AI can absorb real senior living administrative workload without touching care. URBI runs the community and building side only, not clinical care, care records, or medication. This guide draws the line between the non clinical work AI should take on and the judgement that must stay with your team members.
Why is administrative workload a retention problem in 2026?
Because clerical overload is one of the few pressures an Executive Director can relieve quickly, though it is not the largest. The 2026 senior living workforce report from Argentum and Activated Insights found that 62.3 percent of workers named non competitive pay as the top reason they would leave the industry, and 85.4 percent said fair pay is extremely important. Nearly half pointed to poor relationships with supervisors and culture problems. Nearly 95 percent said they are likely to stay where their work has purpose.
Read those numbers together and the conclusion is uncomfortable for software vendors. James Balda, CEO of Argentum, said it plainly in the announcement: "purpose alone is not enough to sustain a workforce". Neither is software. AI does not raise pay, fix a difficult supervisor, or add a person to a thin evening shift.
What AI can do is stop burning the hours your team already has on repeat questions, document hunting, and message sorting. Leaders already expect a lot. A 2025 Argentum technology survey reported by HealthTech Magazine found that 76 percent of surveyed leaders said AI will be a transformative or positive force, up from 58 percent in 2023, while only 49 percent felt somewhat confident deploying targeted AI. Expectation is running ahead of competence. A bounded scope closes that gap.
What does URBI actually touch, and what does it not?
URBI runs the building and community operations layer, and nothing clinical. It holds no care records, manages no medication, makes no care judgements. That boundary is a product fact, not a disclaimer.
Inside that boundary, URBI covers the work that fills an Executive Director's day:
- Maintenance and housekeeping requests, from intake through work order tracking and closeout
- Dining and amenity reservations, visitor management, and parking
- Life enrichment programming and the community activity calendar
- Communication with residents and family members across voice, SMS, email, and in app chat
- Documents, forms, policies, and community announcements
- Accounting, which is In Beta
One more thing, said directly. URBI has no live senior living communities today. Everything here runs in residential and commercial buildings, where the same jobs exist under different names: a resident asks a repeat question, a request needs routing, a family member wants a status update, an operator needs an answer from live records. The capability transfers. The customer list does not yet include a community, and we will not pretend otherwise.
Which non clinical workloads can AI responsibly absorb?
The ones where the output is a draft, a summary, or a routing decision a person can check before anyone acts on it. Reversibility is the test, not how many hours the task consumes. Here is a matrix you can hand to a vendor.
| Workload | What AI should do | What a person still approves |
|---|---|---|
| Message triage and repeat questions | Classify incoming messages, answer known administrative questions from an approved knowledge base, create a callback task | Any message with symptoms, medication, safety, abuse, neglect, or emergency language goes to a person. No model decides a concern can wait. |
| Shift handoff summaries | Transcribe and format operational logistics: completed and open tasks, deliveries, vendor visits, room readiness | Qualified team members validate anything about resident condition, risk, or care instructions. AI formats, it does not assess. |
| Policy and document lookup | Search only approved, version controlled documents and return the exact clause with its source link and effective date | The reader opens the cited source. Compliance leadership interprets exceptions. |
| Family follow up | Draft callback summaries, visit logistics, event reminders, and unresolved questions from approved records | An authorized team member verifies who may receive what, checks the facts, and sends. |
| Ticket summaries | Condense maintenance, housekeeping, and IT tickets, deduplicate reports, identify the asset and location | A person sets safety priority and closes the ticket. Fire, security, elopement, and resident incident tickets skip the queue. |
| Task prioritization | Propose an ordered list for the day from age, category, and stated community rules | The Executive Director or department lead approves the order. Proposal only, never publication. |
Every row shares one shape. AI drafts, retrieves, classifies, condenses. People interpret, authorize, escalate, act.
Which decisions must stay human?
Four categories, none negotiable.
- Care judgement. Diagnoses, medication actions, fall response, level of care, and care plan changes. NIST's generative AI risk profile warns that a confabulated summary of health information could cause doctors to make incorrect diagnoses. A fluent summary is not a verified one.
- Resident health information. A message that turns clinical mid sentence stops being administrative. Build the escalation path first, then the automation.
