Success Story - en
NLP Engineering Division

Retail customers profiling

Motivation for launching the project by the client:  the need to increase the involvement of bank customers in the use of internal products revealed limitations in the current solution: the segmentation and profiling models were of poor quality, the allocated user segments were not interpretable, analysts could not use the simulation results. 

What we had initially:
  • segmentation of the bank's customer base assumes a high conversion of internal products offers to customers; 
  • segmentation requires a deep understanding and analysis of the bank's customers' behaviour, for example, data on user transactions. # nbsp;
  • an increase in the quality of segmentation leads to an increase in the response to the offered products of the bank and affects the business indicators: from customer loyalty to profit. # nbsp;
  • the used segmentation models were not interpretable, their quality was low;  
  • for this reason, there were significant risks when using the current models in the functionality of the products. # nbsp

Project goals: the creation of a model of profiling and segmentation of the client base according to their transactional activity with a high level of quality and interpretation.


MIL Team's solution: the use of the existing solutions of the team in the field of customer analytics and analysis of transactional data made it possible to implement the procedure for training temporal models of the customer base segmentation.


Results:
The pilot project was conducted and integrated into the customer's business process:
  • Identified 80% topics of bank card users;
  • 85% of consumption profiles are interpreted;
  • Predicting socio-demographic parameters of the audience;
  • A two-level hierarchy of customer consumption profiles is constructed;
  • Solved the problem of searching for similar consumers "Look-a-like".

To build the model, we used:
  • Bank customer transactions;
  • Description of MSS codes and merchants;
  • Accompanying information about the user;
  • Results of sales of banking products.

Simulation result:
  • Model the bank users behaviour; 
  • Probability prediction model before purchasing a banking product;
  • Segments of the client base.

Customer: Finance, Banking

Technology stack: TopicNet, BigARTM, gensim, Python