Maintenance Request AI Agent for Property Managers: An Evening Walkthrough
How an AI agent could take tenant repair reports, sort urgency, raise tickets and alert the on-call coordinator, with a person approving every dispatch. A walkthrough of one invented evening.
How can an AI agent handle maintenance requests for a property manager?
It can take the report, ask the follow-up questions a contractor needs, sort it by urgency using rules you wrote, raise the ticket and alert the right person, while a coordinator approves anything that sends a contractor or spends the landlord's money. It cannot diagnose faults or promise attendance times. This page follows one invented tenant message through the system, hour by hour. It is a worked example, not a client project, and no results are claimed. Rules on repairs differ by country and by type of landlord, so treat the legal points as general information.
A request-type guide built with you and your regular contractors, listing emergency, urgent, routine and not-a-repair categories, with the exact questions the agent asks for each.
Intake on the channels tenants already use, such as email, a web form, WhatsApp or SMS, with photo and short video uploads, all landing in one ticket with the same fields.
Tenancy and property lookup, so the agent knows the unit, the building, the landlord, who is responsible for what, and whether the same fault has been reported before.
A fixed safety message for the situations you define, such as gas smells or electrical danger, which tells the tenant what to do first and pages your on-call person at once.
Ticket creation in your property or maintenance software, or a shared board if no integration exists, with a clean summary, photos, access notes and the tenant's preferred contact method.
A dispatch approval screen where a coordinator confirms the contractor, the visit window and any cost limit before anything is booked or promised to the tenant.
Follow-up messages to tenants and contractors, such as visit confirmations and "is it fixed?" checks, each drafted from templates and sent only after your chosen approval rules are met.
A timestamped log of every report, decision and message, which helps if a tenant, landlord or regulator later asks when you knew about a fault and what you did.
What kinds of requests does it sort, and what does it do with each?
| Category | Examples (you define the real list) | Agent action |
|---|---|---|
| Emergency | Gas smell, active flooding near electrics, no way to secure the property | Reads the safety message, pages on-call immediately, creates a ticket marked emergency |
| Urgent | Leak, no heating in cold weather, no hot water, broken lock | Collects details and photos, pages the coordinator, proposes a contractor |
| Routine | Dripping tap, loose handle, appliance fault | Collects details, creates a ticket, proposes a visit window for approval |
| Access or complaint | Neighbour disputes, notice to enter, rent questions | Not a repair. Takes a note and routes it to a person |
The agent's classification is a proposal a coordinator can change in one click, and the change is recorded. When in doubt it goes up a level, not down. The wider approach is described on our AI customer support agent page.
Illustrative scenario: a tenant message at 21:40 on a Thursday
This is a worked example, not a client project. The building, tenant and fault are invented, and no outcome is claimed. Assume a managing company with a request-type guide, an on-call rota and approved plumbing and electrics contractors.
21:40, the message arrives
Tenant, by WhatsApp: "Water is dripping through the bathroom ceiling in flat 3B and there's a brown patch getting bigger. Photo attached. The light in there is still on." The channel gateway creates a draft ticket and wakes the agent.
21:41, what the agent reads
The photo and text, the tenancy record for flat 3B, the building record showing flat 4B sits above, an earlier report of a slow leak from 4B's bathroom, the request-type guide, the on-call rota and the rule that water near a live light fitting is an emergency.
21:41, tools it calls
Read-only: tenancy lookup, ticket history, rota lookup. Then two actions that are allowed without approval: send the fixed safety message to the tenant, and page the on-call coordinator. It cannot book anyone.
21:42, decision and draft
Water plus a light still switched on matches your emergency rule. It sends the pre-approved message: keep clear of the wet area, do not use that light switch, reply if anyone is at risk. The page to the coordinator carries the unit, the photo, the earlier report from above and the category rule that fired. It suggests no cause.
21:48, human approval
The coordinator reads the summary and photo and phones the tenant. They pick the on-call plumber and an electrician, confirm visit windows and tick the cost limit the landlord pre-authorised. Nothing is booked until these clicks. The agent suggested a contractor from the rota; the coordinator swapped them because that firm was on another job.
21:55, write-back
The ticket is updated with the approved contractor and window, the tenant gets a templated confirmation, and a linked ticket is opened for the upstairs unit, with the access request left for a person to send.
Next morning, 08:30
After the visit window passes, the agent drafts an "is it fixed?" message. A reply of "still dripping" reopens the ticket and pages the coordinator again.
Audit log
One timeline holds every timestamp above, showing who decided what and when you first knew.
