How AI is making perfume smell better

In a repurposed shop along Ginnekenstraat, a pedestrian shopping street in the Dutch city of Breda, visitors can fill out a brief personal questionnaire and leave less than 60 minutes later clutching a bottle of perfume that simply did not exist when they walked through the door.

The questions are nothing like what a typical sales assistant would pose. What colour embodies you best. If you could travel anywhere right now, where would it be. Describe your personal style.

You provide your answers, a suite of algorithms interprets them, and a machine inside the room blends a custom scent accordingly, filling and labelling the bottle as you wait.

The firm behind this experience is Scentronix. For nearly a decade, the company has argued that the conventional way people buy perfume is far stranger than most realise. Its founders, Dutch artist and filmmaker Frederik Duerinck alongside scent designer Anahita Mekanik, frame this strangeness as a pointed question: why should roughly 800 individuals dictate how 8 billion humans smell?

They are referring to the exclusive guild of master perfumers, often called “noses,” who formulate virtually every fragrance found on store shelves. It is a deliberate provocation, and like the best provocations, it contains a genuine insight. For much of its history, perfume has remained a closed art, beautiful yet distant. Now, software is quietly prying it open.

That statement often makes people uneasy. We have been conditioned to dread the arrival of code in a craft built on human skill. The common fear is replacement: the algorithm arrives, and the artist is shown the door. In the world of perfume, that is not the reality. The closer you examine who is actually building these tools, the more the opposite picture emerges.

The industry’s biggest players reached the same conclusion years ago. In 2019, German fragrance house Symrise paired its perfumers with an artificial intelligence system developed alongside IBM Research. Named Philyra, from Greek mythology, the AI was trained on a vast database of formulas and performance data. Unburdened by habit or taste, it could suggest pairings no human would instinctively choose. Working alongside it, Symrise perfumer David Apel created two fragrances for the Brazilian brand O Boticário, released under the Egeo line for the country’s Valentine’s Day. They are widely considered the first AI-formulated perfumes ever sold commercially.

Others soon followed with their own machines. Givaudan, the world’s largest fragrance house, developed Carto, a touchscreen system that visualises a formula as a map and feeds it to a robot that mixes a physical sample in seconds. This lets a perfumer test an idea almost as quickly as they can conceive it. Calice Becker, creator of Dior’s J’adore and director of Givaudan’s perfumery school, has stated the tool’s purpose is to encourage perfumers to dare, to try combinations that would never have been obvious choices. Firmenich, now part of DSM-Firmenich, took a different approach with Scentmate, a service designed to help small brands and solo entrepreneurs, who lack laboratories and in-house noses, create a fragrance from scratch.

Not everyone is impressed, and the criticism deserves attention. Jean-Claude Ellena, the former in-house perfumer at Hermès and one of the most respected noses alive, has argued that a machine cannot read the thoughts guiding a perfumer through a composition. He has expressed sadness, pitying the junior perfumer who will one day be handed a machine’s draft and asked to perfect it. Coming from a man who treats perfume as a form of literature, the objection carries weight. There is a real danger that automation flattens a craft into a workflow, optimising away the strange, intuitive leaps.

However, that worry assumes a contest between human and machine. That is not what these tools represent. Every one of them keeps the perfumer in the room. Symrise calls Philyra an apprentice, not a replacement, and seems to mean it. Carto places the formula on a screen, but a person still decides what is beautiful. Even Scentronix, the most automated of the group, routes roughly one customer in 50 to a human perfumer to correct whatever the algorithm has misjudged. The software widens the canvas. It does not sign the painting.

Beneath the commerce, something genuinely new is emerging, and it is the part that should interest anyone who cares about technology as much as perfume. Smell is the sense that has always resisted the machine. We taught computers to see and hear decades ago, but odour, a chaos of molecules binding to receptors we still only partly understand, remained stubbornly analogue. That is changing. Researchers at Google have trained neural networks to predict how a molecule will smell from its structure alone, a first rough sketch of a machine nose. A European project called Odeuropa has used AI to recover the lost scents of historical Europe from centuries of text. Perfume is simply the most commercial edge of a much larger effort to give software a sense it has never had.

The market these innovations enter is large and quietly conservative. Industry estimates place global fragrance sales at around $60 billion a year, a business still shaped by a handful of houses, a rotation of celebrity licences, and the same few hundred noses deciding what the rest of us wear. Set against that, a system that lets a teenager in a pop-up shop, or a small brand with no money for a laboratory, make something that smells like them and only them is not a threat. It is an enlargement. The pie does not shrink when more people are allowed to bake.

Which brings us back to the shop in Breda. The machine there does not know what your perfume ought to smell like. It knows only what you told it, and it is honest about the gap, which is why a human is kept on hand for the moments when you and the algorithm disagree. What it offers is not a verdict but an invitation, a chance to treat the oldest and most intimate of the senses as something you compose rather than something you choose from a shelf. You walk in as a customer. You walk out, an hour later, holding a small bottle that smells like an answer to a question only you were asked, and that did not exist anywhere in the world when you woke that morning.