Personalising Product Recommendations for Mulberry



Tech stack

Google Cloud


Recommender System


AI + Machine Learning

Iconic British luxury brand, Mulberry, has been providing beautiful leather goods for 50 years and its network of stores and e-commerce sites serves over three million customers. Datatonic worked with the client to develop a personalisation solution leveraging AI + ML to score customers and provide predictive analytics. The resulting insights allowed Mulberry to provide relevant, personalised product recommendations to customers.

Our impact

  • Increased click-through rates in personalised email campaigns by +37%
  • Boosted ROAS on Facebook Ads by +110%
  • New Machine Learning model increased the number of customers scored by 112%


The challenge


The luxury fashion industry is traditionally known for high product quality, exclusivity, and a high standard of customer service in-store. With the long-term trend towards online shopping, and more recently store closures due to the Covid-19 pandemic, it was key for Mulberry to rethink how to deliver the ultimate brand experience to customers. To make every customer engagement count, it was crucial to implement personalisation effectively.  

From tailored products and services to highly personalised recommendations and events, focusing on customers at an individual level builds relationships, and drives engagement and loyalty. Empowering Mulberry to do that at scale was our challenge.

Machine Learning

The low frequency of purchase for Mulberry customers presented a challenge for predictive propensity to buy scoring. Our team conducted an R+D project to investigate potential improvements and increase the percentage of customers who could be scored / provided with a personalised product recommendation.

Our goal was to increase coverage and provide propensity scores for a larger proportion of Mulberry customers, without negatively impacting model performance.


Our solution

Datatonic developed a personalisation solution that empowered Mulberry to leverage their data and create predictive analytics with ease. Mulberry are now able to generate propensity scores for customers against a number of product attributes.

Over the course of Mulberry’s campaigns, attributes such as bag family, type and colour have been used to serve the most relevant product recommendation to customers, and personalised content blocks within emails and advertisements (on Facebook) delivered the most relevant recommendation for the customer. 

  1. Email campaign for the Iris bag family: Mulberry ran a personalised email campaign using propensity scores to buy each colour, achieving +37% click-through rates vs. the control group.
  2. Facebook Ads retargeting campaign: Mulberry’s personalised adverts using propensity scores on family and type developed by Datatonic led to +110% ROAS vs. the control group.

By cross-referencing scores against their stock database, additional business rules ensured that recommendations were not served for out-of-stock products.

Using a number of techniques and approaches, including downsampling and model development, we improved the metrics considerably and increased the number of customers scored by 112%.