■ SAFE RISK ■ Creative Arts
AI has become a serious formulation partner in perfumery, proposing accords and predicting consumer response. It still cannot smell, and the decision about what a fragrance should mean stays with the perfumer.
“Electronic noses detect molecules. Your trained nose detects emotion. Different noses.”
Our AI replacement risk score — how we score jobs
A perfumer works from a brief — a brand, a price point, a target consumer, a regulatory envelope — and builds a formula from hundreds of raw materials, naturals and synthetics, each with its own volatility, tenacity and interaction effects. The work is iterative: weigh a trial, smell it on blotter and skin over hours and days, adjust ratios, re-trial. Then come the constraints: allergen and IFRA compliance, cost per kilo, stability in the actual base product, performance in a shampoo versus a fine fragrance, and revisions demanded by a marketing team that cannot articulate what it dislikes.
AI has arrived meaningfully here, earlier than in most creative fields. Major fragrance houses have publicised systems that suggest formulas by learning from decades of internal formula archives and consumer testing data — proposing accords a perfumer might not have tried, and predicting which combinations will test well in a given market. Gas chromatography and mass spectrometry already reverse-engineer competitor products. Regulatory compliance checking, cost optimisation and raw material substitution when a natural becomes scarce are all computational tasks now handled far faster than by hand.
The limits are physical and interpretive. No sensor smells; instruments identify molecules while perception is a construction of the human olfactory system, with anosmias, adaptation effects and context dependence that instruments do not model. Translating 'we want it to feel like the first warm day of spring, but modern' into materials is an act of interpretation. And skin performance over eight hours, on different people, remains an empirical human test. Our score of 20 reflects genuine AI assistance compressing the iteration cycle and reducing junior evaluator roles, while the perfumer's authorship holds.
Automatability: our editorial assessment of current and near-term AI capability
AI formulation assistance is already deployed at the major houses and will be standard by 2030, compressing development cycles and thinning junior evaluation and lab-technician roles. Senior perfumers remain central past 2040 because authorship, brief interpretation and human olfactory evaluation cannot be delegated. Expect fewer people producing more fragrances, with the training path into the profession getting narrower.
It can propose formulas, and major houses use systems trained on decades of formula archives and consumer test data to do exactly that. What it cannot do is smell the result. Every proposal still goes to a perfumer's nose, on blotter and skin, over hours — because perception is not the same thing as molecular composition.
They threaten some analytical work, not composition. Gas chromatography and mass spectrometry identify what is in a fragrance with great precision, which is genuinely useful for reverse-engineering and quality control. But knowing the molecules present tells you nothing about whether a composition is beautiful, or how it will behave on a specific person's skin.
It was always a tiny, competitive profession with a long apprenticeship, and AI has made the junior rungs — evaluation, lab work, compliance checking — thinner. Entry is harder, not closed. Independent and niche perfumery has grown as a route, where a distinctive authored style matters more than a house training pipeline.
It compresses them substantially. Trial iterations that took weeks of manual reformulation and compliance checking can be narrowed computationally before anything is weighed, and consumer response prediction reduces the number of rounds needed. The consequence is more launches per year with smaller teams, which is good for output and hard on employment.