The EHR vendor, the AI vendors, and the model providers are moving to prices set by use. Miso checks each request against a budget by application, department, and route before it is sent, and gives the CIO and finance next month's number while there is still time to change it.
For twenty years a health system bought software by the seat and the module, and the price was known before the year started. AI is arriving on a different meter. The EHR vendor sells its AI features as a suite, with an introductory period that costs less than the years after it and a price that grows with use. Ambient documentation vendors sell by the clinician per month, with a flex tier that bills the AI features by the session. The model providers bill by the token, and the same workload can cost forty times more on one covered model than on another.
The controls a health system already owns assume the old meter. The EHR's admin settings turn a feature on or off. The cloud bill arrives by service, with model calls mixed into everything else. A generic API gateway counts requests and has no idea which department, purpose, or agreement a request belongs to. So the spend report arrives after the month closes, and the only lever left is to turn something off.
A Miso route already names the caller, the purpose, and the agreement. The budget joins them. Before a request is sent, Miso checks all four; a request that would pass its ceiling is refused, recorded, and shown to the owner in the same minute, in place of a line on a bill six weeks later.
Budgets are set where the money is owned: by application, by department, by route, and, where the identity integration is in place, by role. A route can also select the cheapest covered model for the work, inside the policy Privacy approved.
Every request through Miso is already in the record with its caller, route, model, and cost. Eight weeks of that record, plus the routes approved to go live, is a forecast by department for next month, with the reason behind each change and the ceiling it is heading for.
Finance reviews it once a month, the way it reviews any other budget. A department leader sees their own number and the lever that moves it. Forecast against actual is the first measure of whether Miso is earning its place.
Two kinds of AI spend, two kinds of control. Both land in the same console and the same forecast.
Enforced and observed spend together, by department, against the budget the board approved, next to the inventory of what is running.
Next month by department with the levers, forecast against actual, and the vendor invoices due to leave first-year pricing.
Which route selects which model, which ceilings are close, which vendors are inside a budget, and what the identity integration adds.
Spend against ceiling for their applications, refusals with reasons, and the cost of the model behind each route.
A health system already decides who may spend what: approval levels for money, agreements and purpose for data. Miso applies both on each request, so the decision about AI spend is made by the people who own the budget, before the money moves, and the same request record answers Privacy's question and the CFO's.