Holzminden Perfumers are a special profession. They see themselves as artists, composing sensual fragrances by stirring together a few milliliters of high-quality ingredients – and thus creating emotions. There are only 2,000 of these specialists worldwide, and according to statisticians, they are even rarer than astronauts.
Now an artificial intelligence (AI) is supposed to help them with perfume production. Can a computer even do that? The fragrance and aroma manufacturer Symrise is convinced of this. Together with IBM he developed the Philyra program, named after a nymph from Greek mythology.
For the company from Holzminden in Lower Saxony, AI is “one of the company’s most important projects,” says Heinz-Jürgen Bertram, CEO of Symrise. The Dax Group is one of the world’s four major fragrance manufacturers, producing for perfume giants such as L’Oréal and Estée Lauder.
Philyra could shake up the otherwise more traditional fragrance industry. The last major innovation in the industry was the introduction of synthetic fragrance molecules in the late 19th century, says Tobias Koppitz, executive director of the German Association of Fragrance Manufacturers. “By using AI, the industry could make a giant leap.”
Top jobs of the day
Find the best jobs now and
be notified by email.
The AI should not only make perfume development more creative, cheaper and faster. It can also help to use more sustainable raw materials and meet increasingly complex regulatory requirements.
All fragrance manufacturers are working on artificial intelligence
Symrise is part of a trend, says Mirko Waschun, Head of Consumer Goods and Retail Europe at Kearney: “The entire consumer goods industry is increasingly experimenting with AI to increase efficiency, accelerate growth and offer customers more suitable products.”
Because the fragrance business is complex – and big. Market researcher Euromonitor International predicts that the $55 billion market will continue to grow well. More than 4,000 new perfumes hit the shelves every year, and Christmas or Valentine’s Day scents are only there for a few weeks.
All major fragrance and aroma manufacturers are researching AI. The Swiss company Givaudan calls the program “Carto”, Firmenich “Scentmate” and the US group IFF “Codex”. “AI is the key to unlocking innovation,” says IFF.
From the showcase to series production
Symrise hopes for a market potential of 1.7 billion euros and sees itself as a leader in the use of AI in perfume development. To emphasize this self-image, the company created a fragrance brand as early as 2019: Symrise launched the first perfume that was developed entirely by a computer. The large Brazilian perfume manufacturer Boticario distributed the fragrance.
>> Read more: Perfume creation using AI: Symrise’s future is called Philyra
It was more of a showcase for the research project, now the AI is to be used in everyday life. “Now it’s entering the second phase and it will become standard help for all of our perfumers,” says Symrise manager Christian Schepers, who manages strategic projects in the perfumery sector. The group now wants to offer more and more customers perfumes that were developed with the help of computers.
Perfume at the push of a button
The benefit of the technology is reflected in the complex working methods of the perfumers: they create new fragrances from around 3,500 raw materials, such as bergamot, rose oils or patchouli. A perfume usually consists of one to five dozen ingredients, the possible combinations are almost impossible to count. The perfumer keeps mixing different substances together and also adjusts the quantities to meet the client’s wishes. It takes months, sometimes years, to create a finished fragrance.
Thousands of new fragrances come onto the market every year.
(Photo: Photodisc/Getty Images)
Philyra aims to accelerate this and help the perfumers to select and combine the raw materials. The AI uses 3.5 million perfume formulas that Symrise has already developed and some of which have been brought to market. The computer also knows which combinations have sold particularly well in which region, in which age group and with which gender – and which have not.
If the perfumer wants to develop a women’s fragrance for the German market, for example, Philyra suggests twelve combinations of raw materials at the touch of a button, which the AI considers promising. The perfumer can choose whether the fragrance note should be similar to perfumes that are already available or should smell completely different. With the help of the computer, he develops fragrances further or creates new series.
A machine mixes the suggested combinations, and the human expert can smell them, fine-tune or replace individual ingredients. The computer learns which suggestions the human expert develops further and which not. The AI should improve the interaction and thus the creation of the scents. “The AI can give the perfumer a starting point more quickly, from which he can fine-tune,” says Symrise boss Bertram.
“The AI can give the perfumer a starting point more quickly, from which he can fine-tune,” says the Symrise boss.
(Photo: Symrise)
The company also wants to increase the creativity of its specialists in this way, says AI project manager Claire Viola: “The nice thing about Philyra is that she has no barriers in her head and can find combinations that perfumers have not yet come up with.” Perfumers biased, influenced by their culture and background. Colleague KI breaks through such barriers of thought and ensured, for example, that Symrise used the scent of cucumbers in perfumes for the first time.
