The problem
A maintenance request agent answers a problem every property manager knows: repair requests arrive everywhere, at every hour.
A tenant sends a blurry photo of a ceiling stain on Sunday night, another calls about a dripping tap, a third emails about a door that won't lock. Each one needs the same steps: understand it, judge how urgent it is, find a vendor, agree a time, tell the tenant.
When those steps run through one person's phone and inbox, two things go wrong. Routine jobs wait days for a booking, and the rare serious one, a gas smell or water near wiring, can sit behind twenty small requests.
Record keeping suffers too. When a tenant or a regulator later asks when a problem was reported and what happened next, the answer is scattered across chats and call logs.
What we would build
We would build an AI agent that runs on the WhatsApp Business Platform, the channel many tenants already use. Its goal: turn each message into a well-described, correctly sorted job with a booked vendor or a person on it. Its tools are WhatsApp, your property management system, your approved vendor list with calendars, and your on-call alert.
Photo and description intake, with follow-up questions to fill the gaps.
Urgency triage against rules your team writes: emergency, urgent or routine.
Vendor booking from your approved list, within a spend limit per job type.
Tenant updates at each stage, so nobody has to call to ask.
One job record holding the photos, messages, choices and approvals.
How it works
A tenant sends a photo and a line of text. The agent matches the phone number to the tenancy, then asks what it needs: which room, since when, whether water, power or heating is affected, and when someone can let a vendor in.
It reads the photo to help describe the job, for example “brown stain on bedroom ceiling, bathroom directly above”.
Next comes triage. If anything suggests danger, the agent tells the tenant a person is being contacted now, gives your emergency safety message and alerts the on-call manager by phone and WhatsApp. For routine jobs, it checks the approved vendors for that job type and building, offers two or three slots and books the one the tenant picks.
The vendor gets the job details and photos. The tenant gets a confirmation, a reminder the day before, and a message when the job is marked done, with a quick question on whether it's fixed.
What the AI does, and what your team decides
Underneath sits a standard agent design: one goal, a short list of tools, approval points and a full log. The limits are set by your team, not by the model.
The agent may act alone to collect details and photos, ask follow-up questions, sort routine jobs, book approved vendors within the spend limit, and send status updates.
A person must approve any job above the spend limit, any vendor not on the list, any request for access without the tenant present, and every possible emergency.
The agent never tells a tenant that a possible hazard is safe, promises compensation or rent changes, or discusses arrears.
If the agent can't tell how urgent something is, it treats it as urgent and hands it to a person.
What the research says
Tenants are already on WhatsApp. Ofcom's Online Nation 2024 report found WhatsApp reached 87% of UK online adults in May 2024, and 64% used it every day, up from 58% a year earlier (Ofcom, 2024). Taking requests there means no new app or portal login for the tenant.
People will hand routine steps to AI. In customer service vendor Zendesk's CX Trends 2025, 67% of consumers said they are ready to delegate tasks such as tracking orders to AI (Zendesk, vendor research). Booking a slot and sending job updates are the kind of routine steps the agent takes on.
Agents should ask before acting. The OWASP Top 10 for LLM applications recommends that a human approve high-impact actions before an AI system takes them. Here, that covers emergencies and costly jobs.
Guardrails
Data access. The agent sees the tenancy, the property and the job history for the number that messaged it, and nothing about other tenants. Vendors receive only what they need for the job.
Human approval. Vendor lists, spend limits, emergency keywords and message templates are set by your team and stored as settings.
Logging. Every message, photo, urgency choice, booking and approval is timestamped on the job, so you can show exactly when you knew and what you did.
WhatsApp rules. Messages follow the platform's business policies, with approved templates for any update sent outside an open chat.
Related reading: AI in Real Estate Operations: Leases, Tenants and Maintenance
Is a maintenance request agent right for you?
It fits managers whose repair inbox has outgrown one person. Ask yourself:
How do tenants report repairs today, and how many arrive each week?
Do you have written rules for what counts as an emergency?
Do your vendors share availability, or is every booking a phone call?
What spend limit would you trust the agent to book within?
Not sure whether you need an agent or a simpler bot? Read chatbot vs. AI agent. Our AI agent development and WhatsApp and chatbot development pages explain how we build, or read more about our real estate software work.
