One of the most important challenges that any company faces within the retail sector is that associated with having the most accurate estimate possible of its future evolution, both at the level of product sales, billing, delivery… About this, Eroski , hand in hand with Ibermática, has launched an Advanced Data Analytics system to configure a predictive model that allows it to optimize the last-mile transport service. Knowing in advance the number of home delivery orders that are going to be registered, broken down into days and time slots, can ensure significant cost savings (up to 50%), since the supplier himself also saves them by optimizing his service.
Thanks to the use of Machine Learning models, it has been possible to generate a forecast of the demand for orders that each of the transport groups will have to deliver in each time slot and each day with a time horizon of 2 months, without being affected customer satisfaction. Machine Learning algorithms have been specially designed and optimized to understand and correct the influence of different external agents that can reduce the efficiency of predictions: proximity to holidays or the existence of abnormal states, as is the case in the current pandemic context, but especially during the stages of limitation in the mobility of citizens and perimeter closures.
Eroski applies Advanced Analytics to optimize its home shopping service
The project has been developed and put into production through analytical workflows orchestrated through Rocket on the Stratio platform. The tracking of the models, their persistence and comparison between experiments was carried out in MLPojects using mlflow, thus ensuring the quality in the complete life cycle of the machine learning models created. It is an important project since, as a result of the health pandemic, this type of service has multiplied, with many users opting to receive their purchases at home.
Efficiency of online orders
On the other hand, Ibermática has managed to demonstrate in an objective, automatic, descriptive and self-explanatory way, which are the most relevant patterns on different indicators that support, with a reliable statistical significance, the inefficiency in the generation of online orders, in the different distribution centers of the Eroski Group. These inefficiencies translate into very significant costs in terms of returns, refunds, replenishments, and a large catalog of direct and indirect expenses. Likewise, the multifactorial combinations that motivate the appearance of incidents in the delivery of online orders to the client have been discovered, determining if these claims are real when estimating their measurement, and what is more important, predicting and estimating those incidents. not reported, based on the measurement of the direct and theoretical impact on customer satisfaction.