We report on experiences with implementing conversational agents in the recruitment domain based on a machine learning (ML) system. Recruitment chatbots mediate communication between job-seekers and recruiters by exposing ML data to recruiter teams. Errors are difficult to understand, communicate, and resolve because they may span and combine UX, ML, and software issues. In an effort to improve organizational and technical transparency, we came to rely on a key contact role. Though effective for design and development, the centralization of this role poses challenges for transparency in sustained maintenance of this kind of ML-based mediating system.

Transparency in Maintenance of Recruitment Chatbots