Success Story - en

Segmentation of corporate clients

Description:  improving the quality of service for legal entities by the bank involves expanding the range of services provided for the analysis of competing and similar companies, searching for optimal and reliable counterparties and providing general industry reports with distribution across regions of the country. To solve the problem, it is proposed to build a digital profile of a corporate client based on its interaction with contractors and the activities of these contractors.

Context:  there is a lot of information about corporate clients: about transactions with clients, with counterparties, a description from client databases. It is necessary to identify the type of activity of the company based on its transactions to personalize offers.

Decision:  a multimodal hierarchical thematic model is proposed that takes into account all types of information for working with customer data. Simulation results are profiles that correlate with the type of activity.

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

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

Simulation result:
  • Model for building a digital client profile;
  • A search model for similar companies and competitors;
  • Building a sectoral map of the region.

Customer: Finance, Banking

Technology stack: TopicNet, BigARTM, nltk, gensim, Python.
Natural Language Processing Personalization Engineering Division