The State of AI in Leasing: What It Really Saves Operators in Time and Money

Every multifamily vendor with a chat widget now claims to automate the rental funnel. Some of that is real. A lot of it is a rebrand. Before you spend a budget line on either, it helps to separate what the category has proven from what it is still promising.

This is the state of AI in leasing as of 2026: the real numbers behind after-hours leads, vacant days, and the time a leasing team gets back. We will also tell you where the category still has real gaps, because pretending otherwise would not help you or us.

What Is the State of AI in Leasing Right Now?

An AI leasing agent is software that talks to prospects across text, chat, email, and voice, then qualifies them, books tours, and follows up without a person typing every reply. That part of the definition has not changed much in three years. What changed is who is using it.

AI adoption among property managers went from 20% in 2024 to 58% in 2025, according to Buildium's 2026 Property Management Industry Report, produced with NARPM. That is most of the industry deciding, inside a single year, that it could no longer sit this one out.

The same report carries a second number the category tends to leave off the slide. Only 8% of companies have fully automated any single process. Adoption is a mile wide and an inch deep. A lot of operators bought a tool. Far fewer have finished turning it on.

The strongest read on that comes from outside the vendor world. The National Apartment Association reported on MRI Software's Multifamily Pulse Check 2026, a survey of more than 700 North American real estate professionals, which found only 7% of operators use no AI at all. A trade association reporting a 700-person sample is about as solid as this category's numbers get.

EliseAI's 2025 survey of 280 multifamily executives, director level and above, found 78% say they have already lost new business to AI-enabled competitors. That is a vendor-sponsored, self-reported figure, not independently audited, so treat it as a data point from a company selling the answer to the question it is asking. Still, the direction lines up with the adoption numbers above. The conversation has moved past "should we try this" to "who is still catching up."

How Much Money Do Operators Lose to Slow or After-Hours Follow-Up?

Operators lose real leases to slow follow-up because a lot of leasing interest shows up after the office lights go off. ShowMojo reported that 61% of the inquiries reaching its own AI agent in the first quarter of 2026 arrived outside business hours, and more than 60% again in the second quarter. That is traffic to one vendor's agent, not a count of the whole market, so take it as a signal about when renters shop.

Renters have also moved the goalposts on what counts as a reply. EliseAI's 2026 survey of 350 multifamily decision-makers and 500 renters found 73% of renters expect a response by the end of the same business day, and more than 60% expect some degree of round-the-clock responsiveness. A vendor paid for that survey, though an outside research firm ran it, which is better than most numbers in this category get. It is still a company asking the market about a product it sells.

What the current data does show is that the operators actively running AI on this problem report real results. EliseAI's 2025 survey of 280 multifamily executives, director level and above, found that among operators with active deployments, 77% report moderate to significant reductions in operating expenses and 85% report measurable improvements in lead-to-lease conversion. Both figures are vendor-sponsored and self-reported, not independently audited, so read them as a company describing its own customers rather than a neutral study.

Put those two together and the gap is obvious. Renters shop at night and expect an answer the same day. Most leasing offices are dark for two thirds of the hours those leads arrive. The lead that comes in at 9pm on a Tuesday does not wait politely until Wednesday morning, because three other communities are a tap away.

What Does a Vacant Unit Actually Cost While a Lead Waits?

A vacant unit costs about $50 a day on a $1,500 rent, which works out to roughly $1,600 for a turn that runs the national average. That number climbs the longer your time-to-lease runs. Apartment List's National Rent Report, dated August 26, 2026, put the national vacancy index at 7.1% with a median rent of $1,390. Time to lease averaged 32 days in August 2026. In January 2026, the same measure hit a record 41 days, the slowest pace since Apartment List started tracking in 2019.

The U.S. Census Bureau's Quarterly Residential Vacancies report for Q2 2026 put national rental vacancy at 7.3%, not statistically different from Q2 2025's 7.0%. The two count different things, so treat them as two readings pointing the same way, not as one number arguing with itself.

Run the simple math on a $1,500 unit: $50 a day. At today's 32-day pace, a slow-to-lease unit costs roughly $1,600 more than it should. At January's 41-day record, that climbs to about $2,050. Multiply that by however many turns your portfolio runs in a year and the number gets real fast, which is exactly why we like reporting it per turn instead of guessing at your portfolio size for you.

A lot of leasing AI gets built for the demo, not the funnel. Watching a chatbot answer three sample questions well tells you nothing about what happens when a real lead goes quiet for four days because nobody followed up. The vacancy math above is why that gap matters more than it looks like on a sales call.

How Much Time Does a Leasing Team Waste That AI Could Save?

