I have been asked this question, in some form, in basically every customer call for the last 18 months. "What's your AI strategy." Honest answer first, then the longer version.
Honest answer: we use AI in three narrow places where it works and we ignore it everywhere else.
Where it works for us
One. Summarizing long resident emails into a one line gist that a property manager can read in three seconds. This is genuinely useful. Resident emails are sometimes 800 words of context for a request that boils down to "the dishwasher is broken." The summary saves time.
Two. Drafting first response replies that a property manager can edit and send. Saves about 4 minutes per response. Across a 300 unit property with maybe 20 substantive resident emails a day, that's an hour of saved time. Worth it.
Three. Categorizing maintenance requests so they route to the right person automatically. This used to be a human task. It is now a model task, and the model is better at it than the human was, because the human gets bored after the 40th request of the day.
That's the whole list.
Where it doesn't work for us
The thing every PropTech demo right now wants to do is "AI leasing assistant." A bot that handles tour scheduling and prospect questions and qualifies leads. I have seen probably 20 versions of this and I think the technology is genuinely not ready, and I want to explain why.
The problem is not that the AI can't have the conversation. The AI can have the conversation. The problem is what happens when the conversation goes sideways. A prospect asks a question the AI doesn't have data for. The AI hallucinates an answer. The prospect, eight weeks later, tries to move in to an apartment that has a pet policy the AI got wrong. Now you have a legal exposure and an angry resident and a property manager who is mopping up.
The expected value math, in my view, doesn't work. The AI saves you maybe $40K a year on a leasing role. One bad hallucinated answer that results in a Fair Housing complaint costs you, conservatively, $200K and reputation damage. You need a really low error rate before the math turns positive.
I think the technology gets there. I just don't think it's there yet for production. I might be wrong.
The thing I keep being wrong about
I have predicted, on at least three occasions in the last two years, that "this would be the year" some AI thing changed multifamily operations. I was wrong every time. The changes have been smaller and more incremental than the hype cycle suggests, and they have happened in places I didn't predict (categorization, summarization) rather than the flashy ones (leasing, predictive maintenance).
I think the actual pattern is that AI gets adopted in multifamily where it does one narrow task and gets out of the way, and gets rejected where it tries to be a "platform" or an "experience" or any of the other words PropTech marketing decks like to use.
If I had to predict where the next useful AI feature lands, I'd guess it's something boring. Renewal lease summarization in plain English for residents. Or automated translation for non English speaking residents at a property. Both of those are real problems that nobody is currently solving well. Both of them are unsexy.
I will probably be wrong about this prediction too. Check back in a year.