Mail.ru Predictive Analytic Solutions PREDICT

PREDICT (Predictive Analytic Solutions) is a project of the Mail.ru Group, the key task of which is to create products and services for customers based on predictive models built using machine learning methods. First of all, these are projects designed to seriously increase the efficiency of marketing and sales processes, optimize the internal processes of large companies, risk management, human resources and much more.

Some PREDICT projects are:

• CRM segmentation
• segments for digital advertising
• geo-analytics
• audience profiling
• site personalization
• personalization of direct communications

2022: Creation of a service for residential real estate developers

PREDICT (project) VK has developed AI a service for residential developers. real estate VK announced this on July 13, 2022. With its help, you can automatically calculate the optimal accommodation of the residential complex and the recommended prices. This will allow you to implement project in the right time. According to to data the company's tests, the MR Group accuracy of predicting the cost of apartments averaged 91%.

When designing a residential building, the developer needs to determine the number of apartments and rooms in them, the area, as well as the price per square meter at each stage of construction. This helps to get the maximum profit and realize the entire volume of housing on time. Forming a plan manually is not always effective: the market is changing, the amount of data for analytics is growing. Machine learning technologies and analysis of data arrays minimize risks and form the planning of residential complexes as efficiently as possible. Joint experience with MR Group has already shown the high efficiency of this approach. The service can be used both to make a decision at the design stage and to test the hypotheses of experts, - said Roman Styatyugin, director of analytical services PREDICT, VK.

Service algorithms analyze the accumulated data on demand and supply for residential buildings at the point of development, prices and rates of sales of apartments in similar complexes. In addition, the models take into account information about the social and transport infrastructure of the region.

The mathematical model offers the optimal option of housing for new LCDs based on the available data with an assessment of the optimal cost of housing. The algorithm also allows you to change the distribution of apartments taking into account the design restrictions of the development and determine the economic effectiveness of this solution. We use the software package from VK on all new projects, including in such large-scale residential complexes as Mod, City Bay, Symphony 34, Paveletskaya City and Seliger City. In the future, we also plan to expand the tool's capabilities and add an option of detailed recommendations on the functional characteristics of housing in order to plan, for example, the number of parking spaces and the dimensions of individual rooms and premises in accordance with demand, "said Maria Litinetskaya, CEO of MR Group.

The system calculates how many apartments - studios, one-room, two-room and three-room - should be in the complex. The models also allow you to balance sales at all stages depending on the specified pace of project implementation. If necessary, the breakdown by apartment shares can be adjusted manually, after which the system recalculates the key figures of prices and sales dates.

The first tests of the service were carried out at residential complexes in Moscow. In the future, the geography of the project may expand.

2017

The service is able to conduct deep segmentation of the audience, predict potential interest in a particular category of goods, choose the right communication channels, predict outflow probabilities, and build audience segments of Internet users for their further use in advertising campaigns.

The developers of the company said that to create the product core - a predictive mathematical model - machine learning technologies and new information processing methods created by Mail.Ru Group specialists in the field of data science were used, as well as an algorithm for building Multiclass Look-alike models, which is a development of the PU Learning method.