Symrise: AI to create new perfumes
For a long time, fragrances have been created by talented perfumers all over the world. Now, AI analyzes the formula and creates new scents that have a high success potential.
Once thought to be the art perfumers carefully crafted, fragrances are now created with a help of algorithms and artificial intelligence. In 2019, a Brazilian cosmetics giant O Boticário started selling perfumes which were created by AI.
Symrise, the No.1 supplier of fragrance raw materials, was the one who supplied its new AI-created scent to O Boticário. The company has been serving fine fragrances in the world, from DKNY Be Delicious to Caroline Herrera 212 VIP Men, with its strong, creative and collaborative team of perfumers, scientists and artists. [1]
In 2018, the company has collaborated with IBM to develop an AI-based system called “Philyra” to help perfumers develop new fragrances. The system was trained as human perfumer would: it was trained to recognize similarities among the formulas for 1.5 million existing fragrances. Then, these formulas were combined with various other data, such as the application areas, sales figures and so on. With the algorithms, the system can design a new fragrance that has the highest potential for success within a second. [2]
The fragrances that were created by Philyra are then treated in the same way as the human-created fragrances. They will go through the rating process by human experts and then the feedback was fed back into the system to improve the scent.
This new way of creating fragrances certainly creates new values. Now, with the algorithm, it has become possible to create a very targeted product which is predicted to have a high potential for sales. Instead of relying on the “intuition” of the perfumers and requiring a long period of try-and-errors, it relies on a solid database which can come up with different combinations in few seconds. In addition, the level of accuracy would improve further with the increase in new data and the system’s knowledge about perfumes will definitely surpass that of human perfumers. Therefore, it might be more cost and time efficient for the company to further develop the system than training new human perfumers.
Would this be the winning asset in the fragrance space? That is still questionable. First, as mentioned, the process of selecting the final product is still human based. The AI-based system is now only providing additional candidates which can be rejected by human perfumers. If the scents created by the AI constantly lose against other human-created scents, the investment put into developing the AI may not be justified.
Second, especially in the space of fine fragrances, the winning recipe for fragrance is not just the scent itself. It is also about the brand, the story behind the product, the aesthetic of the bottle, how the high-esteemed celebrities talk about it etc. It is not clear how the algorithm can incorporate these qualitative aspects to provide the best fragrance for the clients.
Third challenge is more on the industry culture. Would their traditional clients, which are big luxury brands all over the world, appreciate the scent created by AI? Or would they trust the craftmanship of human perfumers? It seems that many of the designers in the traditional luxury brands still value the art of craftmanship more than anything and may not want to use the scent created by data. In addition, would the brands be able to price its fragrances at a high price if the customers know that the product is created by the AI-system? If not, then Symrise may not be able to sell the scents to the brand at a high price neither.
Influencer, Brazilian Actress Giovanna Lancellotti, holding the perfume
To overcome the third challenge, Symrise has chosen a bold path to work with a big company to push this innovative approach. It collaborated with O Boticário, the world’s number three in the industry, to bring the AI-created perfumes, Egeo ON Yout and Me, to the market. As the perfume was targeting millenniums, O Boticário put a lot of energy into the launch to create a buzz by leveraging top influencers. The performance of the perfume is not certain, but the fact that it received the rating of 3 on Fragrantica[3], an online encyclopedia of perfumes, poses us the question whether the algorithm’s prediction capability is accurate enough to get trust from human perfumers.
Although it is not clear whether the application of data analytics will take off in the fragrance industry, having such capability would definitely act as a competitive advantage. Especially now that new generations value personalization more and more, the value of database is high. The company should consider what is the best way to leverage its digital assets to further increase its value proposition.
[1]”Fragrance”. Symrise.Com. https://www.symrise.com/scent-and-care/fragrance/.
[2]Woodie, Alex. 2018. “The Scent Of An AI”. Datanami. https://www.datanami.com/2018/10/23/the-scent-of-an-ai/.
[3]”Egeo On You O Boticário Cologne – A New Fragrance For Men 2019″. www.Fragrantica.Com. https://www.fragrantica.com/perfume/O-Boticario/Egeo-On-You-55431.html.
Thanks Kanako, very interesting read! It’s an interesting parallel to the art vs science question we saw in a couple of our cases, including the futbonaut case. I don’t know much about the art of creating a fragrance, but I’d venture to say that a vast majority of the customer’s perceived value of a fragrance comes from the brand, packaging, container design, etc. While in this case AI certainly speeds up the development process of the fragrance itself, I wonder if science will continue to gain ground vs the “art” portion of the product by eventually suggesting fragrance names, color schemes, designs, etc.
Hey Kanako, I found this so interesting. Reading this, I couldn’t help but drawing parallels to how Netflix used big data to identify to create TV shows like “House of Card”. I definitely believe that there’s room for disruption as an algorithm like this is able to tease out the preferences of masses. It’s like using wisdom of the crowd to define what sells best. However, I wonder if this would ever replace perfumers in high-end fashion houses. I think a lot of the value of high-end fashion and fragrances comes from the process – a well-trained perfumer goes through the elaborate process of testing and iterating different fragrance notes to get to the perfect one. I think shoppers like the idea that their perfume did not come by easily. I’d be curious to see what the future holds for this algorithm.