Banks use telemarketing to contact potential customers for their products directly. This sales channel is complex, requiring large databases of possible prospects, and is subject to time and personnel restrictions. This article has three objectives: to compare five prediction models based on machine learning algorithms to find the one that offers the best predictive accuracy, deploy a pilot of this model, and recommend a roadmap for the future architecture that supports it. The comparison results show that the selected algorithm considerably improves the identification of customers who accept the product, which went from 11 % to 94 %, so its implementation can contribute to the competitiveness of these organizations.

Banks use telemarketing to contact potential customers for their products directly. This sales channel is complex, requiring large databases of possible prospects, and is subject to time and personnel restrictions. This article has three objectives: to compare five prediction models based on machine learning algorithms to find the one that offers the best predictive accuracy, deploy a pilot of this model, and recommend a roadmap for the future architecture that supports it. The comparison results show that the selected algorithm considerably improves the identification of customers who accept the product, which went from 11 % to 94 %, so its implementation can contribute to the competitiveness of these organizations.