The Future of AI x Fragrance Showcased at World Perfumery Congress 2022

The Kaorium system enables a poetic interactive experience that invites consumers to smell, reflect and experience, building a new relationship to fragrances in a retail setting.
The Kaorium system enables a poetic interactive experience that invites consumers to smell, reflect and experience, building a new relationship to fragrances in a retail setting.

Computational power and large amounts of data have led to breakthroughs in areas such as drug discovery, autonomous driving, machine-generated faces and voices, as well as the customization of your favorite social media feeds. But when it comes to computationally created fragrances, can we smell them yet?

In June 2016, I asked the president of innovation at a large fragrance and flavor supplier whether they believe that robots will soon be able to create perfumes. They said they hoped to no longer be around if it ever were to occur. 

In July 2022 at the World Perfumery Congress (WPC) in Miami, Symrise senior perfumer Dave Apel told me of the moment he knew the company’s digital perfumery assistant, Phylria, worked: “I could relate to what the machine was doing. I watched it do something I studied 40 years for.” 

Futurist Roy Amara once said, “We tend to overestimate the effect of a technology in the short run and underestimate the effect in the long run.”  My two conversations, six years apart, and Amara’s quip suggests that we have underestimated the potential of what lies ahead in the world of fragrance. Specifically, groundbreaking innovations are in the making when it comes to machine learning (ML)--a subfield of AI--that gives computers the ability to learn without explicitly being programmed. 

This is not only great news to our noses, but also to many pharma CFOs because applications for the chemical industry are already leading to a tremendous reduction in time and R&D costs. For designers of flavor and fragrance, increases in creativity, sustainability and scent discovery are to be expected, with a vision to use ML for molecular synthesis as well.

The Early Days of F&F & AI

More than a decade ago, an article systematically identified the ingredient combinations responsible for the taste palette of regional cuisines from a molecular level (F-1). It was the most downloaded essay across all Springer Nature journals, exceeding 100,000 PDF and page views in the first four weeks of its publication. The paper also buzzed across academia; I witnessed it cited at presentations from Oxford University to MIT. 

A few years later, Chef Watson, powered by IBM’s Watson, an AI supercomputer, was born. It understood food at the molecular level; learned about culinary traditions, people’s likes and dislikes; and could compute unique flavors that would elicit surprise and enjoyment in consumers. Thereafter, however, progress was largely silent for almost a decade. 

Meanwhile, in the fragrance realm, the first AI-developed fragrance hit the market in 2019, courtesy of Symrise and IBM. At the same time, companies investing large sums into research and manufacturing facilities catering to AI have made headlines. Which makes me wonder: what is brewing behind the scenes, waiting to make the next big impact in fragrance x AI?

The World Perfumery Congress 2022 (WPC) in Miami, FL, showcased three technologies that highlighted how the business of scent is being digitized from the supplier through to the consumer. What was once the province of intuition, taste, acquired skill and institutional knowledge is now being enhanced by emerging technologies. 

Augmenting the Perfumer

Creativity is hard to achieve, yet easy to recognize. A successful fragrance fulfills human desire and adds an element of surprise. This element of surprise is where creativity comes in. Therefore, if a machine can identify and create something that has never been done before, then it can be argued that computational creativity exists. Google’s Alpha Go, a program for the board game Go, beat the world master and led him to admit that he has never before seen the playstyle chosen by his artificial opponent.

One example is Philyra, a program jointly designed by IBM and Symrise to augment a perfumer’s expertise and achieve higher creativity and sustainability in the creative process. The platform uses machine learning to help identify patterns and generate novel combinations of fragrance formulations. The program is led by Richard Goodwin, Ph.D., principal researcher of the Computational Creativity Research Group at IBM T.J. Watson. At Symrise, an internal cross-disciplinary group runs the effort, including VP senior perfumer Dave Apel. 

Symrise VP senior perfumer, Dave Apel, is part of an internal cross-disciplinary group that runs the Philyra program.Symrise VP senior perfumer, Dave Apel, is part of an internal cross-disciplinary group that runs the Philyra program.

The platform’s data is encrypted, consisting of 20-plus dimensions of more than 2,000 raw materials and 3.5 million fragrance formulas. The algorithm learns from existing formulas, raw materials, historical success data and industry trends. Use cases include creating a new signature fragrance or flanker strategy based on white space analysis, as well as focusing on a micro-target or sustainability.

The starting point for this tool is the fragrance formula, known as the “seed formula.” Once the formula is added to the system, the perfumer can mark how far away from the original they are looking to push the boundaries in their creation. 

To illustrate, Apel draws a set of concentric circles on a piece of paper and marks dots in the circles. Each dot represents a fragrance. The further away from the center, the further away we move from the success criteria. He then draws another set of concentric circles that overlap with the prior set. Apel proceeds by adding dots to the areas where the circles overlap. These, he says, are all unique possibilities of fragrances. Do we want the new fragrance to resemble what has been deemed successful in the past or are we willing to accept larger deviations and potentially be rewarded with entirely new creations? 

