The essentials
An algorithmic perfume is a composition designed with substantial assistance from machine learning software that proposes ingredient combinations, accord structures, and finishing choices based on training data drawn from previous formulas, sensory evaluations, and market performance records. The perfumer then evaluates, refines, and finalizes the composition through standard skin-testing and modification cycles. The term distinguishes such compositions from traditional formulation, where the perfumer operates without algorithmic suggestion at the ideation phase (Perfumer & Flavorist, accessed 2026-05-29).
The two most documented platforms in industry use are Symrise Philyra, developed in partnership with IBM Research and deployed from 2018, and Givaudan Carto, deployed from approximately 2019. Both systems are trained on the very large internal databases of historical formulas, sensory evaluations, and market performance that the major fragrance houses have accumulated over decades. They generate candidate accords that perfumers test, modify, and eventually deliver as finished compositions. Firmenich, IFF, Mane, and Robertet have developed analogous internal systems with less public communication about their specific architectures.
Algorithmic perfumes have appeared in publicly identified commercial launches including O Boticario Egeo On Me (Brazil, 2019), developed by Symrise using Philyra outputs as ideation starting points. The integration of algorithmic suggestion into formulation does not eliminate the perfumer's role: it shifts part of the initial ideation phase toward software while preserving evaluative judgement, skin-testing iteration, finishing decisions, and the final aesthetic responsibility with the human perfumer who signs the formula (Symrise corporate communications, accessed 2026-05-29).
Defining the algorithmic perfume
The algorithmic perfume category covers compositions where machine learning software contributes substantially to the initial ideation phase by generating candidate accords, suggesting ingredient combinations, or proposing structural propositions that the perfumer then evaluates and modifies. It does not cover compositions where software is used only for data management, market analysis, ingredient cost optimization, or formulation verification against IFRA Standards, all of which are now operationally standard across the industry and do not change the creative process.
The boundary between algorithmic and traditional formulation is interpretive and increasingly porous. Major perfumery groups use software extensively for ingredient databases, regulatory compliance verification, historical formula search, and stability testing simulation as part of normal workflow. The algorithmic-perfume designation typically applies when software contributes to the creative ideation phase itself rather than only to the operational and regulatory support around it, and the term is most often used by communications teams to signal innovation rather than as a precise technical classification.
The dominant industry platforms
Symrise Philyra was developed jointly with IBM Research and deployed in industrial use from 2018. The system trains on the Symrise formula database, which contains decades of compositions and the associated sensory and market data, and generates candidate accords matched to client briefs through machine learning models that the company has not fully disclosed in technical detail. Several Symrise perfumers have publicly discussed working with Philyra in industry interviews. The Brazilian commercial launch Egeo On Me from O Boticario in 2019 used Philyra outputs as ideation starting points and remains the most extensively documented public example (Perfumer & Flavorist, accessed 2026-05-29).
Givaudan Carto, deployed in industry use from approximately 2019, operates with a comparable architecture trained on the Givaudan formula database, with a user interface oriented toward perfumer workflow rather than data science workflow. The platform generates accord suggestions and visualizations that perfumers refine through their normal iteration process. Firmenich and IFF have developed analogous internal systems alongside dedicated machine learning teams, but they have communicated less publicly about their specific implementations and the commercial launches that have used them.
Documented algorithmic launches
Egeo On Me by O Boticario (Brazil, 2019), developed by Symrise using the Philyra platform, is the most documented algorithmic launch publicly identified as such by both the brand owner and the supplier. The composition was finalized by Symrise perfumers working from Philyra outputs as initial creative propositions, then refined through traditional evaluation cycles including skin testing, sillage measurement, and stability assessment under standard conditions. The launch was positioned by O Boticario as a deliberate communication around innovation rather than as a hidden internal experiment (Symrise corporate communications, accessed 2026-05-29).
Other launches across the major suppliers and their downstream brand clients are likely to have incorporated algorithmic ideation without being publicly identified as algorithmic in the marketing communication, particularly in fine fragrance briefs where the language of innovation around AI is not always commercially attractive to the target buyer. The industry trend is toward integrating software-assisted ideation as part of standard formulation workflow rather than treating it as a distinct category requiring disclosure. This integration makes the algorithmic-perfume designation increasingly fluid in practice.
The role of the perfumer
The perfumer retains evaluative judgement, skin-testing iteration, finishing decisions, and the final aesthetic responsibility in algorithmic compositions. Software proposes a starting point or a set of starting points; the perfumer evaluates against the brief, modifies the accord, runs the standard iteration cycles, and selects the final formula that will be signed and registered. The training data on which the software draws comes from human perfumers' previous work, so even algorithmic suggestions ultimately reflect human aesthetic choices embedded in the historical formula database and in the sensory annotations attached to it.
Perfumers who have discussed working with these platforms in industry publications describe them as accelerating ideation cycles, surfacing combinations they might not have considered through individual exploration, and freeing creative time for the evaluative and finishing phases where human judgement remains decisive. The technology compresses the time between brief and first candidate accord, which has measurable commercial value in a market with shortened development cycles. It does not replace the iterative skin-testing, the brief reinterpretation, or the aesthetic refinement that constitute the core of perfumery practice (Perfumer & Flavorist, accessed 2026-05-29).
The debate around algorithmic creation
The debate around algorithmic perfumery centers on authorship, originality, and the aesthetic identity of the perfumer in an industry whose communication has traditionally emphasized individual creative voice. Critics argue that algorithmic suggestions tend toward established patterns reflected in the training data, potentially reinforcing safe and commercially proven choices over creative risk and contributing to homogenization of the contemporary olfactive landscape. Supporters argue that the technology liberates perfumers from repetitive early-stage exploration and allows more time for the evaluative, refining, and finishing work where individual perfumer judgement adds the most value.
Niche perfumery has so far adopted algorithmic platforms cautiously, and the technology is structurally less compatible with the editorial positioning of most niche houses than with the optimization pressures of designer and functional perfumery. Most independent niche houses, including artisan operations and the established names of the segment, formulate without significant machine learning assistance, preserving the integrated creative process that defines their editorial identity and their communication. Algorithmic perfumes remain primarily an industrial and mainstream category as of 2026, with limited but growing presence in the niche segment for specific briefs (Now Smell This, accessed 2026-05-29).
Sources
- Perfumer & Flavorist, articles on Philyra, Carto and machine learning in perfumery. Accessed 2026-05-29.
- Symrise, Philyra platform communication and case studies including O Boticario Egeo On Me. Accessed 2026-05-29.
- IBM Research, Publications on Philyra and AI-assisted fragrance design. Accessed 2026-05-29.
- Now Smell This, editorial articles on algorithmic perfumery and authorship. Accessed 2026-05-29.