Given that telecommunications networks are constantly growing in complexity and heterogeneity, management systems have to work with incomplete data, handle uncertain situations and deal with dynamic environments. In addition, the high competitiveness in the telecommunications market requires cost cutting and customer retention by providing reliable systems. Thus, improving fault diagnosis systems and reducing the mean time to repair with automatic systems is an important area of research for telecommunications companies. This article presents a Fault Diagnosis Multi-Agent System (MAS) applied for the management of a business service of Telefónica Czech Republic. The proposed MAS is based on an extended Belief-Desire-Intention (BDI) model that combines heterogeneous reasoning processes, ontology-based reasoning and Bayesian reasoning. This hybrid diagnostic technique is described in detail in the article. The system has been evaluated with data collected during one and a half years of system operation on a live network. The main benefits of the system have been a significant reduction in both the average incident solution time and the mean diagnosis time.