The industrial project partner offers online brokering of loans, mortgages and insurances. The brokering process involves several phone calls of a call agent with a potential client. This is time-consuming and highly repetitive. For this reason, we want to develop a dialogue system which can take over part of the tasks: it will initiate calls to potential clients, aggregate missing data and answer user questions.Brokering loan and mortgage applications is a very sensible and delicate topic, and requires a huge amount of trust between client and service provider. This has to be accomplished also by an automated dialog system. For this reason, one main innovation in this project is to guarantee a “trustworthy” appearance of the system. For the implementation, existing tools for Speech-to-Text (STT) and Text-to-Speech will be used for the spoken dialogue processing. The STT component will be extended to handle typical accents of German in Switzerland (“Swiss High-German” and non-native speakers, but not Swiss-German). A dialogue manager will be trained using reinforcement learning, and utterances are generated with a combination of data-driven natural language generation or template-based text generation with slot-filling. In addition, an information extraction component will be trained on a newly generated spoken dialogue corpus, to identify relevant data entries in arbitrary textual formulations.The solution is available 24/7 for the customers and will be deployed in three sub-companies of the industrial partner.