We study the effectiveness of virtual and in-person care for diabetic patients. The COVID-19 pandemic significantly accelerated telehealth adoption. Policymakers and hospital systems must now determine how to best utilize telehealth in patient care moving forward. We tackle this question from a causal machine learning and optimization approach. We consider the visit modality, virtual vs. in-person, as a treatment and use causal inference methods to estimate individual treatment effects. These effects inform a scheduling model that optimizes a provider’s virtual/in-person mix. We vary the prioritization of operational vs. clinical outcomes and overall virtual visit limits. Our findings suggest a benefit to increasing virtual care and motivate further policy investigations.