Operations9 minute readLast checked

AI for serviced accommodation: an operator's guide

AI for short term rental management works when it sits on top of an operation you've already written down, and that's the step it's easiest to skip. In a short let or serviced accommodation business it earns its place in four layers, in this order: how your operation runs, the jobs that run themselves, AI for what no rule covers, and all of it joined up in your team's own channel.

In this piece
  1. Why AI on a messy operation makes the mess faster
  2. Layer one: write down how your operation runs
  3. Layer two: the jobs that keep slipping, run on their own
  4. Layer three: AI for what no rule covers
  5. Layer four: all of it joined up, in your team's Slack
  6. What we've seen since 29 June
  7. It won't talk to your guests until you say so
  8. Where to start at 15 doors, and at 100
  9. Questions
  10. Sources

LET'S SAY A GUEST WANTS TO DROP THEIR BAGS

A guest messages at nine asking if they can leave their bags at the flat at eleven. The calendar says last night's guest checks out at ten, so an AI reading the calendar says yes, warmly, inside a minute. What it can't see is that the cleaner isn't booked till one, because that's in a WhatsApp thread, or that nobody goes in until the clean's signed off, because that rule lives in your head. At eleven the guest is standing outside a flat that hasn't been turned, and someone on your team has to explain why they were told yes.

Why AI on a messy operation makes the mess faster

Nothing went wrong with the AI in that scene. It answered from what it could see, quicker than anyone on your team would have, and any AI reading that calendar would've said the same, ours included. The rule that mattered just wasn't anywhere it could read.

Most people think AI is what sorts out a messy operation. In reality it inherits it. Whatever your team does by habit, it can't do at all, and whatever's written down badly, it does badly and quickly. That's why AI comes third in what we build, not first, and every layer below runs in our own operation before it goes into anyone else's.

Layer one: write down how your operation runs

LET'S SAY YOUR CLEANER DOESN'T TURN UP

Check in's at four and the flat hasn't been turned. In most operations that ends in a phone call to someone senior, because nobody else knows who the backup cleaner is, whether the guest can be moved, or what they're allowed to offer. In a written operation your VA opens the decision tree for it. Is the backup free? Can the guest move to another free home? If neither, they send the holding message and one line to a founder. The guest hears something sensible within minutes, and it reaches you as a line in the handover rather than a call.

None of that is AI. It's the layer that makes AI possible, because a tree with the message to send at every end point is something a new VA can follow in their first week and something an AI can follow every single time. Roles with an owner and a backup work the same way, and so does a written line between what your team decides and what comes to you. You can click through that tree one branch at a time in layer 01 on the home page.

Layer two: the jobs that keep slipping, run on their own

Most of what goes wrong in a short let business isn't a mystery. Your team knew the damage deposit needed checking. It lost out to ten enquiries and a broken boiler, and more training won't fix that, because the problem was never what they knew. It was what they had time for.

In our own operation, a pre arrival check runs at eight and again at four against every arrival that day, looking for what's behind the worst check ins: a deposit that should be in and isn't, no check in message, no access details. Whatever it finds lands with the team alongside what to do about it. The first dry run caught 5 arrivals with no deposit recorded, on a team that knew perfectly well deposits mattered.

Strictly, that's not AI either. It's a scheduled job reading live data, and plenty of what operators want from AI turns out to be exactly that. It's worth having whatever you call it, and we've written up the deposit check on its own if that's the job that keeps slipping for you.

Layer three: AI for what no rule covers

Automation does the same thing every time, which is fine until the situation isn't the same. The awkward enquiry, the guest outside your policy and the clash nobody spotted don't fit a rule, so they either wait for someone senior or get decided by whoever happens to be on shift.

A MONTH LONG REQUEST

A month long request comes in and the dates are open, so a rule would accept it. Your VA asks the AI layer whether the money works. It answers that, then flags something nobody asked about: the guest already in that property has asked in writing to stay on. That took about 12 seconds, the money check included, and instead of two guests and one front door, your VA has a call to make before anything's confirmed.

A rule answers the question it was given. The AI layer holds what the question was for, a booking that actually works, which is why it reads the thread a rule never opens. You can see the same request run both ways on the home page.

Every answer also comes with a line on why, so your VA learns the call and not just the answer. When the same situation keeps coming up, it drafts the procedure from how your team resolved it, and nothing joins your operation until you approve it. That feeds straight back into layer one, so your operation gets written down a little more every week without anyone sitting down to write it.

Layer four: all of it joined up, in your team's Slack

OpSmithOs is the full version: an executive ops manager in your team's Slack, plugged into your PMS and your guests' WhatsApp. Your team asks it instead of waiting for someone senior, and the call gets made the way you'd make it.

AN EIGHT WEEK ENQUIRY

An enquiry lands for eight weeks, with a head count and the beds they need. By hand, your VA opens each calendar, checks the beds and writes the reply. The searching alone used to take us 45 minutes, and the doors nobody had time to open never got offered. In Slack, your VA asks once. OpSmithOs checks every door against its live calendar, how many it sleeps and its bed setup, ranks the ones that fit and writes the reply, in about 40 seconds. Your VA reads it and presses send while the guest's still shopping around.

