An AI agent that checks your cleaner is confirmed

Carlos is a made-up host with six listings. This example follows one Friday evening and the three homes that turn over on Saturday. The agent is set up once with this brief.

The eight chat bubbles in the first scene are the plain-language way of filling this in.

1. Read tomorrow's check-outs and check-ins from the booking calendar Carlos already keeps.
2. Match each home to its cleaner and text them: "Hi Marta, Lakeview Cabin checks out tomorrow at 11:00, next guest at 16:00. Can you confirm?"
3. Read the replies and count only clear yeses. Anything vague gets one polite follow-up question.
4. At 5:00 pm, nudge anyone who has not replied, once.
5. At the 6:00 pm cutoff, if a home still has no yes, message Carlos with the facts and two buttons.
6. If Carlos says offer, check the calendar once more, text the backup, and tell the first cleaner the job has been covered so nobody turns up twice.
7. Log every step. When all homes are confirmed, stop until the next evening.

The agent does not guess whether a cleaner will turn up. It reports what people have said and what they have not.

A made-up Friday evening. This table is also the data behind the timeline chart.

• Sam replies at 6:02 pm, before Carlos taps. The agent tells Carlos and lets him choose: keep Sam, or go ahead with Lena.
• Lena doesn't answer either. The agent tells Carlos straight away so he can phone someone. It does not text a long list of strangers.
• A booking is cancelled or a check-out moves. The morning check catches changes, tells Carlos, and drafts the update to the cleaner for him to approve.
• A cleaner declines. The agent doesn't wait for the cutoff; it tells Carlos at once.
• A reply with a condition, such as "can't do 11, can do 1pm". The agent does not count it as a yes. It shows Carlos the exact words and asks whether 1 pm still fits the 16:00 check-in.
• Carlos doesn't answer his phone by a set time. He can choose in advance what happens: keep waiting and remind him, or offer the job to the backup automatically. The default is to ask first and keep reminding.

Possibly, if you have one dependable cleaner and a simple reminder already works. Scheduling and turnover apps already cover the standard schedule: sending jobs, letting cleaners accept or decline, and assigning backup cleaners. If yours does what you need, use it.

What this example adds is the follow-through around a cleaner who stays silent:

• chasing at set times (ask at 4 pm, nudge at 5 pm), on the channel you already use with your cleaners;
• escalating by elapsed time and urgency, so at your cutoff you get one message and one tap to ask the backup;
• reading free-text replies such as "can't do 11, can do 1pm" or "probably", and showing you the cleaner's actual words instead of counting them as a yes;
• telling you, so you hear tonight and not at 10:45 tomorrow.

Whether that is worth building depends on how often a turnover goes wrong for you today. We would ask you that before recommending anything. We have not compared this against every app on the market, so check what yours does first.

For the example, the agent reads a booking calendar (for instance a calendar feed or your property management software) and sends ordinary text messages to your cleaners. What is possible for you depends on the tools you use, and we would check that on a discovery call before promising anything.

The texts go to your cleaners only, not to guests. It does not log in as you or post on any booking site.

• It asks first. Reassigning a job, changing a rate or telling a cleaner they are off the job needs your yes. You can later choose to let low-risk steps run on their own.
• Only a clear yes counts. The agent treats "probably" as "not yet".
• Cleaners only. It messages your cleaners, never your guests.
• Limited data. It sees the dates, home names and cleaner contact details it needs. It does not need door codes or guest details, and we would keep them out. See how we handle your data.
• Logs everything. You can see every ask, reply, nudge and tap.
• What it can't do. It cannot make anyone turn up. It cannot clean the home. A confirmed cleaner can still be sick, and it will not predict that.
• Its own errors. If it misreads a reply, the log shows it, and the cutoff message always shows the cleaner's actual words.

These are illustrations too, not client results.

• Home improvement. A supplier's price list arrives with the lumber 12% higher. The agent recalculates the margin on open quotes, finds one below the owner's floor and drafts a revised quote. The owner reads it and decides whether to send.
• Short-term rentals. This page.
• Translation agencies. A rush job is offered to a linguist who does not answer within the agreed time. The agent offers it to the next linguist on the list and tells the project manager. It works with the job's admin details, not the text being translated.
• Editing and small presses. A project goes past the agreed rounds of changes. The agent notices from the tracking list and drafts a note to the author for the editor to approve.

The pattern is the same each time: watch something, check on a schedule, act when a threshold is crossed, and keep a person in charge.

We'd start with one workflow and a measurable goal, not a big project.

1. Discover: a free 30-minute call about how your turnovers actually run.
2. Assess: we map what you already use and where things slip.
3. Pilot: a small version with your own cleaners and calendar, with the agent asking first.
4. Scale & run: extend it once it has earned your trust.