Phylria can suggest alternative materials, considering dosing for each ingredient based on usage patterns, pleasantness and novelty of the fragrance by comparing it to a large set of commercially available fragrances. The technology also takes into consideration the technical qualities of raw material, including the molecular bloom, weight, vapor pressure and threshold of human perception.

The clean interface presents 12 new fragrance formulas per input. For a creative exploration, a perfumer can analyze the output formulas and choose a selection of them to be compounded in order to smell them. The operating perfumer can also direct the Phylria platform to generate new combinations of fragrance formulations that fit specific design objectives—for example, creating a unique fragrance for Brazilian millennials.

Niche vs Mass Applications of AI Perfumery

O Boticario, a top global beauty and fragrance company, launched the first AI fragrance in collaboration with Symrise in 2019. At the time, the database contained more than 2 million fragrance formulas. Phylria tapped a warm milk ingredient that was available to perfumers from the flavorist palette and added fenugreek for the coumarin sweetness always found in a fougere accord in a way that has never been used before in this context. It was a moment of surprise for Apel; he did not know the flavor ingredient existed and never would have thought of adding it to create this olfactory effect. 

Phylria thus proved that machinal creativity in the world of fragrance is possible. With minimal intervention of the perfumer, this fragrance composition was tested against a demographically pleasing benchmark, succeeded, and was brought to market. Alas, the sales figures of this fragrance did not meet expectations. When speaking to Symrise about this case, Apel emphasized that AI-assisted fragrances are in their infancy and fragrances created for broad commercial appeal probably do not meet the criteria of generating great surprise with novel notes. 

On the other hand, the application of this technology to niche fragrance creations that generate tremendous love/hate reactions may be more appropriate. Having said that, Philyra offers insight into accords and combinations that play upon our emotions in a very targeted way. This type of development is the cutting edge of commercial development.

AI-driven Regulatory & Sustainability Compliance

In addition to pushing the creative boundaries, Phylria algorithm also suggests ingredient substitutes for formulas to comply with new safety and regulatory standards. It can, for instance, substitute in more sustainable materials with higher biodegradability and renewability scores, all while keeping a similar olfactory profile. This offers an unparalleled opportunity to design fragrances to be more environmentally conscious, potentially offering significant cost savings and increasing consumer demand.

An Olfactory Compass

While Philyra’s starting point is a fragrance formula composed of fragrance ingredients and their respective weights, the Aryballe odor analytics platform’s starting point is the fragrance itself. 

During a demo at WPC, a tube attached to Aryballe’s technology was placed inside a bottle to assess the headspace, i.e., the air above the fragrance. Within a second, the names of the molecules present appeared on an attached display. Just as quickly, upon removal of the odor, the graph showing the molecular reading reduced to zero, demonstrating the swiftness of the computational analysis and the sensitivity of the measuring device.

Aryballe’s odor analytics platform is designed to assess the headspace of a fragrance.Aryballe’s odor analytics platform is designed to assess the headspace of a fragrance.

Aryballe is building a digital library of odors. The company has raised $18 million in total, including $7 million in its series B in 2020, with investors including Hyundai, Samsung, Asahi and IFF. Its suite of clients comprises several large fragrance houses. 

Sam Guilaume, the co-founder and CEO of Aryballe, shared, “We don’t know yet how many dimensions the odor space will have.” 

With one sniff, the company’s sensor technology can identify an odor or fragrance not just based on single compounds but based on the way our nose detects scents through our receptors. It then situates this odor or odor composition on a digital odor map in relation to other fragrances or odors. One can zoom in to see similar odors and zoom out to view clusters of different odor groups. 

Guilaume further explained the nuances of the technology, “Imagine you are using your GPS, it knows where you are on a continent, in a city and will help you navigate the space. Here, there are different continents, the vanilla continent, and the fruity continent for example.” 

The technology can be a compass. As a perfumer’s tool, it can eliminate the step of a perfumer inputting a formula. It also allows them to work with more formula candidates and visualize the odor profile before executing a new fragrance, thereby greatly reducing time and cost associated with fragrance creation.

Aryballe’s technologies are designed to smell and identify an odor through a combination of biochemical sensors and advanced optics coupled with machine learning. The company’s latest device,The NeOse Advance, introduced in 2021, is the first product based on Aryballe’s silicon photonics-based platform. 

The company’s proprietary peptides, once grafted onto silicon, are able to respond to a wide variety of odors. Designed with heightened stability, Aryballe’s peptides are said to be appropriate for a wide range of operating conditions and mimic the sensitivity of the human sense of smell. If your nose can smell it, so can the sensor. 

