Creating a new perfume has traditionally been the work of a small number of highly trained perfumers, who may spend months or years balancing hundreds of raw materials. The large fragrance houses are now using artificial intelligence to shorten that process, with systems that analyze data on thousands of existing formulas and propose combinations a human might not have tried.
How the systems work
The AI tools in use at fragrance companies are trained on proprietary databases of formulas, the chemical ingredients they contain, how those ingredients behave together, and how consumers in different markets have responded to finished products. From that data, machine learning models learn which ingredient pairings tend to work, how intense and long-lasting each component is, and which profiles appeal to particular audiences.
The models do not smell anything. They identify patterns in historical data and generate candidate formulas that a perfumer then evaluates, adjusts and refines. The companies describe the role as an assistant that broadens the search for new ideas, not a replacement for the trained nose that makes the final decisions.
Symrise and Philyra
German fragrance and flavor supplier Symrise developed its system, called Philyra, in partnership with IBM Research. Symrise describes Philyra as AI-powered perfumery that sifts through its formula library to suggest new fragrance structures and unusual ingredient combinations. The first commercial perfumes created with Philyra were launched for the Brazilian cosmetics company O Boticario in 2019, and Symrise says the tool has since become part of how its perfumers work, helping them explore directions faster and reach novel results.
Firmenich and Scentmate
Scentmate, created by Swiss fragrance house Firmenich (now part of dsm-firmenich), takes a more customer-facing approach. It is an online platform aimed at smaller brands and entrepreneurs that want a custom fragrance without the long development cycle of a traditional brief. A client answers a set of questions about the target market, the emotional impression they want and the product the scent is for, and the platform draws on Firmenich’s expertise and data to propose a selection of fragrance options. The company positions it as a way to compress a process that can take months into days.
A faster path to market
For the industry, the appeal is speed and breadth. A new fragrance that once took years to develop can be prototyped in weeks, and perfumers can test more directions than they could by hand. Firms also say the tools help spot combinations with commercial potential that would be unlikely to emerge from conventional trial and error.
Skeptics within the fragrance world note that the data these systems learn from reflects what has sold before, which risks steering output toward the familiar. The companies’ answer is that the human perfumer remains in charge of the creative decisions, with the AI expanding the options on the table.
Beyond perfume
The same techniques are spreading to adjacent products. Flavor houses, including Symrise and dsm-firmenich themselves, apply similar models to beverages, confectionery and other food products, and cosmetics companies use AI to tailor formulations to consumer preferences. The direction is toward scents and flavors that are increasingly data-driven in their design, with human experts deciding which of the machine’s suggestions are worth putting in a bottle.
