Artificial Intelligence for Customer Service in Tourism: Prototyping an AI Chatbot at Wonderland Viaggi

The rapid development of Generative Artificial Intelligence and Large Language Models is creating new opportunities for customer service automation. This thesis investigates how Generative AI can be integrated into the customer service process of a small travel agency to support routine and information-intensive interactions while preserving human involvement in activities requiring professional judgement. The research adopts a process-oriented approach based on Wonderland Viaggi, an independent Italian travel agency, combining Business Process Management, Customer Relationship Management and Artificial Intelligence. Drawing on direct observation conducted during an internship, the existing customer service process was reconstructed through an AS-IS analysis to identify recurring activities and operational criticalities. A TO-BE configuration was subsequently developed to define the role of conversational AI, together with an operational profiling framework supporting the classification, prioritisation and routing of customer requests. Based on this analysis, a conversational AI prototype was designed using a Retrieval- Augmented Generation approach, combining a Large Language Model with an organisational knowledge base, behavioural instructions and human-handover mechanisms. The prototype was assessed through scenario-based testing representing recurring customer-service situations. The results show that the proposed approach can support selected routine interactions, particularly when requests are associated with explicit and well-structured information. However, the evaluation also identified limitations in factual and source adherence when available knowledge was incomplete or fragmented, highlighting the importance of knowledge-base quality, prompt configuration, operational boundaries and human oversight. Overall, the thesis shows that conversational AI adoption is an organisational rather than purely technological challenge. The proof of concept provides a framework for integrating Generative AI into customer service while preserving human expertise.