The same sensor can be used to distinguish lavenders as well as to detect malodors in cars. Databases of odors can be created to fulfill specific client and industry needs. 

Aryballe’s odor sensor can be utilized to identify counterfeit fragrances, for example. It does this by not only analyzing the odor composition but by capturing each step of the perfume’s evolution from top, mid to base note, based on volatility. This enhances the capability to distinguish between an original and a counterfeit because the counterfeit will diffuse differently, which can be captured better over time. 

For automotive applications, Aryaballe can help capture the odor data of materials to aid in vehicle part standardization and assist auto manufacturers in meeting their sustainability goals by monitoring recycled materials, which are known to emit smells that are undesirable to consumers. 

In consumer appliances, the technology can enhance cooking appliances by adding automation features based on odor analysis to, for example, know when cookies have finished baking or when food in a refrigerator has gone bad. 

Translating Scent into Words

Even the most novel, sustainable and beautiful fragrances will accumulate dust on product shelves if they are not embraced by customers. With almost 5,000 new fragrances launched between 2020 and 2021, it has become increasingly difficult for consumers to find fragrances they will love. In addition, fragrances are able to elicit highly emotional responses, yet are difficult to put into words.

In a retail setting there still has not been much innovation beyond transferring an online quiz onto an in-store touch display or creating a blind smelling experience in which consumers can smell fragrances without knowing which brand or fragrance they are currently smelling—thereby eliminating visual and brand biases. This is where Kaorium by Scentmatic from Tokyo, Japan, enters the stage. 

Kaorium is a digital table system with perfume bottles displayed in a semi circle on top. The system enables a poetic interactive experience that invites consumers to smell, reflect and experience, building a new relationship to fragrances in a retail setting. 

Attendees at WPC who interacted with the display gave it a 9.4 (out of 10) satisfaction rate and a net promoter score of 75. The technology is currently part of Shiseido Open Innovation Program and R&D center and is being used to develop new fragrances based on consumer data. 

Scentmatic collaborated with Nose Shop in Shinjuku & Ginza, Tokyo. The Kaorium display took the center stage and increased store entry rate by 139% and sales conversion rate by 287%. Scentmatic CEO Toshiharu Kurisu’s mission is to help people appreciate scent more and help them more easily discover the scent they like. 

When asked whether there had been additional advertising, Shin Watanabe, director at Scentmatic, says that the company only marketed their display on social media. The post was picked up by an influencer account on Twitter, which generated additional buzz.

Watanabe added, “People are not ready to choose in [the] current buying experience. We are here to help solve this problem.” 

What is the Kaorium experience? Here’s a breakdown:

  • Step 1: The consumer selects their preferred fragrance from a selection of unmarked bottles and places it on a digitally connected coaster. 
  • Step 2: Various words that express the fragrance (such as “romantic” and “warm”) float onto the table display. The consumer selects the word that to them best represents the fragrance that they just picked.
  • Step 3: The display then illuminates a selection of fragrances and asks the consumer to smell these and choose the one that best matches the previously selected word.
  • By repeating this process of consciously smelling fragrances, reflecting and connecting them to words, the system derives a pattern of the consumers individual perception and builds a database of word-scent associations.

The final output is an AI-generated poetic phrase that brings to life a sensory scene based on three fragrances and words selected in the discovery process. The names of the fragrances are only then revealed.

This AI-enabled consumer experience bridges the sensorial and rational mind and when asked what sets Kaorium apart from other AI technologies, Kurisu explained, “It invites consumers to smell, removing memories and biases, and allows them to take a moment to learn about their fragrance preferences.” Such a process that requires presence and focus enables consumers to form a new relationship to fragrances.

As well as allows people to learn an olfactory language, build the confidence to express their preferences and help them find their next scent.

Kurisu continued, “Scentmatic’s goal is to create a dictionary for perfumes that translate words to scents and scents to words.” 

The tech combines gathered data from the consumer experience, coupled with professional, marketing data and Japanese novels. By using Kaorium’s latest version, which can fit on a makeup counter, retailers can increase in-store engagement, education and improve sales while gaining new insight about consumers’ preferences. 

Learning how to smell fragrances without visual influences and how to express preferences in a world where at least 14 new fragrances are launched every single day holds more potential than we can currently foresee.

The World Ahead

Great progress has been made in bringing computational power, data and scent design together. The fact that machines can support perfumers and consumers alike is hardly a surprise on its own. However, the extent to which these novel approaches can expand the range of applications is striking: All aspects—including computer-aided purchase decisions, safety, cost and timing in R&D, digital signatures, certifications, sustainability, and creativity—are affected. The future of smell will involve the synthesis of expert knowledge, data-driven scent development and an increase in consumers’ sensory awareness, even when the scent of this composition is yet to reach our perceptual threshold.

When I ask Apel, “What is the future of AI-assisted fragrance design?” He simply replies, “Everything.”        

More in Regulatory & Research