Coherence before Intelligence.
Vertical AI vs. horizontal decisions in senior housing & care
Vertical AI works inside one department or system: it flags a coding issue, finds overtime, or analyzes payer mix. Horizontal decisions span several departments at once. Vendor AI can improve each lane, but no single vendor owns the trade-offs between lanes. Those decisions need operator-declared authority and an Operator-Controlled Operating Record that preserves the evidence, the disagreement, who decided, and why.
SeniorCRE® Governance Desk · Educational guide · Updated
What is the difference between vertical AI and horizontal decisions?
The difference is not how sophisticated a model is. It is the scope of the problem and the authority behind the decision.
| Dimension | Vertical optimization | Governed horizontal decision |
|---|---|---|
| Focus | One workflow or department | Organization-wide outcome |
| Inputs | Data from one source or lane | Clinical, therapy, workforce, finance and quality together |
| Decision | Local tool or department rules | Shared evidence plus explicit decision rights |
| Example | Flag an MDS coding inconsistency | Coordinate function, staffing, documentation and care |
Why not let each vendor govern its own AI?
Vendors should govern their own tools: identity, access, actions, security and audit logs. Those safeguards do not, by themselves, settle organization-wide trade-offs. A staffing agent cannot decide how much resident-function risk is acceptable to save labor expense. A reimbursement agent does not own clinical discharge readiness.
Vendors can govern their agents. The operator must govern the decisions those agents make on its behalf.
An example: residents declining in walking and daily activities
Illustrative example — not a customer case
More residents in a skilled nursing community need increased help with walking and daily activities. Leadership wants to improve functional outcomes and the related Five-Star quality measures while managing staffing and financial constraints.
- ClinicalADL charting
- TherapyFunction trends
- StaffingCoverage and hours per resident day
- QualityMDS measures
- FinanceResource cost
The vertical answer: One system flags functional decline. That helps, but it does not tell leaders whether the response should be a therapy reassessment, restorative nursing, a staffing adjustment, a documentation review, or several coordinated steps.
How a governed horizontal decision works, in six steps
- 01
Evidence
Combine permitted clinical, therapy, staffing, quality and finance information.
- 02
Disagreement
Surface differences in definitions or observations instead of hiding them.
- 03
Authority
Specify who owns clinical changes, staffing changes and escalation.
- 04
Operator-Controlled Operating Record
Preserve what was known, why a choice was made, and who approved it.
- 05
Intelligence
Let AI compare appropriate scenarios and competing constraints.
- 06
Execution
Act only within permissions, then observe functional and operating results.
The operator chooses the deployment. Keep the clinical, workforce, therapy and finance systems you run today and SeniorCRE governs the decisions that depend on more than one of them, or select SeniorCRE itself, including its own EHR and eMAR, as the system of record for chosen functions. Either way, one authoritative record per function, and the operator decides what governs.
Where should an operator start?
Start with one cross-functional decision, define the data it needs, establish whose authority governs, and give AI narrowly permitted work. Begin with read-only analysis; add approved actions or write-back only when integrations, validation, controls and responsible people are in place.
Not every AI assistant or one-off tool needs an enterprise governance layer. The case is strongest when AI draws on several systems, affects competing objectives, or starts consequential workflows. This is a proposed operating model: no live named third-party connector is claimed, and specific integrations and performance gains must be validated.
Interoperability moves data. An operator-controlled operating record makes enterprise AI possible.