Magnit Tests On-Shelf Product Recognition

Кrasnodar, 27 September 2021: The Company launched the pilot project in 20 convenience stores. The system uses neural networks to monitor whether the products on the shelves are laid out in accordance with the predetermined arrangement (called a “planogram”). It currently boasts a monitoring accuracy of up to 98%.

Based on Magnit’s preliminary estimates, the system can help improve availability of products for the customers by up to 5% (depending on the category) and increase sales. If the pilot is successful, the technology may then be rolled out to more than 20 thousand stores operated by the Company.

Store managers can use the mobile app in their smartphones to take pictures of the store shelves and receive a report on planogram fulfillment within just a couple of seconds. In the future, the Company plans to add this functionality to personal data terminals. The system checks for presence of required products on the shelf and in store stocks, on-shelf arrangement, first-row placement, and many other parameters. If everything is in line with the requirements, the task is marked as completed; if not, the system indicates that the task must be redone and displays the corresponding comments. All errors identified during the analysis are shown on a simple and easy-to-understand graphical diagram.

Previously, store staff had to manually compare planograms to the actual product layout, which required a lot of time and was not sufficiently accurate. The new self-learning system allows the Company to identify whether the product layout conforms to the standards even in complex spaces, remove the human factor from the equation, significantly increase the level of control, cut down on time required for verification, and avoid creating “virtual” stocks.

Ruslan Ismailov, Magnit’s Deputy CEO and Retail Chain Director:

“Today, we have a suite of projects aimed at increasing on-shelf availability of products, and we apply specific solutions depending on the store format, traffic, seasonality, and other parameters. We see the image recognition technology as one of the optimal tools to decrease the workload for our staff while achieving a significant increase in quality of merchandising. The test will run for three months — during this time, we will assess the system’s performance and its impact on our business metrics. At some stage during the project, we also plan to utilize image recognition to monitor correctness of our price tags.”