An AI agent that diagnoses tenant issues and coordinates repairs
We built an AI property maintenance agent for a property management company. Tenants describe an issue in conversation, the system diagnoses it, guides a safe next step and coordinates the right vendor when a visit is needed.
- Guided intake
- every issue captured in one conversation
- Safety controlled
- remote guidance follows explicit rules
- Multi-vendor
- coordination continues until a visit is accepted
- Now rolling out
- with the first landlords and property managers
The maintenance handover
A tenant usually knows what they can see, hear or smell, but not what has failed. Their first message may be urgent, incomplete or difficult to classify. A property manager then has to ask the right questions, decide whether anyone is at risk and work out what should happen next.
If the issue cannot be resolved remotely, the same person must find an appropriate vendor, pass on the context, organise access and keep the landlord informed. Every missing detail creates another call or message before the repair can move forward.
The opportunity was to make this one continuous process, while keeping safety decisions controlled and the property manager visible throughout.
How the agent diagnoses a maintenance issue
The agent identifies the issue category, asks the diagnostic questions relevant to that problem and assesses urgency as the conversation develops. It collects the details, evidence and access information a vendor will need later.
When a safe remote resolution is possible, the tenant receives controlled steps they can follow. When it is not, the agent gives appropriate containment guidance so the tenant and property remain safe until professional help arrives.
Emergency signals take the conversation onto a stricter path. The agent stops routine troubleshooting, gives clear safety guidance and escalates the case instead of improvising.
Engineering judgement into the agent
Property maintenance covers many problem types, and each one needs a different diagnostic path. A heating failure, water leak and electrical concern cannot share the same questions, stopping conditions or tenant guidance.
We designed structured instructions for each issue category. They define what the agent must establish, which questions come next, how priority is assessed and when remote resolution is allowed. They also define what the agent must never suggest.
The system keeps diagnosis, priority, safety guidance and the final ticket connected. Property managers can see how the case was assessed rather than receiving an unexplained AI conclusion.
Issue-specific diagnostic paths
The next question depends on the problem type, the tenant's previous answer and the risk signals already identified.
Explicit safety boundaries
Remote steps are only available inside approved conditions. Unsafe actions, uncertain cases and emergency signals trigger a stop or escalation path.
Explainable priority assessment
The resulting ticket carries the observations and answers behind its priority, giving the property manager useful context for review.
From diagnosis to an accepted visit
When professional help is required, the system matches the ticket with several suitable vendors. The landlord or property manager controls the order in which approved vendors should be approached from their dashboard.
The system contacts each vendor and continues through the list until someone accepts the work. Follow-up timing is driven by ticket priority. Emergency cases can be retried every hour, high-priority cases every two hours, while routine work respects weekday and working-hour rules.
Once a visit is agreed, the appointment and status remain connected to the original diagnosis. The landlord can follow the case in the dashboard without reconstructing progress from separate messages.
Priority rules
Landlord-defined vendor order
Rolling out with the first property teams
The product is now being rolled out with its first group of landlords and property managers. The initial deployment brings tenant reporting, diagnosis, safety guidance, vendor coordination and visit tracking into one operating process.
The first rollout is being used to refine diagnostic paths, priority rules and vendor workflows with property teams before wider deployment.