Places I Would Not Let AI Make the Call in Property Management

An AI can pull an applicant's file together: income documents, references, screening report, the full packet in one place instead of six emails. What it should never do is decide whether that applicant gets the unit.

The reason isn't that AI is bad at pattern matching. It's that a wrong pattern here is a fair housing complaint, and the kind of pattern an algorithm finds useful (this applicant looks like applicants we've approved before) is close to the kind of pattern that gets a company sued. Every approval decision needs a person applying your written criteria the same way to every applicant, and able to explain exactly why in plain language if asked. AI can check that the file is complete. It doesn't get a vote on the outcome.

If a resident calls to push back on something, an AI line shouldn't be the one settling it. A disputed late fee, a "this has happened five times already" complaint, a request for an exception to a policy: all of that goes to a person, every time.

This is less about trusting the AI's judgment and more about what a dispute actually needs. A resident who's upset wants to be heard by someone who can actually change the outcome, not told the same thing twice by a system that can't. An AI handling the intake and giving the resident a clear next step (your property manager will call you back by end of day) is useful. An AI trying to resolve the dispute itself just adds a step before a person has to get involved anyway.

I've built lease question answering that quotes the exact clause and page when a resident or a team member asks something. That's useful because it's fast and it's checkable: you can see exactly which clause the answer came from.

Where I stop it is the moment the question stops being "what does the lease say" and becomes "can we make an exception." A tool that answers from the lease should say what the lease says, including "this lease doesn't address that, here's who to ask." It shouldn't improvise an interpretation, and it definitely shouldn't tell a resident what the policy is when the honest answer is that a person needs to decide. The clause citation is the whole point. The moment it starts answering without one, it's guessing.

AI can go through every tenant file and tell you which ones are missing a document, which income certifications are expired, and which files would raise a question if an auditor pulled them tomorrow. That's hours of checking most teams don't have time to do until the audit notice shows up.

What it shouldn't do is tell you a file is ready. A checklist flags what's missing. A person still has to look at the file, confirm the flagged items are actually fixed and not just reformatted to look fixed, and sign off that it's ready to show. The checklist catches the obvious gaps faster than a person scanning 400 files by hand would. It doesn't replace the person who's accountable if something's wrong.

This one isn't specific to one part of the business. Phone agents, chatbots, document tools: all of them hit calls or questions that don't fit the pattern they were built for. The caller mumbles the unit number. The lease question is really two questions. The inspection photo is blurry enough that the AI isn't sure what it's looking at.

The right move in every one of those cases is the same: hand it to a person instead of guessing. A system that always picks an answer, even a low confidence one, looks more capable in a demo and causes more damage in production. I'd rather have an agent that says "I'm not sure, let me get someone" ten times a day than one that guesses right nine times and wrong once, because the one wrong guess is usually the call that mattered.

Across all of this, the pattern is the same:

• Approvals and screening decisions. AI organizes the file. A person decides, against written criteria, every time.
• Disputes and complaints. Anything a resident pushes back on goes to a person, not back into the system.
• Lease interpretation beyond the text. The tool quotes the clause. It doesn't improvise what the clause means for a specific exception.
• Audit readiness. The checklist flags gaps. A person confirms they're fixed and signs off.
• Low confidence anything. When the AI isn't sure, it says so and hands off instead of guessing.

None of this makes the AI less useful. It makes it useful at the parts of the job that are genuinely repeatable, and out of the way for the parts that need a person's judgment and a person's accountability.

You don't need a vendor to tell you where your line is. Most teams already know, they've just never written it down in one place.

Step 1: Get your operations lead, your compliance person, and whoever handles resident complaints in one room for an hour.

Step 2: Walk through each tool you're considering (phone intake, inspection drafts, file checks, lease Q&A, chat) and ask one question for each: what's the worst version of this tool getting it wrong, and who would have to clean it up?

Step 3: For anything where the worst case touches an approval, a dispute, a lease interpretation, or a compliance sign off, write down that a person decides, full stop. Everything else is fair game for AI to draft, log, or flag.

Step 4: Ask any vendor, including me, to show you exactly where their tool stops and a person takes over. If they can't point to it specifically, that's the answer.

That list is worth more than any demo. It's the thing that tells you whether a tool is built to help your team decide faster, or built to decide for them.