Table 1: Trouble sources

Table 1 shows that a large proportion of the trouble sources which result in bot-initiated user self-repair consists of user inputs of information that was originally prompted by the bot, followed by user-initiated information questions, for example about the procedure, parking information etc. The third main category are user requests to the bot which are often at the start of a new transactional sequence, e.g., to reschedule an appointment.

Step 3: At this step, all instances of user self-repair were further classified by how users conducted self-repair. This resulted in the identification of eight main strategies (see table 2):