- Decisions about people. Hiring, discipline, promotion, accommodation, and adverse scheduling outcomes. The EEOC states that federal employment discrimination laws protect workers when AI systems are used to discriminate. In New York City, covered automated employment decision tools must have been subject to a bias audit within one year of use, with notice given in advance.
- Anything with legal consequence. Notices, resident agreements, incident documentation, and regulatory filings carry a signature, and a signature belongs to a person.
Requirements vary by jurisdiction and license type. That is not a hedge. The National Center for Assisted Living reported that assisted living regulations in 18 states, including the District of Columbia, were updated during 2025, with seven changing staff scheduling requirements. A national AI policy is no substitute for checking your own state and license.
Kristi Stoglin of Parasol Alliance framed it well at a 2025 LeadingAge Oregon session, arguing that AI should "empower staff and enrich resident relationships, not replace the human touch".
How do Arthur and HERO work without acting on their own?
Both operate under one rule: AI proposes and people approve, with a permanent audit trail. Neither takes autonomous action.
Arthur is the resident and family facing AI. It answers residents and family members across voice, SMS, email, and in app chat in ten languages. It acts through PIN verified tool access, so it opens a ticket, books an amenity, registers a vehicle, or pre schedules a visitor only after identity is confirmed. When Arthur does not know something, it does not guess. It escalates to your team with a summary, relays the approved answer in its own voice, then saves that answer to the community scoped knowledge base your team curates and can view, edit, or delete. That is the part that matters for workload: your team's judgement gets recorded once and reused, instead of repeated every afternoon.
HERO is the Executive Director facing AI inside URBI. It answers questions and builds reports from live records, so you can ask what is open, what is overdue, and what changed this week without exporting anything. It reports on the operational side only, because that is the only side URBI holds.
Neither product replaces the person who decides. Keep that in mind whenever a vendor promises full automation. It is the same argument we make about operators and AI.
What auditability should you demand from any AI vendor?
Ask for a record that shows the decision, not just the output. If a vendor cannot produce these, the tool is not ready for a community.
- A named use case boundary. The intended task, the prohibited clinical and employment uses, and the escalation path, in writing.
- A reproducible log. Time, user, model version, input records referenced, documents retrieved, the output, and where it went.
- Source provenance. Every policy answer cites the exact document version and effective date.
- A human approval trail. Who approved, when, what they edited, what they rejected.
- Override and appeal. Who can stop the automation, reroute to compliance, and restore the old workflow.
- An error register. False summaries, missed escalations, duplicate tasks, and the corrective action taken.
- The data chain. If you are a covered entity and the vendor handles health information, HHS requires satisfactory assurances, in the form of a contract or other written arrangement. A consumer AI subscription does not meet that bar.
- Performance by task. Baseline minutes, drafting time, human review time, correction rate. Time saved before review is not time saved.
Frequently asked questions
Does URBI touch care records or medication?
No. URBI runs building and community operations: maintenance, dining and amenity reservations, visitor management, life enrichment programming, documents, and communication with residents and family members. It does not store care records, manage medication, or produce clinical documentation. If a message or ticket contains clinical content, the design intent is to escalate it to a qualified person rather than process it.
Can AI reduce turnover among team members?
Not on its own, and any vendor claiming otherwise is overreaching. The 2026 Argentum and Activated Insights report puts pay, supervisor relationships, culture, and staffing first. Reducing clerical load is a supporting move that gives people more of their shift back for residents. Treat it as workload relief and measure it that way, including review time.
How do we start without risking a resident safety issue?
Start with the reversible rows above: policy lookup, ticket summaries, and repeat administrative questions. Require a citation on every answer. Route every message with clinical, safety, or complaint language to a person automatically. Run it for 30 days, count corrections and missed escalations, then widen the scope. Confirm your state and license requirements before you begin.
What happens when Arthur does not know the answer?
It escalates instead of guessing. Arthur sends your team a summary of the question, waits for the approved answer, relays it to the resident or family member in its own voice, and saves it to the community scoped knowledge base for next time. Your team can view, edit, or delete any saved entry. This is how family communication stays accurate as volume grows.
If you run a community, the honest version of this conversation is the useful one. Start with the URBI overview for retirement communities, then write to hello@myurbi.co and tell us which administrative workload is eating your week. We will say plainly which parts we would automate, which we would not, and what audit record you get either way.
