AI has certainly been a hot talking point of the fragrance industry. Some dismiss it as marketing hype, others praise it as the future of fragrance― revolutionizing its creation, retail, and marketing. Global manufacturers like Givaudan and Firmenich are investing in the space, while startups are offering consumers AI-created scent interpretations of their most cherished images, or hundreds of customizable scent options. A recent report by Future Market Insights predicts that from 2026 to 2029, these technologies will reach mass popularity, aiding in the growth of the global perfume market to a $101.3 billion value by 2033.
Our other senses like sight and sound have been very quickly digitized, but digital olfaction has been a more nascent category. “The thing that’s been missing is what the other senses have already had: a map. RGB is a map of color, low to high frequencies a map of sound. But smell hasn’t had that, and the reason why is because smell is more complicated,” Osmo CEO and neuroscientist Alex Wiltschko tells BeautyMatter. “We have three channels of color information in our eyes, but over 300 odor channels of information in our nose. Whatever map we were going to build of smell, we were going to have to wait for new tools, and in our case, we reached for machine learning.”
Osmo is positioning itself at the front of that movement towards machine learning-driven fragrance. The Massachusetts-based startup, fueled by a $60 million Series A funding round backed by Google Ventures and Lux Capital, has constructed an AI-powered map of odor that can predict the smell of a molecule based on its structure—in short, a computer-rendered, superpowered nose. The map is an extension of the Graph Neural Networks built in 2019 by parent company Google Research, where Wiltschko was Staff Research Scientist. Wiltschko and his team fed the software 5,000 scent molecules from fragrance catalogs to start, followed by 400 molecules that had been designed but not produced, therefore leaving it up to the machine to predict their smells. These predictions were then compared to the scent interpretations of a 15-person-strong panel. While there were differences in opinion, the machine was able to predict the odor perception in half of the cases.
The end result of this initial research is an “almost 300-dimensional” map constructed using pattern discovery on large datasets, which are still growing today. In turn, this equates to exciting aroma molecule-building potential, although the system does have one blind spot, which is molecules that are identical in structure but different in smell, or simply put, subtle nuances differentiating one smell from another. The more data the system will be fed, the more highly attuned it is likely to become.
In the realm of molecule creation is where this odor map truly shines. The industry has been designing aroma molecules for years—think the Iso E Super that made Molecule 01 a blockbuster, or the Cashmeran adding a comforting softness to scents like Dans Tes Bras—but these take extensive investment and testing until they launch to market. Osmo’s approach to creating new molecules is entirely driven by AI tools, driving up the time efficiency of the process. “Our ability to digitally sniff what a molecule might smell like means that we can consider billions of molecules before we even make one. When we make a molecule, we do that with high confidence that it’s going to smell a particular way according to our software system,” Wiltschko states.
In regards to the branch of natural raw materials, there is also potential for replacements. While natural fragrances have witnessed a rise in popularity in recent years, partially driven by the clean fragrance category, the sheer amount of land, water, and human resources needed to harvest these ingredients means that having all perfumes made with only 100% natural ingredients is not a sustainable reality―especially given the extinction potential of certain ingredients like sandalwood. On the flip side, the livelihood of certain communities in harvesting countries is completely dependent on this method of production. If all-natural ingredients were swapped out for synthetic counterparts, ensuring the financial futures of these people would be a resulting challenge. Another challenge is increasing restrictions on ingredients like oakmoss or the banning of them altogether, as was the case with lilial and lyral, which required perfumers and fragrance manufacturers to reformulate their creations in response.
Osmo has the potential to not only replace some of these restricted ingredients but open up an entirely new realm of olfactory possibility. “Can we start to smell, as consumers, things that were previously only found in the rainforest, or that were so expensive that we’ve never smelled them at a price point that makes them broadly accessible? That isn’t just about one molecule or ingredient, it’s about a palette. Ultimately, if we’re successful, we should be able to bring better smells to more people,” he explains.
Individual interpretations in scent perception have been the biggest dividing point between the human nose and the computer nose. After all, the very art of fragrance itself is strongly anchored in human memory and emotion attached to smells, but Osmo’s systems are bringing us closer to a universal language of scent. “People do smell things differently, for the most part though, if you ask enough people, you can get a very stable rating of what something smells like. What is very culturally specific is whether or not they like it. What it smells like, and how potent it smells, is something that we can apply science to today,” he enthuses. “Whether or not we can make something that people like, that’s still in the domain of art. Osmo lives at the intersection of science and art and we try to bring as much science in as possible because science can create better tools to unlock greater creativity.”
Combining technology, human biology, and chemical engineering, Osmo’s unique proposition has provided fertile growing ground for a realm of creative possibility and future expansion. “It’s a delight to work at the intersection of multiple fields because it means that you are never done or a real master of the thing that you’re trying to get good at,” Wiltschko adds. “Today, machine olfaction is a field that is barely existing, and it is at the intersection of neuroscience, computer science and operations, and all kinds of other things. We may one day view this as a single field that requires lots of different skills in order to be a practitioner, but right now we’re sitting at the intersection of lots of different fields that don’t normally talk to each other and bringing them together in one company.”
Osmo’s future potential isn’t limited to the fragrance realm. Together with the Gates Foundation, the team has designed eight mosquito repellants that are more potent than the current existing DEET formulas. Avoiding allergen and irritation potential―a subject related to DEET as well as certain fragrance ingredients―remains an important aspect of these developments. “The bar that we hold our ingredients to is high and getting higher and I think justifiably so. When we come up with an idea for a new ingredient before it’s launched, it will certainly go through stringent testing at international standards in each of the geographies that we will deploy the ingredient into. Now we can speed up the discovery process by predicting molecules that will very likely satisfy all those requirements,” Wiltschko explains. At this point, the team has created a set of roughly a dozen molecules, with this set growing weekly. The company is in talks with fragrance companies for licensing these new ingredients.
Rather than this AI system posing a threat to the manual art of fragrance creation, Wiltschko sees it as a powerful instrument in pushing the possibilities of the medium. “Steve Jobs came up with the idea that computers are a bicycle for the mind. We’ve been able to build machines that make us so much more effective physically. Similarly, we use computers at Osmo to help make our minds do more than they ordinarily would be able to,” he explains. “There’s no way a chemist is going to manually review one million molecules, much less a billion or a trillion, the space of all possible molecules, which is even larger. The way I view software and artificial intelligence is as a tool. To the extent that we can make them better, faster, or more automatic, it means we have more time left for the creative aspect of the craft.”
Osmo’s mission is ultimately about redefining the building foundation of a subset of the current cosmetics system, and with such structural change can come a plethora of new possibilities. As an enthusiastic Wiltschko states: “Today we’re quite focused on the thing that we know how to do well and where we think we can bring value, which is in the creation of new ingredients, but stay tuned. What we’re interested in is whether we can help people produce better ingredients to produce better products to produce meaningful impact in people’s lives. The imagination is the limit.”