The main difference between machine learning and traditional analysis is not programming the algorithm, but training the model to solve the indicated problem on the provided data. These algorithms are called machine learning algorithms, and they are beginning to supplant the previous approaches to analytics. The requirement for the implementation of tasks using machine learning methods is the presence of a certain set of historical data for training a model with a storage depth, depending on the solution being implemented. 

Self-learning algorithms process large amounts of data, remember successful and unsuccessful decisions, and use this information in further predictions. Algorithms are trained on historical data: it can be transactions, customer interaction history, Internet sources, revenue information, etc. The set of data, the quality and duration of the period for which they are collected determine how accurate the model will be in the end.

In the data array, the algorithm finds relationships, tracks how and why the influence of various factors on the process of interest changes. The machine sees even non-obvious patterns and does it faster than a team of analysts.

Machine learning technologies are now automating many business processes and helping retailers make money. For example, a store has been collecting information about purchases for several years. The system analyzes data and finds patterns: how customer demand depends on the season, the appearance of new products, promotions and other factors. Based on this, she makes a forecast: which goods should be purchased more next month, and which no one will buy.

Machines cannot learn on their own, they need high-quality data for this. If the information on the basis of which the algorithm is trained is incorrect, the machine will not be able to make an accurate prediction . At the same time, only 3% of the total volume of information collected by companies can be called high-quality. In order for neural networks to build models correctly, you need to collect reliable data, carefully clean it from extraneous noise and prepare it for machine learning. The preparation stage is called preprocessing – the information is translated into a format suitable for training the algorithm.

Machine learning should be implemented by high-turnover retailers who operate in daily changing markets, track information about thousands of customers, track prices for thousands of items. The higher the store's revenue and turnover, the more profitable it is to use algorithms that optimize prices and predict sales.

It is easier to implement machine learning into business processes if the IT infrastructure is based in the cloud. This makes it easier to scale solutions to regional branches of the network, and simplifies strategic planning.

• Self-learning algorithms can analyze data arrays that a person cannot process, find relationships between various factors and the amount of profit.

• Machine learning in retail makes it possible to more accurately predict demand and sales, segment customers, optimize marketing and advertising, and manage assortment.

• It is worth implementing machine learning for retailers with large annual turnover or high business margins.