Abstract: 

Customer Relationship Management (CRM) comprises the set of processes and techniques used by a company in order to manage and improve interactions with current and future customers. Data mining techniques applied in the CRM environment are evolving due to new technology developments in Big Data management and analytics. More precisely, this application deals with attrition management (churn) and customer developments (cross-selling).
A model is constructed for each churn or cross-selling goal and for each client segment, resulting in approximately 60 industrialized models.
This use case requires promptly dealing with a great amount of heterogeneous data coming from different countries and with different data structures and formats.
Each model takes as an input the list of customers that are considered the potential churners. The
output of the model is the rank of each customer, that is used for ordering the customers from the one who is more likely to churn to the one who has the lower estimated churn propensity. The first
percentiles can then be selected, accordingly with the desired amount of customers that has to be
contacted. On this selection a proper churn retention campaign is applied using several channels.
A similar approach is defined for cross-selling purposes.
The models should be constructed assuring some characteristics for allowing the usage of the results
for CRM purposes. These characteristics are mainly four, i.e. (i) satisfactory predictive power; (ii) integrated with business logic; (iii) reasonably stable across the months; (iv) robust with respect to market changes.
The development has been performed in two different environments, for different purposes. In the
laboratory environment have been managed activities as the model selection, feature selection,
missing value management and feature engineering definition. In the Big Data production environment have been managed several steps as data quality checks, industrialization of the final configuration defined in laboratory environment and automatic periodical launches of the software.

Bio: 

Federica Perugini has been working in UniCredit Services S.C.p.A. since late 2015 as a Data Scientist. Her activities have been mainly focused on Business-To-Customer applications and on Customer Relationship Management within the Big Data framework. She achieved a M.Sc. in Applied Mathematics at La Sapienza University in Rome. She is a teaching assistant at M.Sc. in Data Science at Università Cattolica del Sacro Cuore (Milan).