58% of companies use AI. 8% have automated a process.

A tenant calls about a garbage disposal that stopped. Software picks up, asks what happened, walks her through the reset button underneath, and closes the request. Nobody drives anywhere. The call runs about two minutes.

That is running in this industry today, on real portfolios. It also isn't what most companies mean when they say they use AI.

Adoption here tripled in a single year. Completion did not move. Both of those come from the same survey, and the space between them is the most useful thing anybody has published about this technology in property management.

Adoption went from 20% to 58%. Full automation sits at 8%.

Buildium runs an annual survey of the industry. Its 2026 report states the finding in one line: AI adoption jumped from 20% in 2024 to 58% in 2025, though only 8% of companies have been able to fully automate any processes.

Fifty points of daylight. Almost everybody has switched something on, and almost nobody has taken a process off a person's desk and left it off.

One caution about that 58%, which we raise because it works against the point we are making. Buildium surveys people already on property software mailing lists, so the figure describes a self-selected group rather than the industry. The US Census Bureau runs a nationally representative survey of business AI use and puts all US businesses at 17% to 20% between December 2025 and May 2026, with firms of four or fewer employees under 20%. The truth for a ten person management company is probably nearer the government number than the vendor one.

What they use it for is drafting, listings and a chat box

The earlier wave of the same survey asked what people were using it for. Buildium's 2025 report, drawn from 1,796 property management professionals surveyed in June 2024, lists three answers:

• Generating messaging they can use as a starting point for customer communications
• Writing property descriptions for rental listings
• Answering incoming questions on their website with chatbots

Two of those three produce a draft that a person reads before anybody outside the company sees it. That is why they went first. They are the uses where being wrong costs thirty seconds.

It also explains the 8%. Nothing on that list touches money, screening, or a decision a resident is waiting on. The work still ends at a human, which means no process has actually left the building.

A machine answers the maintenance phone at some companies now

The advanced end is further along than that list suggests, and it has landed on maintenance intake rather than anywhere glamorous.

Property Meld sells MAX On-Call, which answers resident calls around the clock, works out what is wrong, and in their words reports only true emergencies to your team. Their 2025 annual benchmarking report puts the average maintenance call through that product at 132 seconds, scoring 4.55 out of 5 with residents.

Latchel runs the same idea and publishes harder numbers: 23% of requests solved at intake, dispatches cut by around 20%, and emergencies de-escalated. Read the page closely and the emergency figure appears as 48% in one place and 50% a few lines below it, and other pages on the same site put the work order reduction at 30%. No method, period or sample is given for any of them.

The best-sourced version of this claim is smaller and comes from a public company. On AppFolio's July 2026 earnings call, chief executive Shane Trigg said their Maintenance Performer product was deflecting more than a quarter of work orders through AI-led self-help. He was describing one named customer and 2,500 work orders logged since December 2025, of which 49% arrived after hours. That is a single portfolio, not a platform-wide rate, and anybody quoting it as one is misreading a transcript.

Take all of it as the vendor's own scoreboard rather than a finding. It still tells you what the technology is being pointed at, and it isn't the repair. Nobody claims the software fixes the disposal. The claim is that it works out whether anybody needs to drive over, and that a fair share of the time the answer is no.

The leasing numbers are the biggest and the least checkable

On the leasing side the published figures get much larger and much softer.

Landmark Properties manages more than 72,000 student beds. EliseAI's customer story about them reports 8,338 new leases in 2024, of which 3,182 came from after-hours leads, against a 15.15% lead-to-lease rate. More than 108,000 calls were answered across a phased rollout to 90 communities.

Read what is missing. There is no before-and-after, no comparison community running without the software, and no time period attached to most of the figures. The 8,338 leases are leases the company signed in a year when it also happened to be using this software. Whether it signed more of them because of the software is a different question, and the page doesn't answer it.

That is not an accusation. It is what almost every vendor case study in every industry looks like, and knowing the shape of it is most of the defense.

Repairs got slower in 2025, in the same data set

Here is the part that should give everyone pause, and it sits in the same report as the good news above.

Property Meld's benchmark for speed of repair, measured from resident submission to invoicing, went from 6.32 days in 2024 to 6.81 days in 2025. That is 7.46% slower. The share of repairs finished inside seven days fell from 44.3% to 39.4%. Their own commentary points at July, at 7.3 days, and September, at 7.9 days.

So the year AI intake spread across this industry is a year in which the headline maintenance number moved backwards. The report offers a reason. Work order volume rose, which is fair, and it is not proof of anything either way.

One oddity in that report we cannot resolve, and would rather flag than smooth over. Every individual work category in it either improved or held flat between 2024 and 2025. HVAC went from 5.7 days to 3.2, electrical from 5.8 to 4.1, appliances from 8.3 to 6.0, plumbing stayed at 4.7. Yet the overall average got worse. The report never reconciles those two facts, and we cannot either.

Intake got faster and the repair did not. Answering the phone in two minutes does nothing about the four days waiting on a part.

Which points at the real limit, and at the 8% again. These tools compress the paperwork around a repair. They do not compress the repair. If your bottleneck is a vendor who won't call you back, faster intake gets you to that vendor sooner and then stops.

Three things you can check in your own system this week

Every claim above can be tested against your own numbers, and you don't need to buy anything first.

1. How many work orders last month closed without anybody visiting? That is Latchel's 23% claim, measured on you. Most systems will filter for it. If you are already near 20%, somebody on your team is doing that work by phone today, and you now know what it is worth.
2. What is your own speed of repair, submission to invoice? Compare it to 6.81 days, and to the 39.4% finished inside a week. This is the one maintenance benchmark with a large data set behind it.
3. How many after-hours calls last month turned out to be emergencies? Nathan Gesner, who manages almost 350 rentals in Wyoming, wrote on BiggerPockets that he gets fewer than four a month. If your number looks like his, after-hours coverage is not the thing to buy first.

What is genuinely different from the last round of software

The last fifteen years of property management software were about storing things. The record of the work order got better. The work order didn't.

What changed is that a machine can now read a sentence a tenant typed at 11pm, work out that a dishwasher leaking onto a hardwood floor is more urgent than a dishwasher that is noisy, and act on the difference. Messy human input used to be the thing software could not accept. Now it is an input like any other.

That is a narrower change than the marketing suggests and a bigger one than the skepticism allows. It moves the boundary of what can be handed over, and it moves it in one direction only: toward the chasing, the logging and the first reply. Not toward the judgment, and not toward the wrench.

Intake is solved and the middle is not

The first thirty seconds of a maintenance request are now handled better by software than by a voicemail box. That is true rather than promised. The last mile still needs a person with a van.

Everything interesting is in between, which is where the 8% lives, and where this industry's own numbers say the days are still being lost.