This thesis aims to explore the effectiveness of the BERT (Bidirectional Encoder Representations from Transformers) model in processing airline itinerary change emails. The research focuses on two downstream tasks. Classification according to the type of flight itinerary change communicated by email. Named Entity Recognition task for extracting relevant data pieces that indicate specific changes in the itinerary.  The study utilises a dataset provided by Kiwi.com, consisting of a partially labeled sample of two months of continuous airline-to-customer email communication. The thesis will describe the overall problem, the methodology employed for model training, the used dataset, the approaches taken during data curation, and an analysis of the model performance on testing data. This research evaluates BERT's capability in handling real-world airline communication data and its potential applications in the travel industry. The practical part will include the implementation of the whole processing pipeline and a small anonymized data sample to verify the pipeline.