如何通过改造门店来优化收入

New legacy data can now be used by large retailers to optimize performance of its stores. This realization led our client – a major retail player – to rethink the layout of its supermarkets and big-box stores using solutions from data science. The goal: to generate new sources of revenue.

Our client has 350 stores in France with floor space of 600 m² to 17,000 m² that can have up to 40 departments. This means strong disparities that did not allow uniform optimization throughout the chain, because optimal allocation of various areas’ floor space could obviously not be the same at a small urban supermarket and a huge big-box store in a rural area.To approach a store and its potential holistically, other parameters come into play, such as the store's distance from competitors, the type of business (specialized or general), the populations of the areas the store serves, and customer buying behaviour by profile.

The project was handled in two main stages:‍

• ‍A predictive approach toward revenue: We modelled the revenue using multiple data domains, such as department floor space, customer segments, cash register receipts, INSEE-type demographic data, competitive data, and more. This part of the solution is for developing a predictive function. And it is the exhaustiveness of the data associated with the stores, and its preparation, that allow a robust optimization solution.‍
• Determine the best square-metre allocation to various departments: To maximize revenue, our solution generates recommendations on store layout – either increasing or decreasing department floor space.

To ensure effective staff acceptance, the design thinking studios co-built the ability to interact with the algorithms and define the main functionalities of the final product.

The key success factor was the ability to identify and collect pertinent data. A total of 3 terabytes of data (20 billion lines) were collected and used to train the models. The second strategic lever was feature engineering, a process of preparing raw data collected in the first phase. The objective: to help the algorithm extract the right substance by clearly telling it what to look for and where to search – in this case, to understand how revenue is generated."We analysed the customer's journey, the buying behaviour and the relationship with the products and departments. We deduced the variables representing links between the departments. They proved very important and drastically improved the prediction." (Piotr Woznica, Senior Manager Data science)The third success factor: To support data acculturation, we designed an interface that's easy for the staff to use (concept management). The users themselves define the parameters for accessing predictions and iterating optimizations. Concretely, our data science solution generates recommendations and scenarios for rethinking store layout and optimally reallocating square metres according to business constraints. This parametrization allows simulation of the ideal size of store departments to maximize revenue. Based on simulation, the Ekimetrics data team estimate an average gross revenue increase of 4.8 points. Thus, each of the 350 stores can be optimized while considering several dozen dimensions and business constraints specific to reallocation needs.Moreover, thanks to the collaborative approach to designing the solution, our client is now maintaining and developing the solution on its own.