A leasing team can save real hours a week with the right tools, but treat every vendor's headline number as a claim until your own pilot proves it on your own leads. Vendors publish hours-saved numbers freely, and most of them measure a whole platform rather than leasing alone.

The counterweight is the cleanest receipt we have found for "pilot before you believe it." Section's AI Proficiency Report, published January 2026 from a survey of 5,000 knowledge workers at organizations with 1,000-plus employees across the US, UK, and Canada, found 68% of AI-using employees save 4 hours or less per week. The sharper number sits right next to it: only 2% of executives report no time savings from AI, compared to 40% of the workers doing the work. The people reporting the savings are often not the people doing the work. That gap alone is a reason to ask your own team what changed before you ask your dashboard.

None of this is real estate-specific, and it should not be. Time savings from AI tools track similar patterns across industries, which is exactly why a category built entirely around one hours-saved number deserves a second look before you buy on it alone.

Where Does AI in Leasing Still Miss the Mark?

AI in leasing still misses the mark on fair housing compliance and the human handoff, not on how well it holds a conversation. That is a real shift from a few years ago, when the usual complaint about these tools was that they sounded robotic. Most of the category can hold a decent conversation now. The harder problem is what happens when a conversation touches a protected class, a reasonable accommodation, or a voucher program.

The Fair Housing Act itself has not changed and still applies to anything that communicates with a prospect on a housing provider's behalf, automated systems included. What has changed is the guidance around it. A September 16, 2025 memo from the U.S. Department of Housing and Urban Development (HUD) prioritizes intentional discrimination cases over disparate impact ones. Two 2026 Federal Register notices withdrew further guidance: one in April pulled eight HUD Fair Housing and Equal Opportunity documents, including the 2024 digital advertising guidance and both assistance-animal guidance documents, and one in July withdrew thirteen Office of General Counsel documents. The disparate impact rules at 24 CFR 100.500 are proposed for removal. They have not been removed. A comment period on that runs through October 9, 2026.

That rollback matters less than it looks like it should, and the complaint data shows why. The National Fair Housing Alliance's 2025 report, covering complaints filed in 2024, counted 32,321 fair housing complaints nationwide, 54.6% of them disability-related. Of those, 74.12% were processed by private nonprofit fair housing organizations, 20.90% by state agencies, 4.85% by HUD, and 0.14% by the Department of Justice. A softer federal stance does not shrink your exposure, because the people most likely to test your leasing bot were never federal employees in the first place. A fair housing nonprofit can run a test inquiry against a live chatbot remotely, at scale, on any given afternoon.

Two real cases show what that looks like when it goes wrong. In Louis et al. v. SafeRent Solutions, a Massachusetts court gave final approval to a $2.275 million settlement in November 2024. The plaintiffs were two Black women holding housing vouchers, and the case involved a scoring model used to approve or decline applicants, not a leasing chatbot. SafeRent agreed not to issue approve or decline recommendations for voucher applicants based on its score unless the model is validated for fairness by civil rights experts, and the settlement resolved without an admission of liability.

Separately, the fair housing nonprofit Open Communities ran months of testing and found a leasing chatbot issuing a blanket "we do not accept Housing Choice Vouchers" reply across more than 100 properties at once. Its case against Harbor Group International resolved on January 31, 2024 with a source-of-income commitment, two years of monitoring, and nationwide fair housing training. The lesson there is not about any one vendor. It is that an automated reply can scale a single policy error across an entire portfolio instantly, in a way a tired leasing agent repeating the same wrong answer never could.

There is a real counter-argument worth hearing, and it makes this section more honest, not less. Several operators argue AI makes fair housing safer, because it answers the same way every time and a person might not. Tim Kramer of Draper & Kramer put it best on a recorded industry panel: "The good news is you've got consistency in answers. The bad news is you've got consistency in answers, and if it's wrong, it's always going to be wrong." Both halves of that are true at once, which is what makes this hard to govern.

Operators surveyed by Insights by Blueprint's Advisory Council in March 2026 describe the human handoff, not the conversation itself, as one of the main challenges they run into with AI leasing tools. Put those two findings together and the picture gets clear: the risk in this category is not that the bot sounds bad. It is what happens when the bot is confidently, consistently wrong, and nobody catches it until a test inquiry does.

That compliance risk looks like a leasing-AI problem on paper. In practice, for anyone running income-restricted units, it is an affordable housing problem, and it deserves a purpose-built answer rather than a feature bolted onto a general leasing tool.

How Is Fortress Solving the Compliance Gap? (Cora, the Compliance Agent)

Fortress is solving the compliance gap by building a second, dedicated agent for it instead of treating compliance as one more feature on a leasing bot. We're calling it Cora, the Compliance Agent, and it is in active development now, aimed squarely at compliance teams in affordable housing.

