M.Video-Eldorado scales its video analytics system to 50 stores

2021

50 Store Video Analytics Scaling Plan

PJSC M.video (Group) M.Video-Eldorado on August 6, 2021 announced plans to expand the project by to video analytics about 50 stores. According to the pilot's results, video flow analysis proved commercial efficiency - the customer attraction ratio in pilot stores grew a third faster due to an increase in the quality of client service and the implementation of the "lone buyer" scenario. Your own solution, developed by the company, is on average five times cheaper than market counterparts and allows you to recoup the installation from the first month.

M.Video-Eldorado scales the video analytics system developed by the Group's own data office. In June, the company set up a video analytics solution in five additional locations in Moscow, and by the end of 2021 plans to increase their number to 50 in the capital and the region - thus, the number of cameras connected to the neural network will increase to 250. The solution created by the Team's development team uses data from CCTV cameras already installed in stores, which minimizes implementation costs.

As part of the pilot, the retailer has already implemented three main scenarios for using the data received: "lone buyer," "checkout queue" and "store heat card." The first direction was to quickly help customers who have been standing at the shelf for some time or moving around the trading room in search of a consultant. The IT solution quickly identifies such customers and sends a notification to store employees via chat bots. The technology increases the attention of staff to customers and increases the quality of service. Since the beginning of the project, the number of notifications of the need to help the customer or open additional cash desks has decreased by 75%, and the conversion rate has increased by 35% compared to comparable stores.

Шаблон:Quote 'author = said the head of the video analytics department M.Video-Eldorado Maxim Shakura.

The video analytics team of the data office M.Video-Eldorado created a simple and fast algorithm for scaling the system. Now it can be deployed by store employees themselves within one or two days: it is enough to connect the computer to the network, link and configure the positioning of cameras, as well as mark out zones that require analysis.

Data from the IP store cameras are processed by a neural network based on YOLO, a solution for detecting multiple objects in an image. cloudy IT Infrastructure Solutions based on are used to process the video stream. Raspberry An intelligent solution in real time analyzes the flow of data from the store, can distinguish employees from visitors, and then "overlays" the location of people with the store plan. The neural network also analyzes the number of visitors in the area of ​ ​ distribution of goods and cash desks. If the norm is exceeded, the personnel will receive a message and take actions to resolve the situation.

Introduction of corporate neural network

On July 30, 2021, M.Video-Eldorado Group announced the introduction of a corporate neural network to automate communication with 28,000 retail employees. Artificial intelligence through chat bots in Telegram and Viber helps store sellers solve most everyday problems - from vacation processing to customer support, reducing the cost of resources and time to process user calls by four times. As of July 2021, the M.Video-Eldorado neural network processes about one million calls to knowledge bases per month.

The M.Video-Eldorado neural network already processes about 50% of requests automatically, 30% are closed through the correct answer by the operator from the options proposed by the neural network. And only about 20% require the operator to immerse, study issues, search for documentation and prepare a response. Thus, the company spends about four times less resources and time preparing answers, increasing the speed of response to the bulk of incoming questions.

The self-learning neural network on July 3, 2021 is based on a solution AutoFAQ integrated to corporate, chat bots which retail store employees use to issue online orders. As a result, the company created a tool for two-way communication and prompt answer to questions about interaction with IT systems, personnel policies and customer service procedures from more than 28,000 people, without increasing the staff of client support.

A chat bot in Telegram and Viber with a connected neural network receives a request from the user and instantly starts a dialogue session, analyzes the question and looks for an answer in the loaded knowledge base. If the probability of a correct answer is higher than 80% (the parameter is configurable), the system responds if below - transfers the request to the support operators. Questions that the neural network is not yet able to answer on its own fall into the operator's interface, while the system offers it a choice of at least three most suitable answers. The support employee selects the appropriate one, or responds independently in a different way, due to which further training of the neural network takes place. The neural network of support for M.Video-Eldorado employees already operates on about 5,000 articles, the time of further training of the entire neural network with a new article is 10-20 seconds. This allows you to make changes and develop the system as quickly as possible.

Шаблон:Quote 'author = said Dmitry Marykin, Head of Process Development and Change Support Department at M.Video-Eldorado.

Human Resources and Customer Service System Testing

On March 18, 2021, the M.Video-Eldorado Group announced the expansion of the digital technology pool in retail to increase the level of customer service and effectively manage retail. business processes The company tests in stores the management of personnel and customer service based on, data video analysts which are processed neural network in real time. The pilot project is carried out as part of three working analytical scenarios: "lone buyer," "checkout queue" and "store heat card."

The video analysis system was developed by the M.Video- data office. El Dorado Store data IPcameras is processed by a neural network based on YOLO, a solution for detecting multiple objects in an image. Video stream processing uses cloudy solutions and infrastructure based on. Raspberry An intelligent solution in real time analyzes the flow of data from the store, can distinguish employees from visitors, and then "overlays" data the location of people on the store plan.

The first working scenario for testing within the video analytics system was to help customers who have been standing or moving around the trading room alone for some time. The IT solution allows you to quickly identify such customers and sends a notification to the store's chat bot, after which a free consultant approaches the client and provides personalized assistance. The innovation increases the level of attentive staff and improves the quality of their work with visitors - staff engagement has increased fivefold.

The neural network also analyzes the number of visitors in the area of ​ ​ distribution of goods and cash desks. If the norm is exceeded, the personnel will receive a message and take actions to resolve the situation.

Another product created on the basis of the neural network was "heat cards" for analyzing the retail space and managing sales. The software solution builds a dense distribution of store visitors by zone, which allows you to study the behavioral models of customers, evaluate the convenience of placing racks with different groups of goods and select places for placing advertising materials.

M.Video-Eldorado plans to conduct pilots in several stores and in the future the solution can be expanded into more than a thousand stores of the company with confirmed economic efficiency.

{{Quote 'Retail infrastructure is the core of our business, providing customers with quick access to goods, including online orders, can personally test goods and get expert advice. Taking into account the new patterns of behavior of customers, we strive to maximize the digitalization of retail processes, be closer to customers and create a personalized "seamless" experience for them within the framework of the One Retail technology concept. For example, we are developing a mobile platform application with on the side of the seller and customer, and contactless services in stores, as well as actively using systems predictive analytics for managing runoff and delivery, - says Kirill Ivanov, Director of the Big Data Office of the M.Video-Eldorado Group, - We developed a solution in the field of video analytics in less than six months "from scratch," without investing in expensive third-party ready-made platforms, by means of its own team, assembled from graduates and students. HSE The results data allow us to better know customers, their needs and habits and in the future make analytics in the store as rich as on the site or in the mobile application. Behavioral analytics helps us bring customer interaction to a different level, meet their personal requirements, and improve the efficiency and quality of our own business processes. In the near future, we plan to increase the capabilities of this tool by adding work scenarios, for example, registration of group visits. }}