The forecast for number of sales in all potential locations for object placement is based on the data collected from already operating retail chain outlets, more than 50 factors influencing the city space and algorithms of machine learning (more than 100 000 locations throughout the city).

Once a day our system gets updated data on total sales from all operating points of sales. Due to the application of automated machine learning, the model undergoes additional training every day with consideration of the updated data, becoming more accurate and remaining relevant over time.

Beside existing location points in the system, you can import your own into it - the system will automatically collect the necessary data on the loaded location points, implement the model and give a forecast for the number of orders.

In forecasting, we use complex models that allow, on the one hand, to identify non-linear patterns between the value of the number of sales and many factors affecting it, and on the other, to interpret this pattern in a way that is clear to the user.

So, the forecast at each location can be decomposed into contribution level of all factors involved in the model and understand better which features of the surrounding space turn out to be the most significant - which increase the number of orders, and which, on the contrary, decrease it.

The system interface gives an opportunity for the user to select any location from the entire set to simulate the placement of new point of sales in the chain. In the process of such an online simulation, the model will automatically reassemble all data and recalculate forecasts taking into account that new automatic post offices will appear in the selected ones.

If the automatic post office is successfully installed, this location will begin to accumulate data on the number of orders and after a certain time will be included in the training set.

The system is equipped with many filters that simplify the task of finding locations for placing new automatic post offices. With the help of filters, you can quickly select locations with a certain forecast number of sales located in certain areas of the city and in objects of a certain category (supermarket, residential building entrance, library, etc.).

Using advanced filters, you can select locations depending on their position relative to infrastructure objects, for example, locations of other automatic post offices/pick up points or having certain statistics on the neighborhood, for example, on the number of apartments.

All this makes the developed system an innovative and convenient tool for managing and developing the Moscow Automatic Post Office chain. In general, such an application can be used for any other chains of any commercial or socio-economic facilities with a certain number of already operating objects.