Customer Recovery

The challenge of behavioral approach on credit risk management.

Customer Recovery

Background

Now more than ever, the intervention of financial institutions in the consolidation of the comprehensive credit risk management is fundamental due to the delicate balance that characterized them. The bank proactively aims to anticipate the difficult scenarios of its defaulting customers and to propose sustainable recovery solutions, contributing to the financial balance harmonization.

According to the default degree, the nonpaying customers are classified in the customer recovery perimeter, broken down by nine buckets; each one represents 30 days of delay in payments. The relationship with the debtor for credit recovery is based on a 2-step process: in the first step (between bucket 1 and 4) the call approach is adopted. When the debt is more conspicuous and the delay in payment matured a longer elapsed (from bucket 4 to 9), a human interaction is preferred, with door-to-door approach.

The bank should be able to identify in advance each customer classification and to set up the most appropriate and efficient procedure for its recovery in order to reduce costs of credit recovery operations and to implement a smart management of risks. The algorithm is implemented through the deployment of a statistic model that analyzes variables both demographic and behavioral, feeds the Machine Learning mechanisms and enable it to make predictions, building models based on replicable behaviors.

The solution

The solution is developed on Microsoft Azure Machine Learning, the service that allows building and testing powerful cloud-based predictive analytics.

ML Studio allows the control of the end-to-end process: application of data pre-processing modules to raw data, running of the experiments on the prepared data using a machine-learning algorithm and test the resulting model.

Focus On

Now more than ever, the intervention of financial institutions in the consolidation of the comprehensive credit risk management is fundamental due to the delicate balance that characterized them.

The Predictive Approach set up by Azure Machine Learning Studio aims to identify as clearly as possible self-recovered and/or still creditworthy debtors and customers probably doomed to Work Out bank process.


AS-IS processes: issue

The banks operate in a complex scenario regulated by processes requesting a considerable use of human and economic resources. The main issue is intervene in the construction of a robust and efficient network of processes, in order to consolidate, among others, two fundamental pillars:

  • The reduction of operational costs, considering that the both internal and external costs, respectively through phone calls and door-to-door, are very high;
  • Risk management: the risk increases exponentially when defaulting customers overstep the third bucket, with a relevant impact on the risk weight asset (RWA).

Our solution: Machine Learning

The Predictive Approach set up by Azure Machine Learning Studio aims to identify as clearly as possible:

  • Customers entering the first bucket – or up to the third one – destined to leave it in a short-time (self-recovered and/or still creditworthy debtors);
  • Customers over the third bucket probably doomed to Work Out bank process.

For both typologies, no effort is meant to be spent (the first one is expected to be a self-recovered situation, in the second one any effort in this phase is just wasted time), producing therefore an improvement both in time and resources used on the tasks.


Architecture overview

The solution is developed on Microsoft Azure Machine Learning, the service that allows building and testing powerful cloud-based predictive analytics.

ML Studio allows the control of the end-to-end process: application of data pre-processing modules to raw data, running of the experiments on the prepared data using a machine-learning algorithm and test the resulting model. Once an effective model is created, ML Studio allows the deployment of the model on Microsoft Azure.

Azure ML provides several different components, grouped in:

  • Data preprocessing modules;
  • Machine learning algorithm modules;
  • Azure ML API that allows applications access the loaded model once it is deployed on Azure;


DISCOVER ALL THE REPLY DATA ROBOTICS ACCELERATORS


Data Robotics Accelerator

Automated Invoice

Automated Invoice is the solution, also available as a service, which facilitates the automated management of the accounts payable process, from the posting phase to the reconciliation between invoices and purchase orders/delivery notes/receipts, highlighting the differences identified.

Data Robotics Accelerator

Brick Machine Learning

Brick Machine Learning is the solution that makes it possible to simulate various configurations used in automated production lines, in order to recommend the optimal mix of devices required to achieve the overall performance requested by the end customer.

Data Robotics Accelerator

Employee Monthly Expenses

Employee Monthly Expenses is the solution that facilitates the automated creation of expense reports starting from the underlying cost items, quickly and without the need for manual intervention.

Data Robotics Accelerator

Know Your Orders

Know your Orders is the solution that makes it possible to create a simple interface which can be consulted by users using a natural language, thus facilitating access to information while ensuring consistency and accuracy.

Data Robotics Accelerator

Match-up

Match-up is an advanced tool for the analysis, reconciliation and matching of complex data (single and/or multiple). The use of this tool finds application in data-related processes.