AI can distinguish scents even though it cannot smell
Symrise supplies the fragrance formulas for Philyra, the IT partner IBM the algorithms. The basis is the Watson software platform. But how can the computer create scent combinations even though it can’t smell?
Symrise perfumers systematically evaluated the raw materials, creating a standard. IBM partner Otto Krischer, who is responsible for the project at the technology group, explains: “If ingredients smell similar, they have a smaller mathematical difference for the computer.” Philyra categorized the scents in more than 20 dimensions. The AI also knows how different raw materials smell when interacting. In order for this to work, those responsible spent months bringing the data into a suitable form.
The AI helps the perfumers to select and combine raw materials.
(Photo: Symrise)
Symrise boss Bertram admits that “a lot of lessons learned” during development. IBM partner Krischer adds: “Developing algorithms for scents is much more complex than for tastes.” In one case, the AI produced a candle scent instead of perfume. Due to the many degrees of freedom, it is intentional that unexpected scents sometimes arise, says Krischer. With the scent of candles, however, the technique was obviously a bit too creative.
Computer accelerates the basic work
The project is no longer just about ingenuity. In the next Philyra generation, perfumers can also set sustainability goals, explains AI project manager Viola. Then the computer would only suggest raw materials that grow back particularly quickly. “More and more customers are asking for it,” says Viola.
Philyra is already proving useful in the incomplete supply chains. If a raw material is not available, it is often not easy to replace it with another. The entire combination has to be adjusted, but the perfume should continue to smell the same. A perfumer would need weeks for this, the machine minutes. Philyra also helps with regulatory issues, for example if a customer no longer wants a particular raw material.
“The nice thing about Philyra is that she has no barriers in her head and finds combinations that perfumers have not yet come up with,” says the AI project manager at Symrise.
(Photo: Symrise)
And if a cosmetics manufacturer asks for a cheap perfume, the computer can suggest cheaper ingredients to the perfumer. There are also plans to integrate further market data in order to develop perfumes in an even more targeted manner. Symrise boss Bertram says: “All of this makes the basic work easier for perfumers.” They would have more time to take care of the actual perfume development.
More efficient production by Philyra
Philyra should also make development at Symrise more efficient. Companies like L’Oréal usually get suggestions from many fragrance manufacturers when they want to launch a new perfume on the market. For inquiries like this, Symrise develops 60 to 70 samples, and they submit the best to the customer. This happens 100,000 times a year – Symrise always has to wait and hope to get the order. Symrise also has to bear the development costs for rejected fragrances.
>> Read more: Economic miracle of the 2010s: Why the unknown Dax newcomer Symrise has been growing for years
The Holzminden-based company therefore hopes to be able to develop the samples more quickly with the help of AI, to better implement customer requests – and to be awarded the contract more often. However, it is still too early for a record of success.
AI project manager Viola reports further advantages: Customers in Asia in particular are very interested in computer-developed perfumes and even advertise them on the product. This is how you can stand out from the competition. However, most cosmetics manufacturers do not specifically mark such fragrances as AI products.
Detergent and shampoo by AI
So far, Symrise has invested “a decent seven-digit amount” in the project, says strategy manager Schepers. “We are now at a point where the project is financially viable.” The group wants to use the technology in other areas in the future. The group also produces fragrances for detergents and cleaning agents, soaps and shampoos.
Here the use of the computer is much more complex. “Fine perfumery is all about smell,” says Schepers. First and foremost, detergent has to clean and the scent only comes into play at the end of the wash cycle. However, this must remain stable throughout the wash cycle. “The PC still has to learn more.” If Symrise succeeds in taking this step, the group believes that the AI could serve a market worth a total of 13 billion euros.
The Dax group is based in Holzminden in Lower Saxony.
(Photo: dpa)
So will perfumers soon be replaced by computers? This concern is unjustified, insists Symrise boss Bertram: “Humans and machines are more efficient when combined and mutually enrich each other.” The AI is just an additional tool. 50 of the 80 Symrise perfumers use this in their work.
Even the industry association does not fear any layoffs because of the technology. “For perfumers, the AI is a partner, but not a competitor,” says Anneliese Wilsch-Irrgang from the German Society of Perfumers. Man is still necessary as a controlling authority. Especially with luxury fragrances, the perfumer himself is a selling point.
One last, crucial question remains: Who makes the better fragrance, Philyra or the perfumer? AI project manager Viola remains modest: “It’s the collaboration. Technology increases creativity, but it takes humans to read and improve the results. Symrise boss Bertram is also diplomatic: “It’s different. No better and no worse.”
More: The Voice Miracle – How technology creates new business models around language