Models for predicting the monetary value and number of transactions in a Neobank

In the financial and banking services market, financial technology (FinTech) is playing an increasingly important role. Its development is promising and innovative. However, it is still full of challenges, such as the lack of knowledge about customer behaviour, the profile of the potential customer and the dynamics of product changes, among others. A topic of great interest for this industry is marketing campaigns to make profitable, acquire and retain customers. This work proposes regression models that allow predicting the monetary value and the number of transactions for 60 consecutive days for a Neobank, which is a type of Fintech institutions that offer banking intermediation services 100% online. Transactions from 2020/01 to 2022/08 are the main data used. Each regression model includes a novel proposal for the segmentation and characterisation of the customer in terms of his behaviour in the use of six products or services and eight different transactions. It is implemented with an XGBoost algorithm, given its excellent results in recent publications and the discrete and continuous nature of the data. These models are tested with an error of less than 2% and evaluated in predicting the 60 days following the training data, achieving predictions with an error of less than 9% in predicting the monetary value or number of transactions related to calculating product revenue. This proposal would allow marketing and sales to propose strategies that impact the volume of customers with certain characteristics, and then use the model and that data to predict the volume and amount of money in products, thus determining the most appropriate strategy to follow.