Institutional Repository of Vilnius University
Agentais grįstas klientų incidentų valdymo modeliavimas /
Abstract
dc:descriptionThis thesis examines customer incident management by modelling incident assignment, resolution and escalation through an agent-based simulation. The model combines rule-based agent behaviour with an optional large language model (LLM) decision layer. The simulation includes clients, support specialists and a dispatcher, while incidents are represented as tickets with priority, type, complexity, SLA deadline and state. Support specialist behaviour depends on competence, workload, stress, fatigue and error probability. The model is calibrated using a survey of IT incident management employees and an NLP analysis of public posts from the X platform. Survey data define parameters for stress, fatigue, errors, complexity and decision strategies, while NLP analysis estimates incident type and arrival distributions. The simulation is implemented in Python using the Mesa framework. LLM outputs are checked using a Pydantic schema and runtime validation, and incompatible decisions are replaced by a rule-based fallback. The experiments compare dispatcher policies, workload scenarios, L1 escalation thresholds, incident complexity shares and calibration modes. The factorial experiment (4,860 runs, 162 configurations × 30 replications) showed that incident arrival rate (η² = 0.412 for SLA violations) and maximum L1 holding time (η² = 0.550 for errors) are the main operational factors. The calibration mode effect reached η² = 0.963 for the resolved ticket share, confirming the importance of empirical calibration. The LLM layer significantly reduced the error rate by 10.4 percentage points and mean resolution time by 0.61 days (p < 0.001), while SLA violation and resolved ticket share differences were not significant.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2026
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Domarkaitė, Milda,
Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- lit
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.vu.lt/VU:ELABAETD308336630&prefLang=en_US