Headlines about “AI-created perfumes” appear regularly. The reality inside a fragrance lab is more interesting, and more collaborative. Artificial intelligence is becoming a powerful assistant to perfumers – not a replacement for them.
Where AI genuinely helps
- Exploring ideas faster. Models trained on existing formulas and descriptions can suggest starting combinations for a brief, such as “a fresh green fig for summer”, giving a perfumer more directions to evaluate.
- Predicting smell from structure. Research groups have shown that machine learning can predict how a molecule is likely to be described by human panels, which may speed the search for new aroma materials.
- Checking constraints. Software can flag ingredients that exceed IFRA limits for a product category, or estimate cost per kilo while a formula is still on screen.
- Matching and reverse-engineering. Comparing a target profile against a large library of accords helps shortlist bases that get close to a reference quickly.
Where the limits are
- Smell is subjective and contextual. The same formula smells different on different skin, in different climates and in different bases.
- Data is incomplete. Fragrance formulas are closely guarded, so public training data is limited and uneven.
- Performance must be tested. Stability, longevity, discoloration and behaviour in wax or soap still need physical trials.
- Taste and story matter. A memorable perfume has intention behind it – something people bring.
How we use AI at Haveli AI
We use AI where it is strong: understanding what customers ask for, searching our 6000+ oils by profile, reading labels from photos and remembering confirmed answers. Formulation, evaluation and quality control stay on the perfumer’s bench in Mumbai. The combination lets us respond fast without compromising on how the oil actually smells.
Have a brief for a bespoke fragrance? Book a consultation with our team.