Diseño de una herramienta Inteligente para el análisis del cumplimiento de los SLAs de los servicios de clientes

Abstract

This work presents the design and implementation of an intelligent tool for estimating call volume at the Azteca Experience Center (CEA), with a 30-minute forecasting horizon. The solution integrates a workflow from Aheeva → Python → Excel → Power BI, automating data extraction, preprocessing, variable generation, and operational visualization. Several supervised learning approaches were evaluated, and an XGBoost-based model was selected due to its strong balance between accuracy and efficiency. The variable set includes exogenous signals (month, day of the week, time interval, and hourly flag) and t-1/t-2 lags of operational indicators (offered calls, incoming calls, ASA, and AHT), enabling the capture of temporal dependencies and atypical conditions through a failure indicator. In the temporal validation stage (final 20%), the model achieved reference performance metrics of RMSE = 2.82, MAE = 1.59, and R² = 0.672. The tool generates a single output file for Power BI and automatically refreshes the dashboard, allowing the predictions to support staff scheduling and SLA compliance monitoring. The results show that combining lagged variables with calendar features improves predictive capability compared to reactive methods and enhances real-time operational decision-making.