Everybody bought AI. Almost nobody finished

Everybody bought AI. Almost nobody finished

Adoption tripled in a year. Full automation sits at 8%. That gap is the whole story.

Paul Cobb
August 31, 2026

A rehab wrapped and the listing agent needed copy. I knew everything that went into that house, so I asked a model for a first draft.

It gave the house quartz counters and new appliances. Trace those back to the scope and you find granite and one new range.

Nothing shipped. But that is the moment I stopped calling what I was doing a workflow. I was using AI. I had not built anything.

Buildium's annual survey put a number on that this year: among property managers, AI adoption went from 20% to 58% in a single year, while the share who had fully automated any workflow sat at 8%. Three in five are using it. One in twelve has finished anything with it.

Adoption is a purchase. Automation is a build. The first takes an afternoon and a credit card. The second takes writing down what "done right" means, which is the part almost nobody does.

Their report closes with three moves: simple wins, automate the revenue tasks, stay ahead on renewals. Here is the operator version.

1. Simple wins: the description you hand your agent.

Buildium found the most common uses were descriptions and customer communications. Lowest-stakes writing in the business, and there is a lot of it.

The one that earned its keep is not my own listing copy. It is the draft I hand the agent. She was not there for the panel upgrade or the joist sistering, so working from photos she writes around the house instead of about it. I do the first pass. She still rewrites half of it, but from something that already has the panel, the roof age, and the fact that the third bedroom is a real bedroom and not a converted porch.

The prompt is boring on purpose:

Draft a listing description from the scope of work and photos below. Every feature you name has to appear in the scope or be visible in a photo. No adjectives you cannot point at. Anything you want to say that is not in the source goes in a separate "confirm before using" list at the bottom, not in the description. Under 200 words. Neighborhood-level location only.

That last rule is the whole thing. Without it, you get the quartz counters. With it, you get a clean draft plus a short list of what it could not verify, and half of those are things I forgot to photograph.

That is a referral move, not just a time move. An agent who gets a real draft instead of a shrug remembers it. I have listing agents I send people to whether or not the deal comes to me.

2. Automate the task that runs without you.

Buildium's version is revenue tasks, the income that leaks when nobody follows up. The operator translation is any recurring task where the cost of skipping it stays invisible until it is expensive.

Two of mine run weekly. One pulls a market update: rates, local inventory, what sold. The other sweeps for what I should be writing about and leaves a ranked list in a folder before I sit down. Both are what I stop doing the week a deal gets hot, which is the same week I most need to know the market moved.

So pick by that test. The right first automation is the recurring one you have already quietly stopped doing.

Then bound it, and watch it, because a task that runs without you also fails without you. My sourcing sweep is scheduled for Sunday morning. It missed two Sundays in a row, and I only caught it because the file was not there when I went looking. Silence from an automated task is not the same as success. Build it so the absence is loud.

3. Stay ahead on the dates, because the dates are the money.

Buildium says lease renewals. A flipper's calendar is harder, because the expensive dates are irregular and each shows up only a few times a year:

• inspection and financing contingency windows on anything under contract

• the day earnest money goes hard

• assignment and closing deadlines on a contract you intend to move

• permit expiration on an open rehab

• the vacant property insurance binder, short-term by design and quiet when it lapses

These are not hard to track. They are hard to remember to track, which is a different problem and the one AI is good at.

The check: a date is not a date until it ties to a source document. Anything inferred gets flagged unconfirmed instead of landing on the calendar. An inferred closing date will be wrong once, and once is enough.

Where this bites.

That 8% is not the failure it looks like. Most of this should not be fully automated, and I would not want an unattended agent touching an insurance binder. The useful zone is the assisted middle: the machine fetches and drafts, you keep the judgment. The trap is stalling there forever. "I use AI for that" usually means "I paste something into a chat window when I remember to." That is adoption, not a workflow.

And without the check, all three get worse rather than better. Faster wrong is still wrong, and it arrives with better formatting. I keep numbers out of the model's hands entirely: my rehab estimates run on a deterministic engine underneath, and the model handles capture and narration, never the arithmetic.

The play. Take the thing you do most often and, before you automate one step of it, write a single sentence: what has to be true about the output before it is allowed to leave. Mine for the description is "every feature ties to the scope or a photo." For the dates, "every date ties to a document." Write the check first, then build backward. That sentence is the whole distance between the 58% and the 8%.

Source: Buildium's 2026 State of the Property Management Industry Report, written up by Robin Young.

Know an operator who is in the 58% and would admit it? Forward it. And hit reply: what is the one check you would write first? I read every one.

Written by Paul Cobb, real estate operator (Louisville / Kentuckiana), forVEX builder. I run the plays I write about. More depth: forvex.app/journal