Subject indexing, i.e., the enrichment of metadata records for textual resources with descriptors from a controlled vocabulary, is one of the core activities of libraries. However, due to the proliferation of digital documents it is no longer possible to annotate every single document intellectually, which is why we need to explore the potentials of automation. At ZBW the efforts to partially or completely automate the subject indexing process have started around the year 2000 but the prototypical machine learning solutions that we developed in an applied research project over the past few years have yet to be integrated into productive operations at the library. In this short paper, we outline the challenges that we perceive and the steps that we are taking towards completing the transfer of our solutions into practice – in particular, we are in the process of specifying what a suitable architecture for that task should look like and establishing a roadmap for the next two years indicating the milestones that have to be reached in order to build and test that architecture and to subsequently ensure its availability and continuous development during running operations.