The Research Guide EconDesk of the ZBW – Leibniz Information Centre for Economics is to be supported by a chatbot in the future. Research Guide EconDesk stuff answers questions on literature search and library services and support users with their individual data search. On the one hand, the chatbot should support the colleagues from the existing EconDesk chat team in processing common user requests, which have increased due to the pandemic, and on the other hand, it should help to expand the range of services to support users from business and economics on the EconBiz portal. Over the last year, our cross-departmental team has been working with different stakeholders on the use cases and is currently working on our first prototype of the chatbot system. We made use of conversational UX design evaluation methods for designing chatbot persona and conversation flows for later implementation. Our chatbot has been developed based on NLP (Natural Language Processing) techniques. Machine learning and rule-based strategies are the main components of this approach and RASA is our main development framework. Another essential part of our project is preparing data for training and testing the machine learning algorithms. We labelled manually real chat-logs to make use of this data for our purposes. The data include intents or user inputs, actions or chatbot answers, and stories or conversation flows. Stories structure flows of conversation and are fundamental parts of the chatbot. Intents that are bases for training the NLU (Natural Language Understanding) and are used to indicate users’ purposes. They are created based on users’ intent and librarians’ experience, must be unique, and are then assigned manually in the chat transcripts.