Nothing from the earlier layers gets thrown away. The written operation is what it reads, the systems are what it runs and the AI layer is how it decides, and your team gets all three from one place they already work in. An SOP is your judgement, written down once and already out of date. This is your judgement, on shift.

What we've seen since 29 June

It went live in our own operation on 29 June 2026 and reports on itself every day, so these figures come from its own reports, not from anything we've estimated.

In the week to 23 July the team asked it 612 questions, and 88 percent were closed with no human needed. The rest it didn't just leave. It grouped them into patterns and drafted the fixes itself, from availability edge cases to a check on deposit status.

The number that tells you more is the gap rate, the share it couldn't fully close. In the week to 29 June it logged 15 questions and hit a gap on 3 of them, one in five. In the week to 27 July it logged 5,030 and the gap rate was 2 percent. Most of that volume was chatter rather than real questions, so it isn't 5,030 decisions. What matters is the direction: it got far busier in a month, and the share it couldn't close fell instead of climbing with it.

It isn't finished, and the number we watch most says so. Over four days in late July the team still needed a founder four times, then four again, then twice, then once, against a target of about one a day. That's the right direction, and it isn't zero.

For you, that adds up to a team that asks more and needs you less. More than that, it's the confidence that the enquiry got answered properly, the guest got looked after, and the call made while you weren't watching was the one you'd have made.

It won't talk to your guests until you say so

Shadow mode is where it starts. It reads each guest message in the Hostaway inbox and on WhatsApp and proposes the reply or the escalation, and your team decides what goes out. Anything that moves money or can't be undone stays with a person either way.

Nor does it replace your VAs. They get better at the call, and the judgement it works from is still yours.

Where to start at 15 doors, and at 100

The order doesn't change with size. Where you join it does.

Where an operation of each size tends to start, in the same four layer order.
What changesAround 15 doorsAround 100 doors
Who holds the operationMostly you, and it works because you're there.A team, a few senior heads and a lot of WhatsApp threads.
What breaks firstThe week you're ill or away, there's nothing to hand over.The jobs everyone agrees matter lose out to the ones with someone waiting.
Where to startLayer one: the situations that still come to you, written down as decision trees.Layer two: the deposit checks, the enquiry search and month end, off the pile.
Where AI earns its placeOnce there's a written operation for it to work from.Straight after, on the calls that don't fit a rule and still land on whoever's most senior.

At 100 doors the instinct is another hire, often an ops manager. Another person doesn't fix the pace, though. It gives you two people triaging the same pile, and the job that keeps losing still loses. Most operators sit somewhere between these two columns, which is why what gets built is scoped to a conversation rather than picked off a list.

Questions operators ask about this

What can AI do for a short term rental business?

In a short let or serviced accommodation business, AI earns its place on the calls no rule covers: an enquiry that doesn't fit, a guest outside your policy, a clash nobody spotted. Recurring jobs such as enquiry searches, pre arrival checks and month end are better run as systems on your live data. Both work far better once your operation's written down.

Is AI worth it for a small serviced accommodation business?

At around 15 doors the bigger win is writing down how you run things, so the business doesn't depend on you being there. AI earns its place once there's an operation for it to work from. Each layer builds on the one before rather than replacing it, so starting small never means starting again.

Will AI replace my short let VAs?

No. What it takes away is the waiting, whether that's on a calendar search, on somebody senior or on you. Every answer comes with a line on why, so your VA learns the call rather than just the answer, and handles the next one the way you would.

Can AI reply to Airbnb guests without a human?

Several tools can, but whether yours should is a decision to make after you've watched it work on your own inbox. Ours starts in shadow mode: it reads each guest message as it comes in and proposes the reply or the escalation, and a person decides what gets sent. Anything that moves money or can't be undone stays with a person regardless.

What's the difference between automation and AI for Airbnb management?

Automation does the same thing every time, which is right for a pre arrival check or a month end sweep. AI is for when the situation isn't the same, such as a month long request that clashes with a guest who's asked to stay on. You need both, and the automation tends to come first.

Does AI for short let management work with Hostaway?

Yes. Ours runs on Hostaway and reads the same live calendars, reservations and guest messages your team uses, through its API. Hostaway also launched its own AI CoHost in June 2026, which answers questions about your account, drafts and recommends inside the PMS, and waits for a person to approve before anything's applied. A build like ours sits alongside that and adds your operation: your rules, your team, and the calls that would otherwise come to you. On another PMS it's built through that PMS's API, wherever it opens one to your account.

Sources

  1. Hostaway Public API documentationapi.hostaway.com
  2. Hostaway, AI CoHosthostaway.com
  3. Hostaway changelog, AI CoHost updateschangelog.hostaway.com
  4. The Host Report, Hostaway launches an AI CoHost (26 June 2026)thehostreport.com

Last checked . Vendor features change. If anything here no longer matches the vendor's own documentation, theirs is the one to trust.