Magnit retail chain has launched a pilot project using neural networks capable of controlling the layout of products on the shelves according to a declared plan-scheme (planogram).

"Previously, employees were comparing the schematics to the produced layouts on their own, which was time-consuming and insufficiently accurate. The self-training system makes it possible to identify the conformity of layouts even in complex spaces, eliminate the human factor, significantly increase the level of control, reduce inspection time and avoid 'virtual' runoffs," the retailer's press service said.

"The pilot takes place in 20 convenience stores. The principle of the technology is that the administrators of outlets in a mobile application in a smartphone take a picture of the shelves and in a few seconds receive a report on the correctness of the planogram. It is noted that the accuracy of recognition by neural networks is up to 98%. In the future, such functionality will be available in data collection terminals.

"The system analyzes the availability of the necessary goods and their balance in the store's warehouse, the sequence of layouts, the location of items on the first line and other indicators. If they all correspond to the layout, the task is removed, if not, it will give hints and the task will go back to work. In this case, the errors are indicated in a clear graphical scheme, "- explained in" Magnit.

According to Ruslan Ismailov, Deputy General Director, Managing Director of "Magnit" retail chain, such projects help to increase the availability of goods on the shelf. According to preliminary estimates of the retail chain, the system can improve the availability of goods to customers by up to 5%, depending on the category of products and increase sales. If successfully piloted, the technology could be replicated in more than 20,000 company stores.

"We look at the technology of photo-recognition as one of the best tools to relieve staff and at the same time not reduce, but significantly improve the quality of layouts. Testing will last for three months, during which time we will evaluate its effectiveness and impact on business indicators. Also at one of the stages we plan to use recognition to control correctness of price tags, "- he said.

Made in Russia // Made in Russia

Author: Karina Kamalova