A central democratic dilemma in western democratic systems is who decide within public authorities. However, the focus the last years has shifted from who to what with the increased focus on artificial intelligence (AI) and the use of automated decision-making (ADM). There is a strand of research nowadays arguing that AI and the use of ADM is growing rapidly in public administrations in the western world. Scholars arguing that this development is currently changing the settings for administrative procedures and citizens’ encounters with welfare services. However, the focus on technical solutions and decision-making within welfare services in social sciences has put certain human power in the shadow. We argue that, what we call, hidden welfare professionals are more powerful than ever within certain welfare areas, and that it’s time to bring the human back in and discuss the limitations of digital solutions and AI tools as explanations for changed application of welfare systems. For instance, AI-scholars argue that automated systems tend to change forms of work in the public sector and thus raise questions about how central democratic administrative values, such as transparency and accountability, are protected. However, we argue that the determination of democratic administrative values such as transparency and accountability are just as less protected and maintained when hidden welfare professionals is gained power of the decision- making. I this article we focus on the role of what we call a hidden welfare professional within the Swedish Social Insurance System (SSIS): the medical insurance advisor (MIA). The medical insurance advisors can be consulted in different ways when case handlers need advice or support on medical issues. Little is known about what matters they are consulted for, how they are consulted or the quality of their work. Knowledge of when and how the medical insurance advisors are consulted in different cases is therefore deficient. This raises questions of democracy, transparency and legal certainty – values that even has been highlighted within the research field of AI. The paper analyses how MIA:s in different ways has impact on the enforcement of the SSIS, both a more passive one and a more active one. The main findings show that MIA themselves see their role as an advisor, but that their impact is more of a decision maker. The analysis also shows how the MIA:s in most of the time get the final word in the assessment om applications within the SSIS, and that their advice has more influence on the decision making-process that both AI and the formal case-maker at SSIA. Therefore, we argue, it’s not AI itself that raise questions about how central democratic administrative values, such as transparency and accountability, are protected. It’s rather fragmentation in the decision process were more hidden actors have gained more impact on the decision.