Machine learning algorithms use each sites’ data to predict future ED volumes. They are automatically tuned to each site’s nuances accounting for factors such as seasonality, holidays, Covid and local patterns of behavior. This is then fed into a simulation engine that accounts for site specific patient mix, acuity and arrival variation. The simulation outputs a probabilistic demand for provider services for each half hour of the scheduling period.
A simplified heat map view of the month shows users what the quality of “fit” is for their default, suggested and any custom schedules that the user has selected. The system suggests optimized scheduling options (solver) and is designed to promote users to use a repeating pattern of schedule to the extent possible to minimize complexity for the schedulers.
For each day of the month, users can “drill down” to see a comparison of forecasted supply vs demand for various potential schedules. This information helps users to understand why the system is suggesting certain scheduling options and where the fit of supply vs demand is under/over vs forecast.
Ensue partnered with AndrewDayLLC and Novasim to design, prototype and then build this system for SCP. Development was agile and the system continues to evolve both adding new features and improving accuracy over time. Ensue provided service and support throughout the first year of adoption and use while training SCP to become self sufficient.