An AI leasing agent that does not understand affordable housing compliance is exactly the tool that stumbles on an income-limit edge case at 2am with nobody watching. Fortress has been affordable-first since day one, so compliance was never going to be a second product bolted onto a leasing tool after the fact. We showed an early look at Cora at the most recent MultiFamily Insiders Demo Day 2026, and we're continuing to build it out with real compliance teams.

What Should Operators Actually Expect From a Proper AI Leasing Agent?

Operators should expect an AI leasing agent to move a prospect forward without creating cleanup somewhere else in the system. That sounds obvious. Judging by what operators report running into, it is not always what they get.

That table comes from what the Insights by Blueprint survey found operators describing, not a scorecard on any one company. Every row is a workflow problem before it is a technology problem. From what we've seen, the tools that help are the ones that live inside the same system that already holds the lease, the application, and the resident record, so a handoff does not mean opening a second screen.

What Metric Proves an AI Leasing Agent Is Working?

The metric that proves an AI leasing agent is working is Average Days to Lease, tracked before and after, on your own portfolio. Response time and vendor hours-saved figures are easier to report and tell you far less. What counts is how fast a prospect becomes a signed resident.

Apartment List's national average sat at 32 days as of August 2026, up to 41 in January. That is your benchmark, not your target. Your own portfolio's baseline is the number that matters. Pull a full season of that baseline before you switch anything on, covering the slow months as well as the busy ones, or you will not be able to prove the rollout changed anything either way.

Elmington, a Nashville-based operator managing more than 35,000 units, is the clearest proof we have of what a connected system does to that number. Before centralizing on Fortress OS and AffordaPortal, applicant response times ran around 2 hours. After, that dropped to 12 minutes. Approval times went from about 2 weeks to 3 days, sometimes under 24 hours, and file errors dropped 75%. None of that is a Leslie statistic. It is what happened when leasing, documents, and compliance stopped living in separate systems, which is the same principle an AI leasing agent has to be built on to move the needle at all.

One honest caution here: the metric can move the wrong way while the system is working. Lease-up teams run into this when occupancy climbs: fewer units are available, so a smaller share of leads can convert at all, and the lead-to-lease ratio drops while the operation is running better than before. Tour-to-lease holds up better as a measure when that happens. If you judge a rollout by one number without asking what else moved at the same time, you can talk yourself out of a tool that is working.

How Should Operators Compare AI Leasing Agents Without Getting Misled?

Operators should compare AI leasing agents on what each company publishes about itself, checked on the same date, and then confirm the number that matters on their own data before signing anything. What follows is a directory, not a scorecard: figures are vendor claims or our own reading of public information as of this post's date, not audited or guaranteed. Company names and trademarks belong to their owners, and nothing here is an endorsement or criticism of any of them.

A 2026 buyer's guide from Layer3 Labs, an independent firm, suggests two different thresholds depending on what you need. Buy an off-the-shelf tool if you run under 3,000 units on a standard property management system and want to be live in 30 days. Custom builds start making more sense once a portfolio passes 5,000 units, or runs a system that does not fit the standard mold. Those are two separate numbers with two separate triggers, so do not treat "under 3,000 units" as also telling you when to build instead of buy.

Where Fortress Is Taking This (Leslie, and What Comes Next)

Fortress is taking this by building AI leasing directly into the system of record instead of adding it as a layer on top. That is the whole idea behind Leslie: built in, not bolted on. And it does not stop at leasing.

Leslie is built, and it is in beta with real operators right now. The after-hours gap covered earlier in this post, where most of the interest lands while the office is closed and renters still expect an answer that day, is the category-level problem Leslie is built to close. Leslie is aimed at keeping a prospect from going quiet after the first touch, not at replacing the person who tours them, negotiates the lease, and hands them the keys. We are not publishing performance numbers for Leslie yet, on purpose. It is early, it is in beta, and we would rather earn that number than print one before it is real.

Leasing and compliance sit inside Fortress Core at Fortress, which is the whole point of building Leslie and Cora together instead of shipping one and hoping the other gets solved later. If your busiest lease-up season is coming and you want to see where Leslie stands today, that is a better conversation to have live than in a blog post.

Your prospects work all day, same as you. The only time they get to sit down and hunt for an apartment is after they clock off. Same as you. Leslie is built to keep those prospects on the line at 9pm on a Tuesday, so more of them get moved in and your occupancy goes up.

AI will not save a leasing team that already answers every lead in twelve minutes flat and never lets a follow-up slip past day three. If that is your team, congratulations, you do not need this post nearly as much as you think. For everyone else still letting leads go cold after the first try, the math above is not abstract. It is next month's vacancy report with your name on it.