Over the course of a 36-year career at Procter & Gamble, master perfumer Zerlina Dubois created more than 100 different fragrances for Pantene, Herbal Essences, Head & Shoulders, and other major brands within the consumer-product giant’s $15 billion beauty brands division. She did it all without any help from artificial intelligence.
But Dubois said she was intrigued by a call from Alex Wiltschko, the founder and CEO of Osmo, an AI fragrance company that has raised $130 million in funding from backers that include Two Sigma Ventures and Lumina Partners. Spun out of Google Brain, the startup is building a 60,000-square-foot facility in Elizabeth, New Jersey, supporting manufacturing, research and development, and logistics operations. The company’s mission is fairly straightforward: to digitize the sense of smell.
“This is where fragrance is going in the future,” Dubois told BeautyMatter. “It’s interesting to help create that future for fragrance.”
There are large distinctions in the science that goes into creating a fragrance that’s intended for the shower or a home spray—which provides an intense burst of smell that activates the senses—compared to a cologne or perfume that’s meant to linger more subtly. Dubois says she’s helping Olfactory Intelligence learn more about those differences.
While a musician or a photographer may use AI-enabled tools to manipulate sound or visual images, perfume experts say the sense of smell is unique because it is the most emotional of the five core senses. There’s also an immense trove of data for perfumers to sort through: the various compositions of ingredients for different category formats, like laundry versus perfume, and unique representations for brands. A floral or woody note differs between Calvin Klein, Bath & Body Works, and Chanel. Jasmine from India isn’t the same as jasmine grown in England.
“The world of olfaction is far more complex than the other senses,” Valery Claude, Senior Vice President of Scent Digital at International Flavors & Fragrances (IFF), told BeautyMatter.
But smell’s sentimental appeal also leads to heightened sensitivity around job disruption, or even potential replacement of employees, as AI becomes a more potent force in perfumery. Perfumers spend years honing their skills to understand the science required to concoct unique formulations. Even as AI proliferates, Dubois believes the industry will always require a human touch.
“I don’t want to be one of those perfumers that help do that,” Dubois said, referring to the fear of AI-related job disruptions. “But I want to be a perfumer who teaches others to use this tool and take it to the next level.”
Startups like Osmo, as well as some of the world’s largest fragrance houses, including Givaudan and IFF, have embraced AI in their processes to speed up research and development and train large language models on fragrance data to more accurately predict what will be most coveted by consumers.
At IFF, the earliest investments in AI centered on predicting which scents would be future hits with consumers. Later, they focused on the AI algorithms IFF would rely on to deconstruct and differentiate what “fresh” and “clean” would mean across markets. IFF spent years building a dataset to understand what combination of ingredients would meet a customer’s expectations for those key attributes, which vary regionally.
“The difference between R&D and fragrance creation is the lack of accurate data when it comes to olfaction,” Claude said.
Newer AI-powered tools include Jennie, a digital scent design assistant developed to help scent designers search, select, and optimize fragrances for clients. Scent designers are tasked with understanding consumer data and a customer’s brief and translating that data into a technical brief that is shared with a perfumer. With hundreds of thousands of fragrances within IFF’s portfolio, AI is intended to speed up this process.
Claude added that this tool is particularly helpful for the scent designers who manage a portfolio of hundreds of thousands of fragrances that already exist in IFF’s portfolio, and then take that data to deconstruct consumer desires in any given market. “This AI capability provides some actions and translates the consumer feedback into concrete, actionable insights that she—or he—can take to the perfumers,” he added.
Another tool within IFF is Ernest, an AI-powered digital perfumer assistant that helps streamline fragrance creation by offering real-time insights into up to 800 distinct data points covering performance, cost, regulatory considerations, and other factors.
“When you bring all that complexity together, you can appreciate why AI is fundamental in terms of providing our perfumers with the basic navigation systems that tell them to go through that complexity,” Claude said.
IFF’s AI tools rely on a multimodal approach, meaning the company leverages different AI models from various vendors to match the best outputs to each use case. Data scientists worked closely with perfumers to develop the algorithms and ensure that they would properly understand ingredients, design, and the basic structure of formulations.
