The idea is to use fragrance attributes and consumer behaviors a well as advanced data science to correlate fragrances and individual consumers.

The use of ML techniques allowed to match a user with a fragrance based on certain attributes, with up to 91.4% accuracy in fragrance prediction based on declarative ratings..

1) Long-term outcome : better understanding of fragrances x users correlations so as to drive fragrance category e-commerce growth with hyper-targeted approaches.

2) Industry relevance : COTY demonstrates its category captaincy by providing to a major retailer an innovative AI tool to levarage the level of service to the visitor, as well as improving sales conversion, basket size and frequency.

3) Originality : COTY, as Leader in the fragrances Industry, brings the category into new data driven territories, differentiating from traditional aspirational only advertising campaigns and adding a layer of hyper-personalization.

Over 5,000 respondents has been surveyed in UK which has led to 161k user inputs (2 months)

600 fragrances have been included, COTY and not COTY (1 month)

10 model based Machine Learning (AI) techniques have been assessed (7 weeks)

30,000 different combinations of questions have been tested to get to the final optimum set (1 month)

Six points rating criteria ranging from ‘strongly dislike’ to ‘love’ have been evaluated, to estimate prediction accuracy for a hold out sample within the test set (1 month)

When over >90% accuracy in fragrance prediction has been reached, the tool has been deployed on Boots.com.

The engine works as a multiple choices quizz imported as an iFrame on Boots.com

The user can choose 2 major paths : buy for himself/herself or for a gift.

At the end of the questionnaire the user is recommended a set of fragrances with a % match indicated for each of them. He/she can filter by price.

The algorithm is able to work out how successful each recommendation is at converting sales and adjusts future recommendations accordingly. This means that with each consumer that uses the app, the technology becomes ever more successful at driving engagement.

Results :

The Click Through Rate on product pages is 21% with historical peak of 30% in Valentine's Day versus a usual average of 6% Click-Through-Rate.

The quizz completion rate is as high as 94% showing high levels of engagement and interest.

Results : 131,491 people have taken the quizz since launch.

The Click Through Rate on product pages is 21% with historical peak of 30% in Valentine's Day versus a usual average of 6% Click-Through-Rate. This confirms the accuracy of the fragrance prediction and its impact on purchase intent.

The quizz completion rate is as high as 94% showing high levels of engagement and interest.