AI Leasing Handoff: Lessons From 1,000+ Mystery Shops

A prospective renter asks to speak with a person. The AI leasing assistant says it will pass the request along, then keeps sending tour times. In another exchange, a leasing professional steps in, but the automated messages continue alongside theirs. Either way, the prospect is left guessing who, if anyone, is handling the conversation.

Lisa Trosien, president of ApartmentExpert.com, has seen both outcomes repeatedly. After more than 1,000 mystery shops of apartment communities, she says the AI leasing handoff often fails in one of three ways: no one picks up, no one follows through, or the AI never stops. Multi-Housing News asked her what operators should check to make an AI leasing handoff work—so that when a renter needs a person, a leasing professional actually steps in.

Should the task or the prospect’s cues determine when a leasing professional steps in?

Trosien: I’d put more weight on what the prospect is saying. A task might look routine until something in the conversation changes.

The hallmark of just about every AI response is the request for the tour. That’s fine. Getting the tour is the AI’s job. But it’s almost always the same offer: two times, usually on the same day. Then the prospect writes back, “I’m out of town. That doesn’t work for me.” And the AI responds with something like, “How about tomorrow at 11 or 3?” It isn’t reading what the prospect is actually saying. When that happens, a leasing professional needs to step in.

When does AI add capacity, and what suggests it isn’t listening to the renter?

Trosien: To me, added capacity means the team has more time to help people, and the renter can still get the help they need. If AI handles an inquiry well, that’s useful. If the inquiry needs more attention, there should be a workable way to get it.

The warning sign I see consistently is the AI not reading what the renter is saying. I typically ask for a move-in date more than 30 days out, because I want to see whether they can accommodate someone planning ahead. It has gotten better, but about half the time they still try to push me into something available now, when I’ve told them I don’t want something now. I’ve asked for a December move-in and received nothing but November information.

It happens with pets, too. I mentioned my cat, and then I received an email with the subject line “Your dog will have room to roam.” I don’t have a dog. That’s clearly a standardized message that went out because the AI recognized a pet and sent the pet email.

I shopped six properties with one management company in the D.C. area and asked each of them for one-bedroom information. Every single property sent me two-bedroom information. Assuming they had a setup problem, I tried the same thing in their Denver market. Again, I asked for one size and received information on the wrong size.

These are setup issues, and operators need to make sure their setup is right. Whether that’s an operator issue or an AI issue, I honestly don’t know, because I’ve never done a setup myself. Either way, those exchanges deserve attention, no matter how quickly the responses went out.

What signals show that AI has stalled, and can it recognize them?

Trosien: If someone repeats a question or corrects you, they’re telling you the answer didn’t help. If they ask for a person, it doesn’t get any clearer than that. I’ve asked for a human over and over and gotten nothing but more AI.

I don’t know what every system can recognize. I only know what showed up in my inbox. So, I’d tell suppliers to prove it. Show me what your system does when someone asks for a person. Then show me what happens when they say it a different way, or when they’re frustrated and don’t say it at all.

How should operators disclose AI and make it easy to reach a person?

Trosien: I want people to know they’re communicating with AI. I’ve seen Aimee, Ivy, Rosie, Corie—and in one definitely different scenario, Frank—all these one-named AI leasing personas, signing their emails “Best,” followed by the name. I even had some emails signed “Meet Elise.” More are now identified as AI leasing professionals, and that’s an improvement.

That also connects to the move toward AI disclosure in state legislation. Colorado has enacted conversational AI disclosure requirements that take effect January 1, 2027. The language used in the conversation should support that transparency.

The AI should say, “I will hand this over to a leasing team member,” not “my colleague on the leasing team,” because they’re not colleagues. One’s an AI, one’s a person. Why do we have to make it sound like they’re all really people when they’re really not?

The option to reach a person should be easy to find. Once the prospect chooses it, I’d want the automated leasing exchange paused, with a confirmation explaining what happens next.

What does a successful handoff look like, and what role should AI play once a leasing professional takes over?

Trosien: A handoff is done when a person has actually responded to the prospect. Occasionally it works. I get turned over to a human, and the human solves the problem. But I’ve also seen it go wrong three ways. Sometimes I’m told I’m being turned over to a human, no human ever picks up, and the AI keeps responding. Sometimes I’m told I’m being turned over to a human, and then I never hear from anyone again. And sometimes I do get turned over to a human, but the AI keeps going anyway. The person is moving the conversation forward, and the AI is still asking me for an appointment. At most properties, it’s just not smooth.

The leasing professional should get the whole conversation and a quick rundown: what I asked, what I already told them, what was promised and what’s still open. And I should hear back that someone has it, what happens next, and when. Tell me who it is, too.

AI can still help behind the scenes. It can summarize the conversation or help draft a reply. But once a person takes over, the person decides what goes to the prospect.

I’ve had exchanges where the AI kept answering while the human was answering. That’s confusing. Somebody has to sit down and decide, across the chatbot, the CRM, email and text, what stops when a person steps in, what keeps going, like appointment reminders, and who’s allowed to turn it back on.

Where should operators look when prospects receive duplicate or conflicting messages?

Trosien: I’d start with the full contact history rather than assume the cause. Identify which system or person sent each message and what triggered it.

Sometimes when I send an inquiry, I’ll get three responses. One comes from the property management system that’s running the CRM. Another comes from whatever the management company has set up. Then the AI responds too. Three responses to one initial request is absolutely ridiculous, and it tells me nobody has looked at how these systems are connected.

If the problem happens after a handoff, check whether the automation paused and whether an employee accepted the request.

What should the system record to show who took responsibility and whether they followed through?

Trosien: I’d want to see who took the conversation, what they were supposed to handle, when they had to respond by, and whether they did. Assigning someone isn’t the same as contacting the prospect.

The dashboard can show all green because yes, the prospect got a response, and yes, they’re getting follow-ups. But that doesn’t tell you whether the response was worthwhile or whether the follow-ups were relevant. A response going out isn’t the same as the prospect getting help.

There also has to be a way to flag a request nobody answered and get it to someone else. Otherwise the staff thinks the AI has it covered, and the prospect is sitting there waiting for a person who isn’t coming.

How can operators tell when self-service worked, rather than a prospect giving up?

Trosien: Did the prospect get what they came for? Was the answer accurate? Did their question get resolved? If they booked a tour, did it get booked right?

When someone stops responding, you don’t know whether they got everything they needed or just gave up. A conversation count won’t tell you that. Neither will response speed nor how many were handled without a person. Put those numbers next to the unanswered questions and the people who had to ask twice.

What does mystery shopping across properties reveal that a dashboard might miss?

Trosien: Mystery shopping lets you see what a renter sees when they’re shopping several communities at once. For example, I’ve had four different properties respond with the exact same verbiage. They were four completely different properties in the same market, with the same management company, competing against each other. But the management company had set it up so every property said the exact same thing. That’s crazy. Someone reviewing one property’s messages would never catch it. The renter catches it right away, because the same words are sitting in their inbox from four different places.

It also brings the channels together. An email, a text and a call might each look reasonable on their own but create confusion when the prospect receives them together.

I’d test realistic situations: someone shopping ahead, asking multiple questions, correcting information, changing an appointment, or requesting human help. Follow those conversations across channels and beyond the initial response. That gives you a better view of what the renter has to do to get information and move forward.