Claude said IFF has used AI since as early as 2008, with initial applications focused on predictive use cases to better understand consumer preferences. More recently, he said that IFF is still in the process of rethinking the training model for young perfumers early in their careers, as AI will be a factor in how they work. “The challenge is not so much the adoption and usage,” Claude said, but rather, upskilling.
Johan Chaille de Nere, Givaudan’s Director of Digital Transformation for Fragrance & Beauty, told BeautyMatter that the fragrance manufacturer is using AI to “create a direct connection between what the consumer feels and what that means for the perfumers.”
Chaille de Nere’s early career focused on marketing, initially at toothpaste and dish soap manufacturer Colgate-Palmolive and then at Givaudan, which he joined in 2011. But his role later evolved to focus more on digital transformation, including the infusion of AI and ML capabilities, under the guidance of Chaille de Nere’s boss Maurizio Volpi, Givaudan’s President of Fragrance & Beauty.
“We are deeply convinced that AI won’t replace all the perfumers, but we are also convinced that perfumers with good AI support will be better than the perfumers without AI support,” Chaille de Nere said.
In 2021, Givaudan acquired the French-based start-up Myrissi, which developed a patented AI technology that can translate fragrances into color. Another AI-powered tool, called Carto, helps perfumers explore formulations through a visualization-driven creation experience.
“The colors are a language to make a bridge between olfaction and consumer preference," Chaille de Nere said. “This tool is able to predict this color perception and we can also predict the emotional resonance of our fragrances. This notion of prediction is purely AI-driven.”
Givaudan was able to extrapolate those predictions based on 30,000 tests that it had conducted in the past, asking consumers questions like: what color do you associate with the fragrance in front of you?
Myrissi wasn’t immediately embraced, Chaille de Nere conceded. “It takes time for our perfumers, for evaluators, even for our customers, little by little, to be comfortable with this approach,” he said.
But the increased popularity of AI chatbots like OpenAI’s ChatGPT and Anthropic’s Claude has helped create greater comfort with AI. “We have a shift in the mindset versus when I started four years ago,” Chaille de Nere said.
There are also some repetitive, productivity-focused AI use cases that Chaille de Nere said are critical to the business. There are instances when a local government may ban an ingredient, which would trigger a reformulation requirement for Givaudan. But AI is now used to speed up that formulation process, so perfumers can dedicate more of their time to higher-value tasks, including creating new customer formulas.
“We are very interested in using AI to augment our teams, automate recurring tasks, or to make connections between our sources of data to bring more inspiration, more suggestions, and more guidance to our perfumers,” Chaille de Nere said.
Perfumery felt like a far-off dream for Texan Wiltschko. “Because I grew up in South Texas, and not the south of France, I didn’t become a perfumer,” he told BeautyMatter. But he holds a PhD in Olfactory Neuroscience from Harvard University and said he envisions Osmo disrupting the fragrance house business model, which he believes is often inaccessible to smaller, emerging brands.
“It takes half a year to a year,” said Wiltschko of the iterative process from when a client brief is submitted until a fragrance oil is finally selected and then shipped to a factory to be manufactured and ready to sell to retailers. “There is not a lot of precision, or necessarily being able to directly guide what you want,” Wiltschko said. “And nor is there often a lot of data.”
One client that Osmo worked with is the Museum of Pop Culture in Seattle, which wanted to create a unique scent inspired by its guitar exhibition. “We created this beautiful, metallic, twangy, electric, unisex cologne scent,” Wiltschko said.
Wiltschko’s pitch to clients is that Osmo’s suite of AI models can interpret text and even images to guide and speed up olfaction. Osmo’s team has recorded over 5 million “sniffs,” capturing various aromas that the team tracks and inputs into Osmo’s proprietary AI models.
Scent, Wiltschko said, “is the molecular channel by which we communicate emotions and memories. And that’s what we’re trying to teach artificial intelligence.”
