How to turn a client into data: we change video surveillance and video analytics for retail
“The future has already come, it’s just unevenly distributed.” So, the technologies of the future are also penetrating the retail sector, as business success is already determined by the indicators for implementing a variety of IT solutions.
In retail there are two ways: to become technologically advanced or die. In the first case, there is plenty to choose from - technologies are already available on the Russian market to track the availability of goods on the shelves, to recognize customers, and also to prevent queues.
Video analytics is becoming a key competitive advantage - it is a natural evolutionary response to tough competition and an increase in consumer demand for service levels.
Platforms that help boost sales by tens of percent are limited to only one significant problem - the cost of solutions is prohibitive for small and medium businesses.
Affordable video analytics is the free niche in which Ivideon will “distribute the future evenly.”
Computer vision: know the price in person
The world market for face recognition, according to the company MarketsandMarkets, will grow to $ 6.84 billion by 2021. The Russian market in this respect is much more modest - tens of millions of dollars. The main reason for low volumes is the high cost of the final solution for small and medium businesses.
Face recognition for several years as a routine. And when a client came to us with a request to make a tracking service for VIP clients, I was very surprised that we had no such solution. After all, the phone has face recognition, and why not in the IP camera?
The task of monitoring and detecting people in a stream until the boom of neural networks remained extremely difficult.
Let us consider this thesis on the example of one of the world's largest video surveillance systems. In Moscow, cameras are installed everywhere: on the territory and in the buildings of schools and kindergartens, at the MCC stations, stadiums, public transport stops and bus stations, in parks, in underground pedestrian crossings.
Moscow ranks second in the world in the number of surveillance cameras per capita: there are 11.63 surveillance cameras per 1,000 inhabitants, which is almost two times more than in Hong Kong and six times more than in Beijing.
Cameras over intercoms in Moscow recognize faces and send a push notification if a person from a search team gets on the lens. The Moscow metro also has a face recognition zone.
Moscow government spendsabout 5 billion rubles a year for system maintenance, with the records from each camera being kept for only five days. How much does it cost to connect a facial recognition system? According to experts , at least $ 100 for each camera connected. For small businesses, this is an exorbitant cost.
Clouds (not only white-man horses)
Why do we compare video analytics for retail with analytics of a safe city? Especially because of 160,000 metropolitan cameras, an ordinary person is available without a policeman. 0.
The fact is that until recently, the prices for connecting the facial recognition system were taken from the ceiling, confusing customers with an approximate calculation of the flow of people and the size of objects. It is difficult to compare with what has no clear boundaries.
Even the focus of the business itself affects the cost - if you have a casino, be prepared to pay more. If you have a network of 80 stores, be prepared to pay more for each camera. It is not surprising that against this background, up to 2020, the costs of developing, purchasing and introducing innovative solutions from the largest retailers in Europe can reach 17-19% of revenue.
"We would have all this money!"
Opinion of experts in the field of video surveillance
A turnkey system is collected for each client individually, guided by the logic of " how much we can earn right now ." Price lists on sites no one usually places, winding up the price of up to 100 million for large industries and a network of supermarkets.
At first glance, the solution really does not look affordable: to cover the point of sale with intelligent video surveillance, dozens of not the cheapest cameras are required, switching, switch, server, etc.
However, the cloud solution reduces all costs to a minimum. IP cameras will suit any on which you can distinguish the face. The number of cameras is reduced, because cloud analytics allows you to configure multiple zones for analysis on a single camera.
Most outlets have already installed video surveillance systems. But these systems can also be connected to the cloud without overpaying for complex and expensive re-equipment of the hall.
What the buyer needs
The camera recognizes the buyer by sex and age, and then offers something on the display that may interest him. Photos: X5 Retail Group
We make analytics available, but why do we need it at all? For retail, the answer comes down to two bases:
- identification of thieves, on the basis of those who were previously seen by the security service;
- identification for distribution on loyalty bases and stimulating consumer activity.
It is worth the visitor to go to the store, as his face captures through the camera software for face recognition. If the client has been noticed for dangerous actions, the guard will receive a warning about the threat. And if our buyer, for example, is a VIP client, then he can send a discount to the instant messenger. You can also determine that there are more young women in the store at the moment than other buyers and publish a special offer for them on the screens.
In addition, the analyst will show where the individual buyer is looking, what is holding up his eyes, what colorful spots attract his attention. These data will help assess the marketing appeal of the promotion and the work of merchandisers.
