What an operating system actually is
An EHR records care. A CRM records prospects. Payroll records hours. A general ledger records money. None of them is designed to answer a portfolio-level question, because none of them holds the other four domains.
An operating system sits above those systems. It ingests them, applies one governed set of definitions, and produces a single operating record that care, labor, census, compliance, and finance all read from. The test is simple: when your regional director, your CFO, and your REIT partner each pull occupancy, do they get the same number, computed the same way, with the same lineage?
The seven criteria that matter
- Coverage — does it span care, labor, census, compliance, and NOI, or only the domain the vendor came from?
- Governance — can you set the definitions yourself, version them, and see when they changed and who changed them?
- Integration posture — does it read your existing systems, or does it require you to replace them first?
- Lineage — can any number on a dashboard be traced back to the source record it came from?
- Time to first governed number — weeks, or a multi-quarter implementation before anything is usable?
- Portability — can you export the model and the underlying data in a usable form, at any time, without a negotiation?
- Multi-state fit — does it handle differing state survey regimes, payer mixes, and licensure structures without forking your definitions?
Questions to put to every vendor in writing
- Show me one number end to end: where did it originate, what transformed it, and who approved the definition?
- If we change the definition of stabilized occupancy, what breaks, and who has to approve it?
- What is the first governed metric we will have, and on what date?
- What does day one look like if we keep our current EHR, CRM, payroll, and GL exactly as they are?
- On termination, what do we receive, in what format, and how long does it take?
- Which of these capabilities are live today for a customer of our size, and which are roadmap?
Where evaluations usually go wrong
Two failure patterns repeat. The first is buying by feature list: the demo is impressive, but every module carries its own definitions, so fragmentation is reproduced inside a single vendor rather than resolved.
The second is buying by replacement: the vendor requires you to move off the systems your staff already know before anything works. Implementation stretches, operators lose confidence, and the governed record never arrives.
The alternative is to require the record first and the interface second — instrument what you already run, agree the definitions, then decide what, if anything, needs replacing.