AI agents for property management operations
Property management agents automate the highest-volume lettings work: triaging inbound tenant enquiries, dispatching maintenance to the right contractor, chasing rent arrears, and preparing renewals. They read from your PMS and act across email and SMS. Tenant screening and any adverse action must stay with a human for fair-housing compliance.
The volume problem in lettings
A property manager running a few hundred units fields an enormous quantity of small, structurally similar messages: is this still available, when can I view, the boiler is making a noise, can I pay on the 5th instead. Each takes two minutes and none of them can wait, which is why lettings teams spend their day in an inbox rather than on the work that actually needs judgement.
Speed also decides outcomes here in a way it does not in most industries. In a tight rental market the first agent to respond gets the viewing. We saw this directly on a rental-alerts platform where cutting notification delivery from roughly three minutes to five seconds changed who got to the property first — the same dynamic applies to enquiry response.
Which property management workflows to automate first
Ordered by the ratio of volume to consequence. The top rows are where to start.
| Workflow | What the agent does | Human approval needed? |
|---|---|---|
| Inbound enquiry triage | Classifies the enquiry, checks live availability in the PMS, answers or books a viewing slot | No — reversible and low stakes |
| Maintenance dispatch | Reads the tenant's description, assigns a category and urgency, matches to a contractor with the right trade and current availability | Not for routine jobs; yes above a spend threshold |
| Arrears chasing | Detects a missed payment, sends the escalating sequence, logs every contact attempt | No for reminders; yes before any formal notice |
| Renewal preparation | Assembles the renewal pack, proposes a rent based on comparables, drafts the letter | Yes — a human sets the rent |
| Compliance certificate tracking | Watches expiry dates for gas, electrical and fire safety, books the re-inspection | No — booking is reversible |
| Tenant screening | — | Do not automate. See below. |
The fair-housing constraint
Tenant screening is the workflow most people ask about first and it is the one to leave alone. In the US, an automated decision that produces a disparate impact on a protected class is a Fair Housing Act problem regardless of intent, and an adverse action taken on an algorithmic recommendation triggers FCRA notice obligations. The UK and EU have parallel exposure under equality law and the GDPR's rules on automated decision-making.
The defensible design is an agent that gathers and organises application material and a human who decides. Encode that as a hard system boundary, not a policy note — and keep the log, because the log is what answers the complaint.
What a maintenance dispatch agent actually does
A concrete trace, because 'automates maintenance' means nothing on its own.
Reads the report
Tenant texts 'water coming from under the sink, kitchen'. The agent extracts trade (plumbing), urgency (active leak, so same-day), and unit from the sender's number.
Checks the property record
Pulls the unit's history from the PMS. Notes a plumbing callout eight weeks ago on the same fixture — this is a repeat, which changes both the contractor choice and the warranty position.
Selects a contractor
Filters the approved list by trade, current capacity and distance, preferring the firm that did the original job given the repeat.
Books and notifies
Offers the tenant two slots, confirms, raises the work order, and notifies the contractor with the history attached.
Escalates the exception
Because it is a repeat visit within the warranty window, it flags the job to a human to decide whether the contractor is recharged.
Connecting to your PMS
The agent is only as good as its access to the property management system. Yardi, AppFolio, Buildium, Propertyware and Re-Leased all expose APIs of varying generosity, and the practical constraint is usually write access rather than read: several will let you query a tenancy far more readily than update one.
Where write access is restricted, the workable pattern is an agent that prepares the action completely and a thin human step that commits it. That is not a failure mode — it is often the right governance boundary anyway, and it means the integration limit and the control boundary happen to coincide.