Online food retail is a hotly contested growth market. Switzerland is still at the beginning: just 21% of the Swiss population bought groceries online in 2019. In this dynamic environment, the online supermarket LeShop.ch uses product recommendations to make tailor-made offers to customers and thereby increase their appetite for consumption and loyalty. In a survey, however, 38% of respondents stated that items that were suggested to them in online shops tended not to interest them. As a result, they would not find their way into their shopping cart.
In order to improve, speed up and make more relevant individual purchase recommendations, LeShop decided in early 2020 to use machine learning technology, which has been relatively new territory for the online retailer to date. To accelerate the development and the use of the algorithms, the online shop operator decided to set up a DevOps architecture. The close interlinking of development and operations would ensure that machine learning algorithms can be implemented in an industrialised manner, i.e. with a minimum of manual work.
The data scientists at LeShop had recognized that different categories of goods such as alcohol, luxury foods or staple foods require different algorithms. The core task of the project was the introduction of a standardized development and deployment process of machine learning models, which allows the data scientists to focus on creating new algorithms without worrying about their deployment.
In addition, the performance of the website should not be hampered by the new technology. One of the specifications for the project was therefore the seamless integration with the website using the REST interface. LeShop chose Trivadis as the implementation partner in March 2020.
After several workshops, the team of Trivadis and LeShop employees together began developing a DevOps architecture to create the machine learning algorithms. The project was completely implemented in Python on Microsoft Azure Services and thereby used the cloud provider’s extensive functionalities.
We have implemented a multidisciplinary and technically groundbreaking project with Trivadis’ help. There are currently not many machine learning applications that are so production-related and have such a direct benefit. We are therefore able to create tailor-made offers quickly and flexibly and thereby offer a more personalized customer-experience instead.
Maria Gazaki, Data Solutions Team Leader at LeShop.ch
LeShop is able to develop, train and use algorithms in an “industrialised” way through machine learning models. The production-ready machine learning solution developed in cooperation with Trivadis significantly lessens the time until new algorithms are deployed. For LeShop data scientists, the machine-based process means that they can use their resources on the logic instead of working on the infrastructure. The solution is sustainable because future project teams can fall back on the common project structure and best practices. Azure API management also increased security. In addition, the susceptibility to errors has been reduced because automation is now being handled by the computer instead of humans.
LeShop and Trivadis have succeeded in generating quick and concrete business benefits from a project that is state of the art. Within a few weeks, the two teams were able to create a DevOps architecture in Azure that LeShop can use to develop and implement “industrialised” machine learning algorithms within days instead of days.
Founded in 1998 with a range of only 1,500 dry goods, LeShop is now the leading Swiss online supermarket with more than 12’500 products. In the rapidly growing range, customers will find Migros and branded products, fresh fruit and vegetables, frozen foods, fresh meat and fish, alcoholic beverages, home wares and hobby articles. In addition to more than 600 organic products, LeShop also has special categories with lactose and gluten-free foods and a large selection for vegetarians and vegans.