Success Story - en

Segmentation of corporate clients

Motivation for launching a project by the customer:
for better recommendations in building business processes for the bank’s corporate clients, it is necessary to accurately determine the type of business and type of activity of the clients. There were limitations in the current process: the formal business descriptions could not accurately determine the type of activity, and there were many errors.

Description of the initial situation:
  • Increasing the level of quality of service provided to legal entities by the bank involves expanding the range of services provided;
  • Clients need: analysis of competing and similar companies, search for optimal and reliable counterparties and provision of general industry reporting with distribution by region of the country;
  • To provide such services, it is necessary to accurately determine the type of activity of the company using the data about the company available within the bank;
  • Current decisions determine the type of company activity with low accuracy;
  • For this reason, there were significant risks when using current models in product functionality.

Project goals:
  • creating a model for determining the type of company activity based on transaction activity within the bank with high reliability;
  • creation of a digital company profile based on information about the company’s activity within the bank.

MIL Team solution:
the use of existing team solutions in the field of customer analytics and transactional data analysis made it possible to implement a system for determining the type of company activity based on connections between various counterparties.

A pilot project was carried out and integrated into the customer’s business process:
  • The consistency of the company’s digital profile with its OKVED code is 70%;
  • The digital profile was built for 80% of MSME companies (micro and small businesses);
  • Prediction of the company's main products for 75% of companies;
  • A map of interaction between companies in the region was built with interpretation of 60% of the processes;
  • Solved the problem of finding similar companies and “Look-a-like” competitors for 70% of companies.

To build the model we used:
  • Transactions of corporate bank clients;
  • Information about texts in payment cards;
  • Data from the Unified State Register of Legal Entities;
  • History of interaction with a bank client.

Simulation results:
  • Model for building a digital client profile;
  • Model for searching for similar companies and competitors;
  • Construction of an industry map of the region.

Customer: Finance, Banking
Technology stack: TopicNet, BigARTM, nltk, gensim, Python.
NLP Research Engineering
Made on