Intensive outpatient models of need-adapted psychiatric care have been shown to reduce the length of hospital stays and to improve retention in care for people with severe mental illnesses. In contrast, evidence regarding the impact of such models on involuntary hospital treatment and other coercive measures in inpatient settings is still sparse, although these represent important indicators of the patients' wellbeing. In Germany, intensive models of care still have not been routinely implemented, and their effectiveness within the German psychiatric system is only studied in a few pioneering regions. An innovative model of flexible, assertive, need-adapted care established in Berlin, Germany, in 2014, treating unselected 14% of the catchment area's patients, was evaluated on the basis of routine clinical data. Records of n = 302 patients diagnosed with severe mental disorders, who had been hospitalized at least once during a 4-year-observational period, were analyzed in a retrospective individual mirror-image design, comparing the 2 years before and after inclusion in the model project regarding the time spent in hospital, the number and duration of involuntary hospital treatments and the use of direct coercive interventions like restraint or isolation. After inclusion to the project, patients spent significantly less time in hospital. Among patients treated on acute wards and patients with a diagnosis of psychosis, the number of patients subjected to provisional detention due to acute endangerment of self or others decreased significantly, as did the time spent under involuntary hospital treatment. The number of patients subjected to mechanical restraint, but not to isolation, on the ward decreased significantly, while the total number of coercive interventions remained unchanged. Findings suggest some potential of intensive models of need-adapted care to reduce coercive interventions in psychiatry. However, results must be substantiated by evidence from randomized-controlled trials and longer observation periods.