Every few months a headline announces a perfume “created by AI”. It is an exciting idea, and it contains some truth. But anyone who has worked on a perfumer's bench knows that the story is more interesting than replacement.
What AI already does well
- Search at scale. Ranking thousands of materials or finished oils against a brief in seconds.
- Pattern spotting. Learning from large archives of past formulas which combinations tend to work for a given market or price point.
- Constraints. Keeping a proposed formula within cost targets and regulatory limits, and flagging conflicts early.
Large fragrance houses have explored this for years. One of the best-known early examples was a collaboration between IBM Research and the fragrance company Symrise, whose AI system helped perfumers generate formula ideas that led to commercial launches in 2019.
What AI still cannot do
It cannot smell. Models predict how something will smell from data, but the final verdict is always a human one. Perfume is judged in a specific moment: on skin, in heat, next to other scents, through the filter of culture and memory. A rose that reads as romantic in Paris can feel devotional in Ajmer.
It cannot take responsibility. Deciding whether a blend is beautiful, whether it suits a brand, and whether it is safe to put on skin are judgements that brands and perfumers own.
The real model: perfumer plus tools
The most productive way to see AI is as an extremely fast assistant. It shortens the path from a brief to a shortlist, from a shortlist to a first trial, and from a trial to the fix. The perfumer spends less time searching and more time smelling, adjusting and creating.
How we work at Haveli AI
Our name says AI, but our oils are compounded and checked by hand. The AI finds the closest candidates from our 6000+ library and records what customers ask for; our perfumers evaluate, adjust and approve. You imagine. We compound. Book a session with our team to see the process.