AI vs. the Call Center: Did an AI Leasing Agent Lease More Apartments?
Analyzing a pilot of an AI leasing agent that schedules tours and answers prospects' questions, to see whether it actually filled more apartments, not just more inboxes.
The question
Should a multifamily operator replace its leasing call center with an AI leasing agent?
My first project at GID/Windsor was to analyze a pilot of an AI leasing agent from a company unnamed for confidential reasons. The AI leasing agent becomes the “bot” a prospective resident talks to: it schedules tours and answers questions before the tour, at any hour.
Everyone expected more activity, and activity did go up. That wasn’t the real question, though. The real question was: is occupancy higher than it would have been without it?
How I analyzed it
I built the analysis in Power BI, comparing occupancy at each pilot property against similar properties that didn’t have the AI leasing agent. That gave a fair baseline for what “would have happened anyway.”
What I found
- Guest cards went up a lot. A guest card is created when someone first engages with our systems, so more engagement means more guest cards.
- The conversion rate looked worse. With so many more guest cards, the denominator grew faster than the leases did.
- More leases were signed. The total number of leases at pilot properties went up.
So was the lower conversion rate a bad thing? No. It was a side effect of casting a much wider net.
The Moneyball lens
I thought about it like Moneyball: get more people on first base, and the runs follow.
- Guest cards are first base. Every prospect who engages with us is a chance to lease.
- Leases signed are the runs. That’s the outcome that actually matters.
When an AI leasing agent puts far more people on base, the percentage who score can dip while the total runs go up. If you only watch the conversion rate, you could talk yourself out of a tool that’s working.
What it means
When you evaluate AI in leasing, or anywhere, judge it on the outcome: leases signed and occupancy compared with similar properties. Funnel ratios are useful, but when AI changes the size of the funnel, they can point you in the wrong direction.