Expanding access to primary care requires better strategies for assigning patients to providers, both when forming new panels and when rematching patients due to provider turnover. Through partnership with a large regional health system, we identify key operational challenges that contribute to mismatches between patient needs and provider capacity. We address these challenges using a combination of machine learning and optimisation-based approaches.

This talk focuses on the growing challenge of rematching patients in the face of provider turnover, which has become increasingly common and administratively burdensome. We propose a novel patient-provider matching framework in which patients are offered personalised menus of providers and respond sequentially. This hybrid offline-online setting introduces a new variant of assortment optimization that balances patient choice with system-wide match quality. Using a semi-synthetic, Medicare-based dataset calibrated to our partner health system, we show that tailored menu policies improve match quality by 13% over a greedy benchmark. Our results highlight a tradeoff between preserving patient choice and optimising overall system fit, illustrating how limited, well-structured menus can expand access while supporting continuity and autonomy in primary care. This is joint work with Naveen Raman (Carnegie Mellon University).