BACKGROUND The COVID-19 pandemic saw the accelerated uptake of virtual care and a proliferation of virtual ward models as alternatives to facility-based care. Early in the pandemic, our program implemented a virtual mental health crisis ward (vWard) to provide options for individuals needing intense psychiatric and/or crisis support but who preferred to remain in community and were safe to do so. OBJECTIVE This study aimed to identify early learnings from the vWard implemented rapidly in a resource constrained environment, to inform the future state should it be sustained beyond the pandemic. METHODS Mixed methods of data collection were used to evaluate provider perspectives on the vWard, develop archetypes for individuals who are a good fit for the vWard model, and create a driver diagram. Data sources included an anonymous survey of clinical and managerial staff involved in the vWard, a service planning workshop, and analysis of program discharge forms for all individuals admitted between March 2020 and April 2021. Survey responses were coded for themes under categories of “Benefits” and “Challenges.” Discharge forms where the team indicated that the vWard was a good fit for an individual, were examined for characteristics common to these admissions. These findings were reviewed in the service planning workshop and refined with input from the participants into patient archetypes. A driver diagram was created for the future state. RESULTS Survey respondents (N=60) represented diverse roles in crisis services and the vWard team. Ten providers took part in the service planning workshop. A total of 467 discharge forms were reviewed. The vWard was felt to be a model that worked by 39 survey respondents, 1 felt it did not work, and the remaining had no response. A number of benefits for the individual and the system were identified, alongside challenges – some process and material related to the nature of rapid implementation during the pandemic, and others due to lack of fit for certain individuals. The model was felt to a good fit for 67.5% of admissions. Four patient archetypes representing good fit with the model were developed. The driver diagram connected the program aim with primary drivers – (1) reduce barriers to care, (2) improve outcomes, and (3) provide collaborative, patient- and family-centered care – linked to secondary drivers and interventions that leveraged virtual technology among other crisis care interventions. CONCLUSIONS Despite some challenges, the vWard demonstrated high levels of provider acceptance and a range of mechanisms by which the model works for a variety of patient archetypes. These early learnings provide a foundation for growth, sustainability and spread going forward beyond the pandemic.