Most maintenance calls arrive after the office closes, and most of them are not emergencies. An AI agent answers at any hour, works out which is which, raises the work order in your own system, and puts a human on the line only when it genuinely matters.
Gas odour reported in hallway outside unit. Tenant advised to leave the property and not touch switches.
The judgement is the product. Everything else is paperwork.
Voice AI for property management software means answering tenant calls around the clock and deciding what each one is. The agent captures the unit and the problem, classifies the urgency against your rules, writes a work order into your own system, and transfers genuine emergencies to on-call staff while the tenant is still on the phone.
HOW IT WORKS
A tenant with no heating at eleven at night reaches a conversation rather than a beep. That alone changes the outcome, because most people who hit voicemail simply hang up and ring again in the morning, or ring the out-of-hours number twice more first.
It asks the questions your protocol would ask, captures the unit, the location within it, and whether anyone can enter, then decides whether this is a same-night problem or a Tuesday one.
Anything genuinely urgent is transferred to on-call while the tenant is still on the line. Everything else becomes a work order in your system with the detail already filled in.
THE TRIAGE THAT MATTERS
| Priority | Typical call | What the agent does |
|---|---|---|
Emergency | Gas smell, flooding, no heat in winter, broken entry door | Transfers to on-call immediately, then logs it |
Urgent | Boiler down in summer, fridge failed, contained leak | Work order raised with a next-day flag |
Standard | Dripping tap, blocked drain, appliance fault | Logged with access details for the normal queue |
Low | Touch-up paint, sticky door latch, cosmetic damage | Recorded and scheduled with routine work |
Your definitions, not ours. The agent applies the escalation rules you already publish to residents, and asks a follow-up question rather than guessing from one word.
WHAT IT CHANGES
The person holding the out-of-hours phone currently answers everything and decides everything, half asleep. Put an agent in front of it and they only hear from the small share of calls that genuinely need a human decision. That is the change your customers feel first, and the one they will thank you for.
Unit, location within the unit, what is actually happening, whether anyone can enter, and how urgent it is, captured while the tenant was describing it. That is the difference between a job a vendor can accept and one that needs three phone calls before anybody leaves the depot.
Every call, at the hour when voicemail loses them.
Residents explain the problem in their own words.
Weekend enquiries captured, not lost to voicemail.
Hear a difficult call while it is happening.
THE HONEST LIMIT
A gas smell, a fire, water coming through a ceiling or a resident who sounds frightened are not automation problems. The agent should recognise them, say the right thing, and hand over immediately, and the value of the system is measured by how reliably it does that rather than by how many calls it kept to itself. Any product in this category that talks about deflection without talking about escalation is selling you the wrong half.
QUESTIONS
Because that is when most of it happens. Widely reported industry figures put around two thirds of maintenance calls outside office hours, and roughly four in five callers abandon voicemail without leaving a message. The request still exists, it just arrives angrier the next morning, or as damage that has had eight hours to spread.
That judgement is the whole job. A gas smell, water coming through a ceiling, no heat in winter or a broken entry door are emergencies and go straight to on-call. A dripping tap, a squeaky hinge or a noisy neighbour is a work order for the morning. Getting it wrong in either direction costs money, which is why the agent asks follow-up questions rather than pattern-matching on a single word.
Through your own API. The agent calls a function with the fields you specify, so the property, unit, category, priority, description and permission to enter land in your schema exactly as your product expects them. You are the PMS, so we write into you rather than asking you to adopt our structure.
Plenty are prospective tenants asking whether a flat is still available, or existing ones asking about rent dates. The agent handles those from what you give it and can capture a viewing request, which matters because an unanswered leasing call at the weekend is often a lease that goes to whoever picked up.
On outbound calls it does automatically, pulling previous call summaries into the conversation so it knows this is the second time somebody has reported the same leak. Within your product you can also feed it whatever history you hold through a function, so it never asks a resident to explain something they already reported.
The agent handles fifteen languages and switches automatically across ten of them, so a resident can explain a problem in the language they think in. On an emergency call at midnight that is not a nice-to-have, it is the difference between understanding what is wrong and guessing.
Yes. Provision a number per management company or per portfolio over the API and attach it to one shared agent template. Each is isolated with its own call history, so serving hundreds of them does not mean maintaining hundreds of agents.
A flat $0.08 per minute of voice, with no retainer and no per-call minimum. Compared with an answering service that bills whether the call was a burst pipe or a wrong number, the cost tracks what actually happened.
Emergencies still wake somebody. Nothing else does.
Last updated August 2026