Application of machine learning in credit recovery process

• Grant number: 18/01165-7
• Support Opportunities: Research Grants - Innovative Research in Small Business - PIPE
• Start date: May 01, 2019
• End date: December 31, 2020
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Principal Investigator: André Menezes Oliveira
• Grantee: André Menezes Oliveira
• Company: Adimplere Cobranças Ltda. - ME
• CNAE: Atividades de teleatendimento ; Atividades de cobrança e informações cadastrais ; Atividades de serviços prestados principalmente às empresas não especificadas anteriormente
• Pesquisadores principais: Felipe Maia Bezerra ; Leandro Farias Nogueira ; Rodrigo Amorim Ruiz
• Associated researchers: Carlos Stein Naves de Brito ; Glauber De Bona ; Leliane Nunes de Barros ; Marcelo Finger
• Associated scholarship(s): 19/10065-9 - Application of machine learning in debt collection process, BP.TT

Abstract

Adimplere is a "fintech" that aims to develop an innovative process based on the intensive recovery of credit through digital communication channels and to collect and negotiate people in default situation autonomously by the system. This process will have as main tools, data monitoring, analysis and tracking techniques with the aid of artificial intelligence and machine learning. This innovative methodology aims to break the main paradigms of the current market, which has a highly intensive scenario in people and human resources, resulting in a need for high financial contributions and investments throughout the value chain. The company intends that this methodology be made available in multiplatforms, software / web and mobile (smartphone and tablets), which will facilitate the access of the functionalities to all involved. In addition, the entire structure will follow a security protocol for the preservation of information, using restricted access to users, with an audit of all operations performed in the system, seeking to ensure confidentiality between institutions and customers. The solution's target customers are banks, financial institutions, small and medium-sized creditors, and department stores that have credit sec- tors. Adimplere intends to develop a module that will be part of its credit recovery negotiations portal. This module will be responsible for managing "Automated Billing" and will use machine learning techniques to support decision in negotiating with each customer. Using data provided by customers and retrieved by different channels, the program must distinguish individual qualities and interact with customers by sending charges and proposed agreements. In the course work, one of the team members has already carried out a first modeling and experimentation of a batch reinforcement learning model (BRL) in order to automatically identify the optimal policy for debt negotiation campaigns. The next step, the goal of the research project, will be to develop and validate a machine learning model to guide decisions to offer discount and installments of debts along a portfolio of defaulting clients in order to maximize credit recovery in agreements financial resources. Adimplere already has a sufficiently large customer and negotiation database for the design and testing of a prototype model. (AU)