People are still needed
Posted: Thu Jan 23, 2025 8:18 am
Sergey Shcherbakov, Head of the Data Analytics and ML Group at ICL Services, spoke about the results of implementing a system for video analysis of customer behavior during the expert discussion "How to Get the Most Out of Big Data", which was held as part of the Innopolis AI Conference for Business "AI IN 2023".
"Compared to the usual iceland whatsapp resource security that is installed at the self-service checkouts in every store, the system works approximately three to four times better. It is clear that everything depends on the store, on the vigilance of the security, but on average, three to four times better," said Sergei Shcherbakov.
According to him, from a technical point of view, the system combines two data streams - images from surveillance cameras, which are passed through the recognition server, and payment data from self-service checkouts. As a result, it was possible to prevent several thousand thefts and achieve payment for goods. This made it possible to justify the need to develop such a system for the customer.
"This is the fundamental problem of this case, since no retailer knows the exact losses in the self-checkout area. Because there are losses in the entire store and there are analytically derived data on how these losses changed when the checkouts appeared. That is, there is no starting point from which to calculate the effectiveness of the system. It is impossible to say how much the video analytics system will reduce losses, because it is unknown what they are now. Therefore, we reduce 30% of the unknown value. This is a problem. And here we can only measure the fact - that today such and such a number of people were detained, goods were scanned for a conditional 10 thousand rubles," the developer said.
Sergey Shcherbakov noted that the presence of the system does not lead to a reduction in work for the staff, on the contrary - the workload of store security and assistants at the checkout counters has increased by an average of 10-15%.
"In order to turn the detected anomaly into the detention of the thief, the security must be activated, the assistant at the self-service checkout must be notified. For example, many people print goods on two or three receipts, and this is normal. Or they bring some goods with them. It cannot be said that since they brought water with them, they must be immediately grabbed, not allowed in and the police must be called. Of course, this cannot be done. And this logic, how to interact with people so as not to negatively affect their customer experience, despite the fact that they may or may not be thieves, - the development of this logic took two or three months. Customer paths have been built, where and at what point the system reacts, what is displayed at the checkout, what recommendations are given, how security is notified," the developer said.
He added that both the store staff and customers were understanding of the experience. "It was very interesting to watch when the system was put into operation, how people scanned goods and the system told them that they still had three items that had not been scanned. And they either moved these items away and said that they were not mine, or put them in a backpack and said that they had already come to the store with them," Sergei Shcherbakov gave an example.
"Compared to the usual iceland whatsapp resource security that is installed at the self-service checkouts in every store, the system works approximately three to four times better. It is clear that everything depends on the store, on the vigilance of the security, but on average, three to four times better," said Sergei Shcherbakov.
According to him, from a technical point of view, the system combines two data streams - images from surveillance cameras, which are passed through the recognition server, and payment data from self-service checkouts. As a result, it was possible to prevent several thousand thefts and achieve payment for goods. This made it possible to justify the need to develop such a system for the customer.
"This is the fundamental problem of this case, since no retailer knows the exact losses in the self-checkout area. Because there are losses in the entire store and there are analytically derived data on how these losses changed when the checkouts appeared. That is, there is no starting point from which to calculate the effectiveness of the system. It is impossible to say how much the video analytics system will reduce losses, because it is unknown what they are now. Therefore, we reduce 30% of the unknown value. This is a problem. And here we can only measure the fact - that today such and such a number of people were detained, goods were scanned for a conditional 10 thousand rubles," the developer said.
Sergey Shcherbakov noted that the presence of the system does not lead to a reduction in work for the staff, on the contrary - the workload of store security and assistants at the checkout counters has increased by an average of 10-15%.
"In order to turn the detected anomaly into the detention of the thief, the security must be activated, the assistant at the self-service checkout must be notified. For example, many people print goods on two or three receipts, and this is normal. Or they bring some goods with them. It cannot be said that since they brought water with them, they must be immediately grabbed, not allowed in and the police must be called. Of course, this cannot be done. And this logic, how to interact with people so as not to negatively affect their customer experience, despite the fact that they may or may not be thieves, - the development of this logic took two or three months. Customer paths have been built, where and at what point the system reacts, what is displayed at the checkout, what recommendations are given, how security is notified," the developer said.
He added that both the store staff and customers were understanding of the experience. "It was very interesting to watch when the system was put into operation, how people scanned goods and the system told them that they still had three items that had not been scanned. And they either moved these items away and said that they were not mine, or put them in a backpack and said that they had already come to the store with them," Sergei Shcherbakov gave an example.