Russian field (Datanomics Demand Forecast (DDF))

2022: Improved predictive accuracy of auto-order

On August 2, 2022, the company Beltel Datanomics announced that it had previously been tasked with improving the prediction accuracy of the existing auto-order module retails of the Pavlovskaya Chicken branded agricultural holding. To solve it, an analytical service Datanomics Demand Forecast (DDF) based machine learning on the model is used. SaaS The service works in the context of a store - SKU for every day with a weekly forecast horizon. Service results DDF are integrated in the customer's business systems.

According to the results of the pilot project, the accuracy of the created forecast algorithm turned out to be 40% higher than the previous one, and all 110 stores of the Pavlovskaya Kurochka chain were connected to the solution. The service also allows you to automatically forecast sales during promotions.

The transfer of the forecast service to Yandex Cloud made it possible to optimize billing and reduce the cost of maintaining cloud resources.

Distributed calculations for order forecasting are carried out using the Yandex DataProc service, a managed, flexible big data storage and processing tool that allows you to change the number of hosts on a running cluster without interrupting tasks. Compute Cloud has created an additional virtual machine that allows you to control tasks in Yandex DataProc. Data, task logs and forecast update scripts are stored in S3 Object Storage, and Yandex Virtual Private Cloud is used to manage the cloud network infrastructure.

The Datanomics Demand Forecast service is not only a fully automated demand forecast for our retail network, but also expert support from the Beltel Datanomics analyst team. There are plans to further develop the functionality of the service and the transition from regulatory insurance stocks to calculated algorithms, which will increase the turnover rate of inventory and reduce the share of lost profit arising from the lack of goods on the shelf, reported Valery Naumov, Director of the Information Technology Department of the Management Company "Russian Field "

In the network "Pavlovskaya chicken" the main share of the assortment is perishable products. The difficulty of forecasting in the fresh and ultrafresh categories is a known problem for retail, which machine learning algorithms help solve. The use of cloud infrastructure significantly speeds up the launch of the forecast service and does not load the customer's own server capacity, narrated by Anna Plemyashova, director of Beltel Datanomics

Yandex Cloud offers convenient services for working with large amounts of data. Partner solutions empower the platform to implement more complex analytical scenarios, including, for example, predictive analytics of sales, production and logistics processes. As of August 2022, predictive services based on machine learning are actively used by retailers. However, we see that interest in the solution is growing from companies of all sizes and from different industries, commented Maria Anikanova, Director of the Yandex Cloud Partner Department