Design and deployment of Store Adaptor, an algorithmic assortment recommendation engine for 1,200 Carrefour stores.
Vidéo aimablement mise à disposition par Carrefour, présentant Store Adaptor, une solution de la Suite Assortiment conçue et développée par Theodo.
Context
Carrefour, a global player in mass distribution, is undertaking a profound transformation to improve the commercial and operational performance of its stores. The brand had already developed a module for selecting references in a context of reducing its food assortments by 20% and its non-food assortments by 40%.
To go further, Carrefour wanted to optimize assortment choices per store, in order to optimize space, simplify the offer and strengthen profitability.
Despite the richness of its data lake, the largest in Europe in retail, Carrefour was facing obsolete tools, set up manually and without impact monitoring.
Challenges
The first challenge lay in dealing with large data volumes and the management of scenarios of Combinatorials possible factors that led us to build two algorithms, one on the forecasting of KPIs, the second on the optimization of assortment bricks.
The second challenge of the project was the large-scale deployment of the tool, with more than 1,200 stores in very varied formats and contexts. This diversity has complicated the reliability of the recommendations, because the data collected differed greatly between points of sale — both in structure and in quality.
It was also necessary adapt the solution to the business objectives specific to each type of store, while maintaining overall consistency.
Finally, thepost-pilot performance evaluation proved to be complex: in the absence of comparative tests between stores, it was necessary to design alternative indicators to measure efficiency in real conditions.
Our approach
A mixed team of 5 Theodo experts (Data Scientists, Data Engineer, Product Owner) worked alongside the Carrefour teams to design and deploy Store Adaptor.
The keys to success:
.svg)
User-centered approach in the field
Go&See in the field with regional department managers, which allowed a strategic pivot: from a global approach to personalized optimization by department and customizable locally
.svg)
Accuracy of impact measurement
Implementation of Causal Impact to measure incremental turnover despite the absence of A/B testing, based on past trends and comparable stores
.svg)
Rapid and integrated deployment at the national level
Agile deployment and integration into existing tools, guaranteeing adoption and operational efficiency
“With Store Adaptor, our store teams are making better decisions more quickly, and that's directly reflected in the numbers.”
