Beluga Group applies neural network to automation of trade audit

2020: Implementation of ST SmartMerch for automation of trade audit

On August 13, 2020 it became known that Beluga Group is producer of vodka and alcoholic beverage products, importers of strong alcohol — selected ST SmartMerch service for automation of trade audit. On a withdrawal of the company, the neural network defines data from photos with an accuracy of 96%.

Task

Beluga Group set the task to automate process of recognition of names (a brand, the brand and liter capacity of SKU), to define the shelf on which the goods are placed, to recognize price labels.

Audit of points of sale is booked by merchandisers and sales representatives of Beluga Group: they photograph store regiments and fill out merchandayzingovy reports in the ST Mobile Trade application (the supplier – System Technologies). Collected data come to the central office of the company where for their processing the staff of operators is connected. The Beluga Group command analyzed technology innovations and assumed that application of neural networks will allow to optimize and accelerate processing of data.

Pilots

The first pilot on digitization of categories "Vodka" and "Wine" of Beluga Group carried out in 2018. The best result was shown by the SmartMerch system, but the accuracy of data was only 70%.

"We began with low indicators. It was the first project in the industry, and there were no guarantees that the technology can be adapted for our segment. Experts did not know how to use neuronets in the market of strong alcohol, and solution providers made mistakes. The main difficulty — to teach to distinguish a neuronet the volume of SKU. Bottles of vodka of different liter capacity, for example, 0.5 and 1 liter, on a photo often look equally. If they stand nearby, then a system can compare the sizes and draw conclusions. And if is not present? During "pilot" we checked several approaches to quality improvement of data, one of which: calculate the area of a bottle and compare it to surrounding objects" — the project manager of Beluga Group Pyotr Kuzin told.

The second iteration took place in 2019 on limited selection: in the course of recognition not all range, and several positions participated. SmartMerch showed the best result again: the quality approached 90%. But there was a question — whether it will be possible to save such high level on the complete range. The additional pilot at the choice of new devices for "field" employees carried out to the same time of Beluga Group.

And in the third pilot on Image Recognition the company entered with new smartphones and tablets. Testing was held on all by SKU in three retail chain stores. Medium accuracy of determination of SKU the SmartMerch system reached 96%, quality of recognition of price labels — 100%.

"We more than accepted the pilot's results, and, 2 years later after the first testing, Beluga Group announces a project startup", – Pyotr Kuzin reported.

Project Results

In July-August, 2020 the module SmartMerch (it is integrated into the ST of Chicago SFA system) will earn from devices of merchandisers in Moscow and St. Petersburg. It is going to start the project in the territory of all Russia in the fall. Transition to digital monitoring will lead to reconsideration of a share the regiments Beluga Group in retail chain stores.

At manual audit information on a situation in outlet, including about autofstoka, arrives with a delay to 5 days; there is no guarantee that operators investigate all photos; when calculating the preset values for certain categories of outlets are used.

After implementation of a tsifromerch the created report comes already by 8 in the morning the next day; a system with guarantee processes all loaded pictures; the share of the shelf is calculated proceeding from the current situation in specific outlet — the company obtains data on a situation "here and now".

"We implemented the automation system of mobile trade for a long time, and the "field" personnel got used to work with customized mobile application. Development of System Technologies company gives us, among other, functionality of control: we precisely know about visit of outlet, time and duration of a visit, structure of the performed work. In due time it allowed us to raise labor discipline, to fulfill visit steps, to exclude a possibility of a juggling of coordinates and photos. Therefore from the point of view of the merchandiser workflow will not change. The employee needs to take the picture in mobile application of the ST of Chicago complex and to send pictures and reports to office part of a system", - Pyotr Kuzin explained.

Standards of the photoreporting will change — as as under what corner to photograph. At manual audit the operator distinguished even badly exposed SKU. And SmartMerch can not define a bottle which is turned by a side that will lead to fixing of a "false" autofstok. Means, employees should do merchandising more carefully to receive the correct picture of a situation on the shelf".^[1]

Notes