A tenant reports a dripping tap through the portal. Another emails a photo with no unit number. Someone calls about an issue that may already have an open work order. Before your property-management team can arrange the work, someone has to make sense of all three.
AI can help organize those messages, suggest a category and highlight missing information. A person should still review the original request, decide its priority and authorize the next step. I would start with a small internal review queue that makes your existing process easier to follow.
The useful question is where your team keeps repeating the same sorting and copying, and whether a draft summary would actually reduce that work. Our earlier call-to-job-brief guide covers initial intake for trade businesses; here the focus is managing requests across properties and keeping exceptions visible.
Start with the requests you already receive
Before adding AI, choose one place where the team can see every request and its status. Agree who watches that queue during office hours, who covers an absence and what happens outside those hours. Keep your established emergency reporting route clearly visible to residents. It must continue to work independently of the AI process.
Your property software may already cover much of this. For example, Buildium describes maintenance tools that keep tasks, work orders, vendors and repair updates together. Check what your existing plan supports before building a second system that staff would have to maintain.
An ordinary form can require a property and unit number. A simple rule can send a known building’s requests to its assigned manager. Neither needs AI. The more useful AI task is reading varied wording and preparing a consistent draft for the person reviewing it.
Give AI a narrow job
For a first trial, I would ask it to prepare four things: a short summary, a suggested category, the details explicitly provided and a list of missing or conflicting details. Keep a link to the original message beside the result.
Use a small category list your team understands, such as plumbing, heating/cooling, electrical, access and other. Include “unclear” as a valid answer. A model should not have to guess a category to finish its work, and a category should never decide how long a resident waits.
Microsoft documents a Power Automate action that classifies text into categories using a trained, published model. It can return a category and a confidence score. That shows one possible building block, not that a ready-made property-maintenance workflow is included with your email subscription. A high confidence score also does not establish that a request is safe or routine.
A sample request and review worksheet
Here is a fictional example prepared for this article. It is a proposed process, not a customer result or a tested integration:
“The tap under the kitchen window is dripping again. I sent a photo yesterday. Please come after lunch.”
The draft below deliberately leaves gaps. It does not assume which apartment sent the message, whether the earlier photo reached the office or whether “after lunch” authorizes entry.
| Field | Suggested draft | What the reviewer checks |
|---|---|---|
| Original request | Link to the unedited message and any available attachment | Can we see what the resident actually said? |
| Property and unit | Not provided in the message | Match against an authorized resident record; do not guess. |
| Reported issue | Resident reports a recurring drip from the kitchen tap | Keep the wording factual; do not invent a diagnosis. |
| Suggested category | Plumbing | Confirm the category and assess priority using the established process. |
| Possible related request | Resident mentions an earlier photo | Find the earlier record; do not merge or close a work order automatically. |
| Access and next step | Requested time: “after lunch”; date and entry arrangements unclear | Confirm arrangements and approve any reply or work-order update. |
You can use those same six fields without AI. If the team cannot agree how to fill them in manually, that is the first problem to solve.
Keep exceptions visible to a person
AI can miss important wording or produce a neat summary that leaves out a detail. The original request must remain available, and staff need to review requests even when the suggested category looks ordinary. Do not make an AI label the only way a potentially urgent report reaches someone.
For the trial, leave resident replies, contractor dispatch, spending approval and work-order closure under human control. A missing unit number, conflicting address, unsupported language, unreadable attachment or possible duplicate should remain visible for review. If the AI step fails, retain the request in the normal queue and make that failure visible too.
Treat text inside a resident’s message or attachment as request data, not instructions that can change the workflow or grant access. Keep the AI connection’s permissions narrow enough that a mistaken output cannot send a message, delete a record or change an unrelated property’s information.
Connect the tools only after checking the fit
There are several possible ways to build the drafting step. Microsoft’s current Power Automate guidance describes adding a prompt to a flow and putting human approval after its output. It also notes regional and capacity limits. A prompt creates text; it does not by itself establish a supported connection to your property-management system.
Before choosing an approach, check the property software’s supported connections, the permissions it requires, the relevant licenses and usage costs, and how failed or repeated submissions are handled. One resident message should not accidentally create several work orders when a connection retries.
Begin with fictional examples. Before using real requests, agree which information the process needs, who can see it, where it is stored and how long it is retained. Entry codes, financial records and unrelated tenant information should not be copied into a general-purpose AI chat to make a summary more convenient.
Through our AI integration and automation work, IT MGMT can help assess the process, evaluate supported connections, set access boundaries, test the workflow and train the people who will use it. The design and ongoing support need to fit your actual software and team.
Try one queue and measure the whole job
I would begin with one office team and one request channel, keeping the normal process available throughout the trial. Include straightforward requests and awkward examples: missing information, two issues in one message, a repeat report and a failed attachment.
For each request, record how long the complete task takes, including checking the original and correcting the draft. Count important omissions, wrong categories, duplicate records and requests that still needed clarification. Ask the staff whether they trust the process enough to use it and what they keep having to fix.
Agree in advance what would stop the trial, such as a request disappearing from the queue or information reaching the wrong person. Assign someone to maintain the categories, permissions and connections afterward. A useful first result is reliable handling with less repetitive work; the trial has to demonstrate that, rather than assume it.
If you want to explore this for your office, bring one maintenance-request process and the names of the tools your team uses to an IT MGMT conversation. We can start by working out which steps need clearer rules, where AI might help and what should stay with your team.