Research and development of algorithms for prediction of customer churn probability for recurrent revenue companies

Abstract

The proposed project aims to develop machine learning based computational tools to perform customer churn prediction and classification of customers with the highest risk of dropout for companies listed in the a55 database. These companies may be A55 borrowers, or potential new customers under review for lending. A55's customer base is comprised of medium and small businesses that have a recurring revenue-based business model, such as companies that offer subscription software (Saas) services, or companies that provide subscription services. Development of these tools will use financial data from companies from the a55 database (assignors) as well as revenue data from the respective customers of each company (drawees). Database will be complemented with data acquired from external bases that correlates consumer market, geolocation, among other information. Data that will compose the base will be normalized and structured so as to provide interconnected information in the form of time series describing the information that is relevant for the understanding of clients churn rates. The customer prediction and rating system will be integrated into a platform in cloud developed by the a55 (a55 App) where algorithms will consume data from the database in real time or periodically. Information generated by this project´s technological development will be continuously used by the a55 credit analysis team, assisting them in performing default risk analysis processes and credit pricing for new loan operations, enabling the determination of more assertive interest rates to each a55 client company.The information generated by the a55 app platform will also be made available to a55 client companies, allowing them to assess churn rate and identify customers with highest cancellation risk. This will be used by companies to develop strategies for customer base expansion or retention of customers most likely to dropout. The result of this technological development will also be offered via the software platform to Brazilian financial institutions that perform credit analysis (i.e FIDCS), promoting an expansion of credit lines for small and medium enterprises in Brazil. (AU)