being sold everywhere,” she said, adding that algorithms for the model have been designed to “pull apart and create synonyms for ingredients — for instance, vitamin C and ascorbic acid.” The ever-growing dataset is also supplemented with ingredient and clinical research from more than 80 databases including PubMed and ScienceDirect. 

With this, one of the most common ways a chemist might harness Good Project is reverse-engineering an existing product in order to understand what makes it successful — and replicate it with as many or as few changes as they’d like. 

“A chemist can say, for instance, ‘my brand customer told me this is the number-one lip oil at Sephora — how can I recreate it, but with a new fragrance and making it blue instead of red’ — Good Face can show them what raw materials they would need, create a template for the formula, and so on,” Teixeira said. 

A chemist might also use the platform to help rework existing formulas and seek commensurate alternatives for restricted ingredients — a common use case amid the backdrop of today’s shifting ingredient regulatory landscape. “The sandbox for R&D was much smaller before solutions like Good Face arose, which made all of these templates and options available to chemists in mere seconds,” said Teixeira. 

Because the model is trained largely on ingredients and products that already exist, though, an Olaplex-level breakthrough — i.e. a first-to-market molecule or product — can’t be generated via Good Face — at least, not yet. 

“Our aspiration for tomorrow is to enable that kind of molecule design; to shift from formula design to molecule design, because that’s where the future is going to be,” said Teixeira.  

All Hail the Molecule Makers

That’s where Debut, the L’Oréal-backed biotech firm which operates a portfolio skin care brand, Deinde, and this year unveiled an AI-powered beauty ingredient discovery platform called BeautyORB, is focusing its efforts. Like Osmo does for fragrance, BeautyORB uses AI to scan through billions of potential molecules at a speed and scale that wouldn’t be possible for a human to do, in order to vet those which may be beneficial for beauty, and specifically skin care, formulation.

“There are something like 50 billion potential ingredients, with 50 billion different molecular structures, on this Earth — which, when formulated for our skin, could or could not have an effect,” said Debut founder and chief executive officer Joshua Britton. “We as scientists have tested just a fraction of those molecules, meaning it’s unlikely that we’ve found the best molecules for skin, body, scalp health — I believe they’re still out there.” 

Tracking down and developing these molecules is the aim of BeautyORB, which has so far launched three skin care ingredients available for use by brands and manufacturers:  DermCeutical InflammagePRO, DermCeutical Barrier RepairPRO and DermCeutical LongevityPRO. 

“The first ingredient we developed took about two-and-a-half years during which we built our infrastructure; now we’re getting to a point where we’ll release two ingredients per year,” said Britton. “Without AI, it would have taken around 125 years to develop a single one of these ingredients.”

BeautyORB’s computational model is built in-house and currently comprises roughly 100 million data points pertaining to skin health. “It’s able to take data and make predictions; you can go to the model and say, ‘hey, what molecule do you predict can address X concern,’ and it’ll spit out different molecules. We then try to create them, test them in the lab, and go to market with the successful ones,” Britton said. 

Man vs. Machine?

No form of AI can take on the task of conducting clinical trials yet, though Britton anticipates it’s more likely that a model may soon be able to conduct safety and toxicology tests: “Those are a bit more binary — they’re, ‘this chemical structure gave this response’ — whereas with clinical testing, [results] are always different.” 

The reality is, though, AI’s unmatchable ability to process information and predict outcomes will — and is — impacting beauty’s workforce. 

“AI speeds up the research process, it reduces trial-and-error, it speeds up speed-to-market, and it cuts cost because you’re using less manual labor,” said David Chung, the founder of Farmacy, The Rootist and beauty manufacturer/incubator ILabs. “For manufacturing at ILabs, we will dramatically reduce [personel] in production lines, because the price of automation will come down through AI, which we’re already seeing.” 

Added Teixeira: “There’s going to be a reduction of ‘mind-numbing’ activities across all industries…if you think you’re in the R&D cycle but all you do is data entry, you may need to reskill yourself.” 

But still, in some ways — for instance, the roles of perfumers and cosmetic chemists — AI still cannot do the work alone, experts say. 

“AI can never be used in isolation — the craft of fragrance is still deeply needed,” said Wiltschko, adding that OI “produces a wonderful starting point for fragrances, and can be used for certain manual work…so that the creative process, the refining and interacting with the client — that’s where the perfumer can focus their time.” 

This sentiment is echoed at 1895-founded Swiss fragrance house Givaudan, which is increasingly leaning into AI-assisted scent formulation in a different way via proprietary tools like Carto and Myrissi, which allow for instant scent sampling and align colors with fragrances (and, by extension, emotional states), respectively. 

“We don’t want our perfumers to be dedicated to low-value tasks,” said Johan Chaille de Nere, Givaudan’s director of digital transformation, adding that it’s not so straightforward whether implementing AI-driven processes leads to “better” fragrances, but rather, “The difference is, we are able to do what we do in a more complex environment. 

“AI helps us digest this fast-moving regulatory context; it helps our perfumers develop faster, because there is more and more pressure from customers to develop faster — with digital, we can do what we’ve always done, but now in a context where we have more complexity and more constraints.” 

With that being said, though, AI’s continued rapid development does make the future of these dynamics unclear. 

“The relationship between the cosmetic chemist and AI isn’t adversarial, but it is an awkward one — everybody’s telling you that you have to use it,” Teixeira said. “But being able to speak the language and recognizing that generative AI might be the spoon, and predictive AI might be the knife, and with them, you’ll start being able to have a whole meal, so to speak, is incredibly important.” 

At Givaudan, “it’s up to perfumers whether they want to use AI in their creations or not,” said Chaille de Nere, adding that new graduates of the company’s perfumery school have been “trained with AI, so they use it in all of their creations,” while longtime perfumers are more split. “Digital transformation requires changed habits, and that’s quite difficult when you have 30 years of experience behind you — that’s one challenge of AI, is accompanying teams to embrace what it has to offer.” 

But even in less-official capacities, using AI — often, public, generative models like OpenAI’s ChatGPT — has become routine for developers. 

“Most recently, I asked ChatGPT how skin care can be part of the mental awareness category — the response was interesting,” Chung said. 

Cosmetic chemist Ginger King, founder of product development company Grace Kingdom Beauty, also considers ChatGPT a key tool for research purposes. “I’ll ask it for suggestions, to propose different ingredients to solve issues — it’s only 80 percent accurate, so you need to fine-tune it; it still needs that human touch,” she said. 

As for the question of whether these tech-optimized development processes lead to better products, on one hand: “yes, because finding better ingredients will lead to better, more effective consumer products,” said Britton, stipulating, though, that the appeal of a category like beauty — as opposed to, say, the pharmaceutical industry — goes beyond efficacy. 

“Beauty is such an art and a science, and that’s why we love it,” he continued. “So that’s the question — does beauty become something from which people want solutions and nothing else? Does it become that data-driven? Or is there always going to be a very large emotional aspect to this? We just don’t know yet.”