There will always be situations in which the chatbot does not have a correct answer ready - for example for topics it does not (yet) know - or should not give an answer, such as. for a request to terminate a contract or for legal advice.

Our framework can detect this in different ways:

• if the chatbot cannot assign a user input several times.
• when a certain topic is recognized, e.g. Termination of contract or legal advice.
• when a user asks directly for help from a person.

In these cases, the chatbot can transfer the conversation in the background to a service employee ("agent") - a so-called "human handover" or "live chat". He can then chat with the user and continue the conversation, answer questions and obtain further information.

If the human handover is ended by the user or the employee, the chatbot takes over the conversation again and continues to automatically answer questions from the user.

The chatbot also supports the employee during a human handover. The input of the user is analyzed by the chatbot and on this basis suggested answers are displayed to the employee.

In the editorial post-processing, responses from employees can also be adopted as new chatbot content with just a few clicks, so that the chatbot continues to learn from handover conversations.