What counts as ‘good work’? Data-driven approaches and situated judgement in frontline eldercare

As ambitions for an ‘AI-driven public administration’ intensify in Sweden, eldercare is increasingly framed as a site for data-driven governance. In an ethnographic study of a municipal out-of-hours service where an LLM-based feedback system was proposed, we examine how these ambitions confront the situated and uncertain conditions of frontline eldercare. We show that ‘doing a good job’ involves assembling partial information, negotiating ambiguity and attuning to older adults’ needs: an epistemic orientation which stands in stark contrast to the data-driven logic embedded in the proposed system. Using the concept of socio-digital care arrangements, we show how the proposed feedback system materialises a particular care disposition in which care becomes knowable primarily through its documented traces in decisions, case records and outcomes. We further show how competing understandings of care quality become distributed across Problem, Institutional and Social spaces (PSI framework), generating tensions in how quality is defined, recognised and valued. The article contributes an empirically grounded account of how data-driven ambitions intervene in frontline eldercare and identifies conditions under which data-driven systems may support rather than undermine situated professional judgement.