User Experience is one of the main aspects that maintain a customer loyal to cloud based solutions or SaaS (Software as a Service). With the rise of the natural language processing techniques, the industry is looking at automated chatbot solutions to boost and expand their services. This thesis presents a practical case study of the implementation of a chatbot solution to complement a CRM (Customer Relationship Management) software called FOXAIO, and then quantify, following the most appropriate guides and solutions available, the User Experience (UX) optimisation. In order to create a robust and scalable solution based on the constraints created by the company in the case, we reviewed the current deep learning techniques, tools and libraries available to help the development process. The most proven techniques in the field of Natural Language Processing (NLP) will be introduced. To achieve the goals of this solution without "reinventing the wheel", we present possible architectures to use at the top of some open source and available tools on the market, with a special relief in the framework RASA. Also we discussed some of possible techniques to create the intent classifier, where we detail the better performance in the top of the rasa tensorflow embedding pipeline for this particular case. The conversational system, also, required a channel to interact with the final user. To achieve that, we also implemented a basic chat interface created on the top of the socket protocol, which communicate with the conversation system. In any case, it would be possible to extend to the other channel’s available on the market, like messenger, slack, telegram. Finally, we detail with a few use cases, that’s hypothetically possible to improve the user experience of an existing software system (FOXAIO) using a conversational interface on the top of that. Also, we achieved some highlights about the preference to use a conversational interface because of his simplicity, defended by a better score in the SUS scale, 70 against 58 to the traditional UI, and good indicatives by the HEART framework.