AI Market Intelligencefor cosmetics.
AI reads reviews and connects ingredients, texture, fragrance and perceived effects. A queryable data ecosystem for marketing and R&D.
Beyond reviews.Inside the product.
Ingredients, products and reviews each told a different part of the story. Without shared identifiers and a granular understanding of feedback, connecting formulations, perceptions and consumer needs was difficult.
- Data engineering
- Generative AI
- MCP
What we designed.
Shared identifiers and domain taxonomies connect information from different sources. Generative AI turns text into structured attributes; the database preserves relationships; MCP makes them accessible to applications and agents.
Unify the information
We linked ingredients, products and reviews through unique identifiers, separating structured, enriched data from accessible reference sources.
Go beyond sentiment
Generative AI extracts information on perceived effects, such as hydration and blemishes, alongside texture, fragrance, value for money and usage context. Reviews become comparable data.
Bring the data into AI
The MCP server exposes the ecosystem to AI clients and agents with read-only access, enabling specific questions about product design and cosmetics development.
Ingredients, products, reviews
Shared identities + AI extraction
Product analysis through MCP
From conversationsto knowledge.
Follow a review’s journey and discover how it becomes information connected to the product.
One voice, many sources.
Multilingual reviews from e-commerce sites, forums and social channels are brought into the context of products and consumers.
- Review text
- Referenced product
- Available profile and context
A body of feedback ready to be organised.
Text becomes structure.
Generative AI extracts granular attributes and maps them to shared taxonomies: concerns, perceived effects and fragrance dimensions.
- Effects and polarity
- Skin profiles
- Fragrance and use context
Perceptions that can be compared across sources and languages.
Each data point finds its relationships.
Unique identifiers connect reviews, products, ingredients and profiles. This structure makes cross-dataset analysis possible.
- Product ↔ Ingredients
- Review ↔ Effects
- Profile ↔ Experience
A coherent, queryable data ecosystem.
Knowledge becomes usable.
The MCP server exposes read-only data to AI clients and agents. Marketing and R&D can explore product questions with the necessary context.
- Queries through MCP
- Segment analysis
- Product comparisons
Insights to guide research and product design.
One data ecosystem.Many perspectives on a product.
The challenge is turning multilingual opinions into comparable entities, attributes and relationships. The project connects the consumer’s voice to product structure and ingredient knowledge.
The connected data
Multilingual reviews
E-commerce sites, forums and social channels feed collection from multiple sources. The architecture is designed for flows of hundreds of thousands of new reviews per week.
- Multiple sources
- Different languages
- Usage context
Products
Serums, cleansers and other categories, including competing brands, are organised through shared identifiers.
- Category
- Brand
- Product origin
Ingredients
INCI entities are linked to products to explore formulations and relationships with reported perceptions.
- Niacinamide
- Centella asiatica
- Hyaluronic acid
Skin profiles
Skin types are mapped to common categories to make consumer experiences comparable.
- Dry · Oily · Combination
- Sensitive · Normal
Needs and concerns
A normalised taxonomy organises the themes expressed in reviews across languages.
- Acne
- Dullness
- Visible pores
Perceived cosmetic effects
Effects are described by direction, intensity and sentiment: what a person reports after using the product.
- Increased hydration
- Reduced fine lines
Demographic and geographic data
Age brackets, gender and country or region, where available, enable analysis by segment.
- Age bracket
- Gender
- Country / region
Fragrance Intelligence
Fragrance is broken down into separate dimensions. Explore the variables captured and what they mean.
Intensity, pleasantness and duration
The strength of a scent and how that strength is experienced are different pieces of information.
- Perceived intensity
- From absent to very strong: describes how noticeable the fragrance feels.
- Intensity comfort
- Distinguishes a strength considered appropriate from one perceived as excessive or insufficient.
- Pleasantness
- Records the polarity of the judgement, from appreciation to rejection.
- Duration and duration comfort
- Separates perceived persistence from how much the consumer likes that persistence.
Scent families and contexts
The same fragrance may be appreciated in different situations for different reasons.
- Scent family
- Categories such as citrus, floral or woody allow perceptions to be grouped.
- Recommended or avoided contexts
- Daytime, evening, office, exercise and other occasions mentioned in reviews.
- Off-notes
- Descriptors such as rancid, plastic or sulphur make recurring signals traceable for further investigation.
Reported reactions and perceived audience
Consumer reports remain identifiable as such, together with their context.
- Self-reported reactions
- Headache, irritation and other reported experiences can be collected for analysis by the relevant teams.
- Perceived gender suitability
- Feminine, masculine or unisex describes who the reviewer thinks the scent suits, not the reviewer’s gender.
- Perceived audience age
- Terms such as young or mature describe the audience suggested in the review, separately from the writer’s age.
From an identifier to an R&D question
Unique identifiers connect reviews, products, ingredients and profiles. The structure distinguishes enriched domain data from reference sources. The MCP server makes this foundation accessible to AI clients and agents with read-only access, turning a perception into cross-dataset analysis.
Reviews describe perceptions and experiences: ingredient–effect associations support research but do not by themselves establish causation.
Questions the data makes explorable
- How does fragrance comfort vary across usage contexts and audience segments?
- Which effects are reported for products containing an ingredient across different skin profiles?
- Which combinations of texture, fragrance and effects recur in well-liked products?
What it makespossible.
- Reviews, products, ingredients and profiles linked in one ecosystem.
- Granular variables for perceived effects, texture, fragrance and usage context.
- Cross-dataset queries supporting marketing, research and product design.
The next formulastarts with a question.
Want to make your data a resource for product innovation?
Tell us about your project
Industrial data. Connected.
CAD, business systems and Excel. One platform to find projects, compare costs and put the factory’s memory to work.