Abstract One of the most active challenges in the production life cycle is the ability to predict failure to avoid loss of production and increase deferred production. This simple question affects many areas of operation such as maintenance, supply, production, and planning. The digital transformation enables us to disrupt the conventional mode of operation and create space for such solutions. This project aims to improve operational decision making for the logistics of maintenance planning in critical wells with ESP lifting systems of an oil field in Colombia's Upper Magdalena Valley basin. This enhancement is provided by combining Machine Learning techniques for the prediction of downhole pump failures by various agents. This project's scope is predictive failures with Analytics insight. The periodicity of failures and the impact of each of these, measured in terms of deferred production, define the criticality of the wells. The project was developed by conducting an analysis of the state of the art in this type of prediction both within and outside the industry. Following this, historical data is collected, the data to be analyzed is selected and prepared (including sample data balancing), and the data is treated with Python. Following that, the objective is defined based on the criticality of the potential failures, and machine learning algorithms are used to determine the variables with the greatest predictive potential, and then a classification algorithm is run to predict with enough time to support decision making (hours before the event). The outcome of these processes is validated, first with information from the training data and then with hidden data. This paper discusses the advantages of predicting failures in ESP pumps with performance greater than 60%. This is accomplished in two steps, the first of which is determining which variables contribute the most to the prediction of background failures. This analysis produced insights for field decision-making, assisting those who monitor more than 50 variables. The second step took the selected variables and delivered to maintenance planners an algorithm capable of predicting pump failures with high accuracy in enough time to perform preventive maintenance, reducing production deferrals of high-potential wells. This project helps to increase the use of these techniques in Latin America, as well as the development of future research in the field of predictive and prescriptive analytics and the expansion of periods that lead to better tactical decision making. This solution can be extended to other fields with similar failure characteristics, allowing it to contribute to production optimization in a short period.