Machine learning and artificial intelligence are being used in the beauty industry to transform the big-beauty ‘one-product-fits-all’ strategy. Instead, the industry is increasingly recognizing that each individual / customer has different requirements and preferences – leading to personalization in their products and recommendations.
However, in the beauty space, consumers prefer mass personalization over mass customisation or bespoke (individually customised) products. Mass personalization refers to products that are mass produced but can be channelized by the business to satisfy consumer preferences. An example of this are the personalized product recommendations offered by Amazon based on their purchase history. Mass customization also involves mass produced products but do offer the consumer some limited set of options to customize the product they want. For example, Dell allows customers to customize the computer they wish to order (“The Deloitte Consumer Review Made-to-order : The rise of mass personalisation,” 2019).
Consumers, rather than being actively involved in creating customized products, instead prefer passive involvement. This combined with mass personalization presents an opportunity for brand-led personalized recommendation systems (“The Deloitte Consumer Review Made-to-order : The rise of mass personalisation,” 2019).
Recommendation engines allow big-beauty brands the opportunity to identify consumer needs and market specific products to them – be it through personalized emails or dynamic websites. There are various approaches to deriving a recommendation. Content-based filters and collaborative filters are some of the more common types of recommendation systems.
Content-based filters are product-based. A content-based filter will recommend products that are similar to the products that a particular user has purchased or rated highly. The parameters for similarity can be defined – such as aroma, ingredients, etc. However, a limitation of this type of filter is the reduced diversity in the recommendations. For example, if a consumer, Natasha buys a sunscreen with particular ingredients, the recommender system may calculate similarity on the basis of the ingredients, and makes recommendations that are likely to consist of sunscreens with similar ingredients. However, Natasha will not require a sunscreen for a time period, as she has already purchased one recently. Although there are ways to overcome such limitations – by carefully choosing the parameters for similarity.
Collaborative filters are user-based. Consider another consumer Sara – a collaborative filtering algorithm will recommend products that Sara hasn’t purchased but users similar to Sara have purchased, that is, the system recommends beauty products that users similar to the consumer have liked / purchased.
In the following simplified example of this filter, suppose Sara and Natasha both purchased a particular moisturizer, and Natasha also purchased a foundation that Sara hasn’t purchased. Sara has bought for herself a cleanser that Natasha hasn’t bought. The algorithm might deduce a high similarity between Sara and Natasha on the basis of the moisturizer both purchased, and thus recommend the cleanser to Natasha, and the foundation to Sara. This algorithm allows for a larger diversity in recommendations, in comparison to content-based filters, and will cross product categories effortlessly.
Just like in the case of content-based filters, one can decide the parameters for similarity in a collaborative filter recommendation system. Explicit ratings such as user rating data for beauty products may not be very effective in a collaborative filtering system, as available data is often sparse with high deviation across thousands of different products. On the other hand, implicit ratings such as user purchase data and viewed products will lend itself very well to such algorithms. Parameters for similarity can also stem from skin type, skin complexion, age, etc. This is especially beneficial to avoida cold start, such as when a new product may not have sufficient user purchase data – they can simply design a prompt to find out the consumer’s skin attributes.
Another type of filter that can be utilized in the beauty industry are complementary filters. These filters enable the algorithm to learn the probability of two or more products being purchased together. Many beauty products go hand-in-hand, such as lipstick-lipliner, shampoo-conditioner, cleanser-toner, etc. Using complementary filters can yield good results in the beauty industry.
Hybrid filters allow us to combine different kinds of approaches to a recommendation system. A hybrid may be implemented by making two kinds of systems, separately, such as a collaborative filter and a complementary filter separately, and then combining the results of each. Hybrid filters can also be created by adding specific capabilities from one specific approach such as content-based filters to another approach like collaborative filters. Netflix’s recommendation system is a good example of hybrid filters – it uses a combination of collaborative filtering, by comparing the viewing habits of similar users, and content-based filtering by recommending movies that a user has rated highly, an explicit rating. This solves the data availability issue associated with explicit ratings.
This covers some relevant types of recommendation engines, that can be very effective in the beauty space. Developing a high-quality recommendation can lead to a significant boost in sales as evidenced in other industries.
Grimaldi, E. (2018). How to build a content-based movie recommender system with Natural Language Processing. Towards Data Science. Retrieved from https://towardsdatascience.com/how-to-build-from-scratch-a-content-based-movie-recommender-with-natural-language-processing-25ad400eb243
The Deloitte Consumer Review Made-to-order : The rise of mass personalisation. (2019). Deloitte.