Which actions run alone, need approval, or stay with people?
| Action | Level | Why |
|---|---|---|
| Send the fixed safety message in an emergency | Alone | Speed matters, and the wording was approved in advance |
| Page the on-call coordinator | Alone | It alerts a person; it commits nothing |
| Propose a category and contractor | Alone as a suggestion | A proposal is not an action |
| Book a contractor or promise a visit time | Approval | It commits a third party and affects the tenant |
| Authorise spend | Approval | Landlord's money, with a limit set by the owner |
| Diagnose a fault or tell a tenant how to fix it | Never | Wrong advice can cause harm |
| Decide responsibility for a repair, or deny one | Never | Needs the tenancy terms and a person's judgement |
| Handle complaints, disputes, vulnerable-tenant situations | Never | Human conversation required |
Which repair rules and timelines might apply?
This is general information, not legal advice. Check with your solicitor or trade body, as rules differ by country, region and type of landlord.
One example from England. Awaab's Law came into force for social landlords on 27 October 2025, according to the government's guidance. In its first phase, it covers all emergency hazards and significant damp and mould hazards. The guidance on timeframes for repairs says landlords must investigate and make safe emergency hazards within 24 hours of becoming aware of them, and investigate significant hazards within 10 working days. It calls these a maximum. Whether the rule covers your properties depends on your landlord clients.
The practical point for an agent is the clock. Many repair rules run from when a landlord or manager became aware of a problem, and an agent that timestamps every report in the same place gives a coordinator a reliable record of that moment. It does not decide whether a hazard is significant or a deadline met. Those judgements stay with a person, and your legal adviser should say how the rules apply to your portfolio.
What if the report is messy, the tenant is upset, or the agent is wrong?
What if the photo is unclear or the tenant describes two faults?
It asks one question at a time and opens separate tickets when two faults need different trades. The report is already logged from the first message.
What if the tenant is frightened, vulnerable or very angry?
Any mention of a child, an older or ill person, no heating in winter or fear for safety goes to a person right away. An upset message is never answered with a long automated reply.
What if the category is wrong?
Sometimes it will be. Emergency rules lean towards over-escalating, reclassifying takes a click, and a weekly review reads the changed tickets and fixes the rules behind them.
What if a message tries to give the agent orders?
Tenant text is data, not instructions, and the agent's permissions limit the damage: it cannot book, spend or change a tenancy record, whatever a message says.
What if the software is down?
The safety message and the on-call page do not depend on the maintenance software. If ticket creation fails, the agent retries, holds the case in a visible queue and alerts a person.
What data does it need, and how would success be measured?
- Tenant and property data: names, units, contact details and tenancy responsibilities, read-only. Under laws such as UK GDPR these are personal data, so keep access narrow and set retention with your adviser.
- Messages and media: photos can show the inside of homes and people, so store them with the ticket only, with a retention period.
- Rotas and contractor lists: kept current by a person; the agent warns when one looks stale.
- Limits: spend limits and approval rules owned by the landlord or your management team, not by the agent.
Success would be measured with metric names agreed before launch and baselines from your own tickets: time from report to first human sight; time from report to approved dispatch; share of tickets reclassified by a coordinator; tickets reopened after "fixed"; follow-up questions needed per ticket; missed or failed pages; and tenant feedback on the process where you already collect it. We make no promise on any of them.
For the safe-approval pattern behind this setup, see our post on human-in-the-loop approvals, and for a voice version of intake see our AI voice agent for real estate page. To discuss your own portfolio, contact us.
How pricing works
We quote every project individually, from the number of systems, integrations, call or message volume and how much is custom. Tell us what you need and we reply with a written quotation.
Frequently Asked Questions
Can an AI agent take maintenance requests from tenants?
Yes. It can receive reports by email, form, WhatsApp or SMS, ask follow-up questions, log photos, sort the request by urgency using your rules, and create a ticket. A coordinator approves any contractor booking or spend. It does not diagnose faults or promise visit times.
How does it know what is an emergency?
It follows a list you and your contractors write, plus fixed rules such as treating any mention of gas or water near electrics as an emergency. It over-escalates when unsure. It does not judge severity itself, and a coordinator can reclassify any ticket.
Can it book contractors automatically?
We would not start there. The agent suggests a contractor from your approved list and a visit window, and a coordinator confirms. Booking and spending commit other people's time and money, so they stay behind approval unless your management team sets narrow, written exceptions.
Does it help with legal repair deadlines such as Awaab's Law?
It helps by recording when each report arrived, which many deadlines depend on. It does not decide whether a hazard is covered or a deadline met. Awaab's Law applies to social landlords in England as of its October 2025 start; check what applies to you with a solicitor.
Does it work with our property management software?
That depends on whether your software offers an API or webhooks and what your plan allows. If it does, tickets can be created and updated directly. If not, the agent writes to a shared board or email for staff to enter by hand. We check access before quoting.
What does a maintenance request agent cost?
We do not publish prices. Each company is quoted after a call, and the quote depends on how many properties and channels are involved, which software must connect, how many request categories and contractors you use, and how detailed your approval rules are. Model usage costs are separate.
Related services
Want this for your business?
Free consultation, no pressure: just a clear plan and a written quotation.
Get a Free Consultation