Thesis Mejora del desempeño operacional del Service Management mediante la implementación de un framework AIOPS empresarial: caso de una organización de transporte aéreo
Loading...
Date
2026-07
Authors
Journal Title
Journal ISSN
Volume Title
Program
Ingeniería Civil Informática
Departament
Campus
Campus Casa Central Valparaíso
Abstract
La gestión de incidentes tecnológicos en organizaciones de alta criticidad enfrenta ineficiencias estructurales que impactan directamente la continuidad operacional: tiempos de resolución (MTTR) superiores a 15 horas, triage manual que consume más de 15 minutos por ticket, duplicación de aproximadamente el 20 % del trabajo y subutilización de más del 80 % del conocimiento histórico resuelto. Estos problemas, identificados y cuantificados en el contexto real de una organización de transporte aéreo latinoamericana, motivaron el desarrollo de la presente solución. Esta memoria presenta el diseño, implementación y evaluación de un sistema AIOps enfocado en el módulo de Service Management, desplegado sobre infraestructura cloud-native y orientado específicamente a resolver las brechas operacionales descritas. La propuesta integra cuatro capacidades principales: clasificación automática de tickets mediante Random Forest con TF-IDF (87,3 % de accuracy en categoría), búsqueda semántica de casos similares con Sentence-BERT y pgvector (Precision@10 de 78 %), enriquecimiento contextual en tiempo real desde una plataforma de monitoreo y un bot ChatOps para notificaciones y resolución colaborativa. Los datos utilizados corresponden a un dataset anonimizado con fines académicos. La evaluación se realizó en un programa estructurado de 10 semanas (fase técnica, piloto controlado y rollout extendido), con hasta 55 analistas en la fase final y aproximadamente 1.840 tickets reales. Los resultados reportan una reducción del MTTR de 24,6 %, aceleración del triage en 57 %, disminución de la duplicación en más de 10 puntos porcentuales y una adopción del 89 % en la cuarta semana. El ROI estimado alcanza el 3.072 %, con un periodo de recuperación inferior a un mes. Todos los resultados son estadísticamente significativos (p < 0,05), evidenciando la viabilidad e impacto operacional de AIOps aplicado a la industria aeronáutica latinoamericana.
IT incident management in high-criticality organizations faces structural inefficiencies that directly impact operational continuity: resolution times (MTTR) exceeding 15 hours, manual triage consuming over 15 minutes per ticket, duplication of approximately 20% of the work, and underutilization of over 80% of resolved historical knowledge. These problems, identified and quantified in the real-world context of a Latin American air transport organization, motivated the development of this solution. This thesis presents the design, implementation, and evaluation of an AIOps system focused on the Service Management module, deployed on cloud-native infrastructure and specifically aimed at addressing the operational gaps described. The proposal integrates four main capabilities: automatic ticket classification using Random Forest with TF-IDF (87.3% accuracy in category), semantic similar-case search with Sentence-BERT and pgvector (Precision@10 of 78%), real-time contextual enrichment from a monitoring platform, and a ChatOps bot for notifications and collaborative resolution. The data used corresponds to an anonymized dataset for academic purposes. Evaluation was conducted through a structured 10-week program (technical phase, controlled pilot, and extended rollout) with up to 55 analysts in the final phase and approximately 1,840 real tickets. Results report a 24.6% MTTR reduction, 57% triage acceleration, duplication decrease of over 10 percentage points, and 89% adoption by week four. The estimated ROI reaches 3,072% with a payback period of less than one month. All results are statistically significant (p < 0.05), demonstrating the feasibility and operational impact of AIOps applied to the Latin American aviation industry.
IT incident management in high-criticality organizations faces structural inefficiencies that directly impact operational continuity: resolution times (MTTR) exceeding 15 hours, manual triage consuming over 15 minutes per ticket, duplication of approximately 20% of the work, and underutilization of over 80% of resolved historical knowledge. These problems, identified and quantified in the real-world context of a Latin American air transport organization, motivated the development of this solution. This thesis presents the design, implementation, and evaluation of an AIOps system focused on the Service Management module, deployed on cloud-native infrastructure and specifically aimed at addressing the operational gaps described. The proposal integrates four main capabilities: automatic ticket classification using Random Forest with TF-IDF (87.3% accuracy in category), semantic similar-case search with Sentence-BERT and pgvector (Precision@10 of 78%), real-time contextual enrichment from a monitoring platform, and a ChatOps bot for notifications and collaborative resolution. The data used corresponds to an anonymized dataset for academic purposes. Evaluation was conducted through a structured 10-week program (technical phase, controlled pilot, and extended rollout) with up to 55 analysts in the final phase and approximately 1,840 real tickets. Results report a 24.6% MTTR reduction, 57% triage acceleration, duplication decrease of over 10 percentage points, and 89% adoption by week four. The estimated ROI reaches 3,072% with a payback period of less than one month. All results are statistically significant (p < 0.05), demonstrating the feasibility and operational impact of AIOps applied to the Latin American aviation industry.
Description
Keywords
Industria Aeronáutica, Gestión de Incidentes, Service Management, Machine Learning, Inteligencia artificial
