# AI in property management: what it does, and what it must not _Published 2026-09-16 by Proppely. Canonical: https://www.proppely.com/ai-property-management_ ## Short answer AI in property management reliably does four jobs today: it answers questions about your own records, drafts communication, triages and routes maintenance, and pulls structured data out of documents like leases and invoices. It is much weaker at anything requiring judgement about a person — who to approve, what to charge, when to file — and those decisions carry legal consequences that fall on the landlord or manager, not the software. The useful question about any AI feature is not how smart it is. It is what it is allowed to do without asking: read only, write with confirmation, or act on its own. Money is the clearest line. No AI assistant should be able to move it unattended, and rent should never be paid inside a chat window. ## The four jobs AI does well in property management ### Answering questions about your own records Who owes what, which leases end this quarter, what a unit has cost in repairs this year, which work orders are stalled. This is retrieval over your data, not general knowledge, and it is where AI earns its keep first. - Working well: It cites the record. An answer you cannot click through to is an answer you have to verify by hand, which costs more time than it saved. - Where it stops: It is only as current as the data. An assistant reading a ledger nobody reconciles gives confident answers about numbers that are wrong. ### Drafting communication Renewal offers, late-rent notes, maintenance updates, replies to prospects, owner emails. Drafting is the lowest-risk, highest-volume use in the industry. - Working well: The draft is attributed and a person sends it. Volume matters less than whether you would have been happy to write it yourself. - Where it stops: Anything with legal force — a notice to pay or quit, a lease violation, an adverse action letter — is a document with statutory content and timing rules that vary by state. Draft it if you like; do not send it unread. ### Triaging and routing maintenance Reading what a resident wrote, working out urgency, deciding whether it is a plumber or a handyman, spotting that four requests in one building are the same underlying failure. - Working well: It routes to the right trade and escalates genuine emergencies fast, and it tells you when several requests are one problem, because that is the finding that saves real money. - Where it stops: Habitability and safety calls are yours. An assistant that quietly downgrades "no heat" in January is worse than no triage at all. ### Reading documents Pulling terms out of a lease PDF, line items off a vendor invoice, figures off a statement — turning paper into records you can actually query. - Working well: It shows you what it extracted next to the source before anything is saved, so a misread rent escalation is caught in the preview rather than in next year’s books. - Where it stops: Extraction accuracy falls off a cliff on scans, handwriting and addenda. Treat every import as a draft until a person confirms it. ## Every AI feature is one of three things - **Read only** — It can find and summarise, and cannot change anything. Safe to turn on everywhere; limited on its own. Ask: Does it see everything, or only what the signed-in person is allowed to see? - **Writes with confirmation** — It proposes a specific change and a person approves it before it happens. This is where most of the day-to-day value is. Ask: Does the confirmation show the exact change, or just a summary of intent? - **Acts on its own** — It completes work unattended. Appropriate for reversible, low-stakes actions, and a genuine risk for anything else. Ask: Which actions can it take unattended, who chose that list, and can the account owner shorten it? ## Rent is never paid through chat Residents pay in the resident portal, where card and bank details are typed into the payment provider’s own hosted form; that data is transmitted to the provider and never passes through, or is stored on, Proppely infrastructure. Card details are never entered in a chat window, and rent cannot be paid over WhatsApp, Telegram or any connected AI client. The one money-adjacent thing an assistant can do in Proppely is record a cash or check payment already collected in person. Anything that moves money becomes a proposal a person approves, and a vendor payout needs two different approvers. ## Decisions that stay with a person - **Approving or declining an applicant** — Tenant screening is regulated by the Fair Credit Reporting Act, and a decline triggers adverse-action duties. The reasoning has to be a person’s, documented, and consistent between applicants. - **Serving notices and starting an eviction** — Content, delivery and timing are set by state and often city law, and a defect can restart the whole process. This is a decision to make with counsel, not a draft to send. - **Keeping any part of a security deposit** — Deposit rules are statutory, deadline-driven and unforgiving of vague itemisation. The itemisation is a factual claim you have to be able to defend. - **Moving money** — Payouts, refunds and vendor payments are irreversible in practice. An approval step is not friction here; it is the control. Nothing here is legal advice; rules differ by state. ## Questions to ask any vendor about their AI - Can the assistant change records, or only answer questions? - What exactly happens before it moves money or sends a message to a resident? - Is the AI included in the price, or on a higher tier, or metered on top? - Can I use it from the AI tools I already pay for, or only inside your app? - Does it respect the permissions of the person using it, or run as an administrator? - Is every AI action logged in a way I can show an owner or an auditor? - What happens when it is unsure — does it guess, or does it say so? ## How Proppely answers them One assistant across the web app, the desktop app, WhatsApp and Telegram, working within the permissions of the person using it rather than as an administrator. In the web app and chat channels it confirms each change before running it; in the desktop app and MCP clients, routine changes run directly. Actions that move money, send an AI-written message to a resident, or cannot be undone always become proposals a person approves. An account owner can require approval for every change. Every action is logged. Proppely publishes a remote Model Context Protocol server with OAuth sign-in and documented setup for Claude, OpenAI Codex, Cursor and OpenClaw: https://www.proppely.com/mcp ## FAQ ### What can AI actually do for a landlord today? Four things dependably: answer questions about your own records, draft communication, triage and route maintenance, and extract data from documents such as leases and invoices. It is weakest at judgement calls about people, and those carry legal consequences that fall on you rather than on the software. ### Can an AI assistant collect rent? It should not, and in Proppely it cannot. Residents pay in the resident portal, where card and bank details go straight into the payment provider’s own hosted form. Card details are never entered in a chat window, and rent cannot be paid over WhatsApp, Telegram or any AI client. The one money-adjacent thing an assistant can do is record a cash or check payment you already collected in person. ### Is it safe to give an AI assistant access to my portfolio? It depends entirely on what the assistant is allowed to do without asking. Look for three things: it acts within the permissions of the person using it rather than as an administrator, actions that move money or send resident communication become proposals a person approves, and every action is logged. Ask for those specifically — they are the difference between a useful tool and an unsupervised one. ### Can I connect my property management software to Claude or another AI app? With some software, yes. Proppely publishes a remote Model Context Protocol (MCP) server with OAuth sign-in and documented setup for Claude, OpenAI Codex, Cursor and OpenClaw, so those apps can work with your Proppely data within your own permissions. Protected actions still become proposals you approve inside Proppely. AppFolio has announced a connector between its Realm-X assistant and Claude. ### What is MCP, and why does it matter for property management? The Model Context Protocol is an open standard for letting an AI application work with an outside system’s data and tools. It matters here because it decouples the assistant from the vendor: instead of being limited to whatever chat window your software ships, you can work with your portfolio from the AI app you already use, under your own account and permissions. ### Does AI in property management cost extra? It varies by vendor — included on some plans, reserved for higher tiers on others, metered on top elsewhere. In Proppely every workspace includes $5 of AI usage a month with a daily message limit, including free workspaces, and heavier use runs on prepaid credits that never expire. You can also connect an MCP client and work through the AI subscription you already have. ### Will AI replace property managers? Nothing in the current generation suggests it. What it removes is the volume work around the job — the lookups, the first drafts, the retyping of documents into records, the triage of a full inbox. The decisions that define the job, about people and money and risk, are exactly the ones you should keep. ## Related - Property management software with AI, compared: https://www.proppely.com/property-management-software-with-ai - Free workspace details: https://www.proppely.com/free-property-management-software - Published prices of other property management software: https://www.proppely.com/compare