Generation Z and the younger ones are particularly sensitive to personalization, the online transition to offline, digital chips that increase the attractiveness of shopping. Ultimately, targeted offers save customers time, money, and kill boredom from shopping.
The Rive Gosh network recently officially announced that it is successfully using a trained system for predicting customer behavior based on data from cameras, cash registers and machine learning. Already at the initial stage of its work, the accuracy of personal product recommendations for specific article numbers was 33%.
The system is able to identify from all retailer loyalty card holders (and this is 2.6 million people) those who will potentially make a purchase in the next two weeks, while it makes a forecast of the most likely purchases for each of the customers. In Magnet, the machine learning technology allows you to always have in the assortment exactly the products that customers need.
For the business owner, according to X5, video analytics and computer vision technology reduces by 10% the number of people leaving the store without purchases, and by 20% - the loss of stores.
Products on the shelf
Smart people in trade marketing have come up with a special term for the situation when you come at two o'clock in the morning for your favorite Agusha and see that the damn yogurt disappeared from the store shelf.
OSA (On-Shelf Availability) is the availability of goods on the store shelf at any time. In turn, the situation when the product is not on the shelf and you want to buy it is called OOS (Out-of-Stock).
Full OOS is an unpleasant situation. According to our data, 25% of people go to the store just for what they need - no situational purchases. Such a man-flint will unfold and leave, without looking at the chocolate bar at the cash register.
There are several reasons for the emergence of OOS: an analyst in the head office has become mistaken with the forecast of orders, a logistics problem has arisen somewhere, the goods are on another shelf. 20% of all goods in Russia are constantly in the state from OSA to OOS, and retail chains lose from 2 to 4% of their turnover.
Computer vision elementary determines the presence, quantity and correctness of the location of products on the shelf, eliminates the human factor and automates the process of supplying the store.
The system recognizes almost any goods on the shelves with an accuracy of more than 95%, takes into account the shape, color, logo, type of packaging, and sends employees to the application reports with information on the range and display within 10 seconds.
Image Recognition technology defines:
- what goods are on the shelf and the number of units of products;
- whether the product has a price tag, and whether the price for it is approved in the central office;
- is there an approved promotion for this product;
- what products are missing on the shelf despite the approved range.
An additional element of protection from "I forgot" the manager - the automatic setting of tasks for the display of the goods and the specified frequency analysis of the availability of goods. Access to the system is configured at once for several responsible persons in order to completely eliminate the lazy human factor.
Testing of such a system in Pyaterochka led to an increase in sales of goods, which showed how effective the video and photo control displays.
The potential benefits from the introduction of automated control over the availability of goods on the shelf are up to 5% in turnover due to an increase in the availability of goods on the shelf.
We wrote several articles about the queues , the essence of which boils down to one sentence: customers are negative, and any negative leads to the fact that another time the customer does not come to this store.
Video analytics determine the number of customers in the queue. When the camera records that more than five people have accumulated, the manager receives a notification about the need to open an additional cash desk.
At the same time, one camera is able to control several areas of the trading hall at once, which reduces the cost of a small business solution.
Manager in real time monitors the dynamics of visits to the store throughout the workday, knows the time and day of the most active customer activity. It is useful to find out objectively, and not from the words of the sellers, when in the store the peak of visits, on weekends or weekdays, do customers prefer to shop early in the morning or late in the evening.
More advanced analytics will be able to track shoppers' movement down to the meter. In accordance with this information, you can send them push-notifications with relevant promotional offers.
Over time, even a small business will be able to find a solution in which it will be possible to figure out how often a particular customer goes to a particular store and, by comparing this data with the interests of the customer, form regular promotional offers.
What is under the tree
Video analytics Ivideon has long been like a unicorn - everyone heard about it, but no one saw it. The problem was not to make analytics and train neural networks. The problem was to make this system as possible as professional solutions, but as much as… no one, that is, for people who see the phrase “the cost of introducing innovations will reach 17-19% of revenue” from a scientific quotation fiction novel.
Raising $ 8 million in investment from the Rusnano Sistema SICAR and Skolkovo Ventures funds allowed us to actually complete this task. Now we not only give access to the online video stream and video archive, but also connect any third-party video analytics, adapting it to the scale and needs of the business.
What this means in practice: the analytics payment is calculated on the basis of traffic (face detector, goods detector on the shelf) or directly at fixed rates for each camera (queue detector). Counting visitors, queues and goods on the shelf are already available.
Face recognition, we will present the New Year. If you want to connect video analytics modules right now or sign up for a beta test of individuals (for business users only), please contact us at firstname.lastname@example.org .