Short answer: AI should read the request, check the property's history, draft the response, and get real emergencies to a human immediately. What it should never do is trap a resident with a burst pipe in a chatbot loop.
Every property manager knows the two failure modes of maintenance. Under-responding: the request that sat in a portal for four days and turned a repair into a habitability fight. Over-automating: the "AI-powered portal" that makes residents describe a gas leak to a dropdown menu. Residents remember both. Renewals are won or lost on how the hard moments were handled.
What good triage actually is
Triage is three decisions, made fast, on every request:
- What is this? A leak with active water is not a "maintenance ticket" — it's an emergency, and the clock started when the resident hit send.
- What do we already know? Has this unit had this problem before? Which vendor handled it? What part was installed? The property's own history usually contains the answer — if anyone can find it.
- Who acts next? Emergency → human, now. Routine → drafted response, scheduled, resident informed. Ambiguous → a human looks at it, not a form letter.
Software does the reading and the drafting. A person makes the call. That's the design.
What "the AI knows your properties" changes
Generic tools treat every request the same because they know nothing about the building. An assistant connected to your records is different in ways residents feel:
- It recognizes the recurring issue — "this water heater was serviced in March by Carter Plumbing; here's the work order" — and drafts the vendor follow-up instead of starting cold.
- It matches the resident's actual address and unit, not a half-remembered conversation.
- It drafts updates in your voice — "part ordered, arriving Thursday, installer scheduled Friday morning" — for your team to review and send, so residents stop wondering whether anyone read their request.
- It escalates by your rules: active water, no heat, safety words — straight to the on-call human, with the property history attached.
What stays with humans, always
- Emergencies. The assistant's job is to recognize and escalate in seconds, not to handle.
- Habitrustability and legal calls. Habitability disputes, reasonable-accommodation requests, anything touching Fair Housing — people decide, always.
- Vendors and money. Dispatch approval, invoices, what gets spent: your team.
This is also where the Fair Housing line lives: AI that never decides outcomes can't discriminate, and a human-approved workflow creates the record that shows it.
The questions to ask before you automate any of this
- How fast does a real human see an emergency? (Seconds should be the answer.)
- Does it read our property history — work orders, vendors, unit records — or does it treat every request as brand new?
- Who approves messages that go to residents?
- Where does resident information live, and can the whole thing run on hardware at your office if that's your policy?
A realistic first step
Take one building or one portfolio segment. Route maintenance through the assistant for thirty days with your team reviewing everything. Watch two numbers: emergencies-escalated-fast and resident complaints-about-bots. The first should be perfect. The second should be zero.
That's the deployment: learn the portfolio, connect to the systems you already run, draft everything, escalate instantly, decide nothing. Talk with us — or read the Fair Housing boundaries every PM should set on AI.