Start with the cohort, not the staffing model
The first decision is which patients the program exists for. A heart failure cohort, a frequent ED user cohort and a post-overdose cohort produce different protocols, different visit cadences, different escalation paths and different metrics. Programs that begin with 'we will hire two community paramedics' and choose the population later spend their first two quarters improvising.
Pick one cohort with a clear referral source and a countable outcome. Add a second only once the first runs without a spreadsheet.
Agree the referral path before the hire
A program with no reliable referral path idles. Decide in writing who refers, from what system, into what queue, and what happens when nobody accepts within the window. If the hospital is on Epic, the referral should originate in the chart. If it is not, a secure web form is a complete answer for v1.
Fax is a fallback, not a path. It leaves no status and no record either party can report on.
Define the outcome fields with the funder in the room
The state or hospital sponsor will eventually ask what the program changed. Every field they will need should be a required field on the mission template from the first visit, and no field they will not need should be required at all.
Four to six outcome fields per protocol is typical. More than that and crews will work around the system.
Budget coordination and evaluation as line items
Under a Rural Health Transformation subaward, coordination software and program evaluation are ordinary budget lines alongside workforce and telehealth equipment. They are also the lines that make the workforce line defensible, because they are what produces the evidence.
Write them as new program infrastructure. That framing is what non-supplantation asks for.