Sistema de priorización comercial inteligente: integración de analítica predictiva para la gestión de ventas cruzadas en banca de personas (Honduras, 2023–2025)
No Thumbnail Available
Date
2026-02-27
Journal Title
Journal ISSN
Volume Title
Publisher
Universidad Tecnológica Centroamericana UNITEC
Abstract
La baja tasa de aceptación de productos financieros en los canales físicos representa un desafío operativo relevante para la banca hondureña, particularmente en contextos donde los esfuerzos comerciales no se encuentran priorizados de manera analítica. Ante esta problemática, la presente investigación tiene como objetivo desarrollar un modelo de Machine Learning que permita estimar la probabilidad de aceptación de productos financieros en los canales físicos de una institución bancaria, utilizando variables anticipables disponibles antes del contacto con el cliente. El estudio adopta un enfoque cuantitativo, aplicado y explicativo, bajo un diseño no experimental y correlacional-predictivo. Se emplearon datos históricos anonimizados correspondientes al período 2023–2025, los cuales fueron sometidos a procesos de limpieza, análisis exploratorio, pruebas de hipótesis y modelado supervisado, siguiendo el marco metodológico CRISP-DM. Se entrenaron y compararon cuatro modelos de clasificación: regresión logística, árbol de decisión, Random Forest y Gradient Boosting, considerando el severo desbalance de la variable objetivo. Los resultados evidencian que el modelo Random Forest presentó el mejor desempeño predictivo, alcanzando un AUC de 86.6% en validación y 85.3% en datos reales. Asimismo, el modelo demostró una alta capacidad de priorización, concentrando el 70% de las ventas reales dentro del 20% superior del ranking de propensión. Las variables más relevantes estuvieron asociadas a factores operativos y de interacción, como el resultado del contacto, el tipo de producto y el riesgo del cliente, mientras que las variables sociodemográficas mostraron bajo poder discriminativo. En conclusión, el modelo propuesto constituye una herramienta analítica robusta y operativamente viable para mejorar la eficiencia comercial en los canales físicos, aportando evidencia empírica sobre el valor del Machine Learning aplicado al contexto bancario hondureño.
The low acceptance rate of financial products in physical banking channels represents a significant operational challenge, particularly in contexts where commercial efforts are not analytically prioritized. In response to this problem, this study aims to develop a Machine Learning model to estimate the probability of acceptance of financial products in the physical channels of a banking institution in Honduras, using exclusively anticipatory variables available prior to customer contact. The research adopts a quantitative, applied, and explanatory approach, with a nonexperimental and correlational-predictive design. Anonymized historical data from the period 2023–2025 were analyzed through data cleaning, exploratory analysis, hypothesis testing, and supervised modeling, following the CRISP-DM methodological framework. Four classification models were trained and compared: logistic regression, decision tree, Random Forest, and Gradient Boosting, accounting for the severe class imbalance of the target variable. The results indicate that the Random Forest model achieved the best predictive performance, with an AUC of 86.6% in validation and 85.3% on real data. Additionally, the model demonstrated strong prioritization capabilities, capturing 70% of actual sales within the top 20% of the propensity ranking. The most influential variables were related to operational and interaction factors, such as contact outcome, product type, and customer risk, while traditional sociodemographic variables showed limited discriminatory power. In conclusion, the proposed model represents a robust and operationally feasible analytical tool for improving commercial efficiency in physical banking channels, providing empirical evidence of the value of Machine Learning applications within the Honduran banking context.
The low acceptance rate of financial products in physical banking channels represents a significant operational challenge, particularly in contexts where commercial efforts are not analytically prioritized. In response to this problem, this study aims to develop a Machine Learning model to estimate the probability of acceptance of financial products in the physical channels of a banking institution in Honduras, using exclusively anticipatory variables available prior to customer contact. The research adopts a quantitative, applied, and explanatory approach, with a nonexperimental and correlational-predictive design. Anonymized historical data from the period 2023–2025 were analyzed through data cleaning, exploratory analysis, hypothesis testing, and supervised modeling, following the CRISP-DM methodological framework. Four classification models were trained and compared: logistic regression, decision tree, Random Forest, and Gradient Boosting, accounting for the severe class imbalance of the target variable. The results indicate that the Random Forest model achieved the best predictive performance, with an AUC of 86.6% in validation and 85.3% on real data. Additionally, the model demonstrated strong prioritization capabilities, capturing 70% of actual sales within the top 20% of the propensity ranking. The most influential variables were related to operational and interaction factors, such as contact outcome, product type, and customer risk, while traditional sociodemographic variables showed limited discriminatory power. In conclusion, the proposed model represents a robust and operationally feasible analytical tool for improving commercial efficiency in physical banking channels, providing empirical evidence of the value of Machine Learning applications within the Honduran banking context.
Keywords
Machine Learning, Analítica Predictiva, Banca, Probabilidad de Aceptación, Canales Comerciales, Priorización Comercial