In the last article, I wrote about why I stopped shopping for a PMS and started building my own. The short version: every platform I tried was built by people who’ve never actually run a property. Vendor coordination is where that gap shows up the clearest, so it’s the first place I built for.
If you run vacation rentals, mid-term rentals, or furnished housing, you already know: managing the cleaner calendar is one of the hardest parts of the job.
I have my horror stories. The next guest is standing at the door, and the room inside is a mess — because my cleaner forgot they had a job that day. Or worse: I misread a booking, or a last-minute reservation came in, and I forgot to schedule a cleaner at all. It’s manageable at 5 rooms. At 50, it isn’t.
1. Why the fancy tools don’t actually help
Traditional PMS platforms are proud of their calendar integrations — email notifications, third-party cleaner management tools, all of it. I don’t use any of it, because it misunderstands who’s on the other end.
Cleaners aren’t sitting in an air-conditioned office checking emails from 9 to 5. They’re on the go, moving from one property to the next, all day. They don’t have time to open an app and read a notification. They use WhatsApp, or they call. That’s it.
2. What I tried before AI
At 5 rooms, I didn’t need software at all. I wrote out each cleaner’s jobs every week and sent it over WhatsApp. When a job was done, I sent payment manually, one cleaner at a time. It was a lot of work, and the payment side especially was a mess — but at that scale, it was doable.
Then I tried the obvious tools: Google Sheets, Google Calendar, a couple of rounds with Todoist, paid or free tools, I tried them all. None of it mattered — they were all about equally bad for what I needed. I was still tracking everything manually behind the scenes, and every one of these tools still required copying jobs over into a WhatsApp message by hand, which usually meant another round of editing once they were there. It was a mess dressed up as a system.
That’s the pattern that stuck with me: low-value work, high stakes if you get it wrong, and you professional you could reasonably hire just to do this. So it fell on me — until I could hire AI to do it instead.
3. What I built instead
So I built the version that actually matches how my cleaners work. The system now handles cleaner scheduling, bed type and job scope per room, pricing, and — the part that used to eat a full afternoon every other week — calculating bi-weekly payments per cleaner and emailing each of them an itemized breakdown, no manual math required.
The optimization piece is where the real money is, and it’s also the part that’s too specific to my operation for any off-the-shelf PMS to have built. Some of my cleaners price by location, not by room — if they’re already on-site, an extra room in the same building costs less than a separate visit, because they’re saving their own travel time too.
This is where property type matters, and it’s something I had to teach the AI explicitly. On a short-term or vacation rental, you generally don’t have room to optimize — guests book back to back, sometimes last-minute, and the clean has to happen right after checkout no matter what. But on furnished housing or mid-term rentals, turnover is much lower, and there’s often a day or two of slack between one guest leaving and the next arriving. That slack is exactly what the system looks for: a chance to shift a clean by a day and combine it with another unit at the same address. One recent example: a room with a checkout scheduled for a Saturday got pushed to Sunday so it could be combined with another unit in the same building checking out that day. Same cleaner, one trip instead of two, and a good discount for the combined job instead of two full-price visits.
The daily piece is the one that matters most operationally day to day: every morning, it generates each cleaner’s task list and sends it straight to WhatsApp, so I’m not the one manually forwarding reminders anymore.
4. The pros: this is exactly the job AI should take over
This is the category of work I now think AI is genuinely built for. It doesn’t require a brilliant mind — it requires an okay brain and complete reliability, applied to a job that’s high-stakes if you get it wrong but too low in value to justify hiring premium talent to do it. That combination used to mean it fell on me. Now it doesn’t, and that alone has made my life considerably easier.
5. The cons: I still can’t fully let go
AI makes mistakes. Constantly. And the more I improve the system, the more new bugs I introduce along the way — every change is a chance to break something that was working.
It saves me an enormous amount of time. But I still can’t treat it as fully autonomous — everything gets a review pass before it goes out. The mental model that’s worked best for me: this is my AI employee, and my job now is supervision, not execution.
Where this leaves me
Vendor coordination was never the interesting part of running COZi. It was just the part that broke the most, and the part no off-the-shelf tool was built to fix — because it was never built for how cleaners actually communicate. Replacing it didn’t require a smarter system. It required one narrow enough to fit the job.
If you’re an operator reading this, I think this is your era. You hold something AI doesn’t have and nobody else can replicate — years of hands-on, day-to-day operating knowledge, the kind that only comes from actually doing the job. That’s the part that used to be hard to scale. Now it’s the part worth building around.
Start building.
