Let's say it's 10:47 p.m. and a guest calls reception. They say the room is too hot, the shower drain is not working, and whoever is in the room needs extra towels.
To the guest, it's basically one conversation.
But what's actually happening inside the hotel? For them, it becomes different pieces of work. At least three. Engineering gets to deal with the AC and the drain. Housekeeping has to take care of the towels. And reception routes it and decides how urgent this whole thing is, whether to inform the manager on shift right now, and how quickly it has to be solved.
So the guest just asked one question, and reception now deals with three departments.
And that might be where one of the most useful applications of AI in hotel operations sits. The goal is not to replace reception, and it's not about adding another chatbot. In this situation, AI can become the operating layer between the guest, the hotel's tools and systems, and the people responsible for getting it all done.
A response is not a resolution
Right now, hotels have different ways for guests to ask for something. It could be the phone, it could be WhatsApp, it could be their own app, some kind of web chat. It could be an email. It could also be the front desk.
But the most interesting part starts after the request arrives.
Because reception has to understand what the guest wants, what they mean, identify them, decide if it's urgent or not, and work out who has to deal with it right now. Which department. Get the right context. And then follow up with the guest and explain to them what's going on.
And the delay is usually not the task itself. The problem is usually in the handoff. The message could go to the wrong channel, or a certain department could be unavailable, or some context is missing. And at the moment, nobody knows who has to own this task. As a consequence, the guest has to ask twice, or call again, or ask for some kind of update on what's going on.
So let's say AI answers in two seconds. It happens really quick. But housekeeping never gets the request. Then in this situation nothing actually got improved.
Reception is often the human coordination layer
Many hotels already have all the systems they need. Usually the problem is that those systems don't always behave like one system, one operation.
Let's say the PMS knows the stay. They also have messaging that holds the conversation. Housekeeping could be using another platform. Maintenance has something like their own queue. And restaurant information could also live somewhere else.
So if those systems don't connect properly and cleanly, someone has to bridge the gap. And very often it all goes to reception. It's usually reception who has to deal with all that.
They get the request, they look at the reservation in whatever tool they're using, they reach out to housekeeping, they talk to engineering, they wait for them to get back, and then they call the guest back.
Imagine how tight this capacity is. HOTREC said earlier this year that European hospitality is still missing around 10% of the workforce it needs. And very often, almost every time the operation gets busy, the answer is that they need another person. That's not a very well designed strategy. I think giving the existing team a much better operating layer is a much better solution.
What AI can change now
So what's already possible? An AI system can understand a guest request. It can identify multiple intents. It can get the right context, use tools, take permitted actions across the different systems the hotel already has. And then it keeps track of what happens next, which is the most important part.
There's this hotel, THE FLAG Zürich. It's a 101-room property. They have an AI agent that reads special requests in reservation comments and creates the relevant tasks in their housekeeping system. Apaleo says it currently creates around 30 tasks per day and saves an estimated 3.5 hours every week, just in direct task creation time. And that's before counting the reduced back-and-forth between reception and housekeeping.
That's basically how AI can handle this work.
What the system should actually do
So what, in my opinion, should the system actually be able to do? Take the same call. The room is too hot, the shower drain is backing up, and let's say someone needs extra towels.
A useful, working AI operating layer, what is it supposed to do?
It's supposed to identify the guest and their active stay. Separate the request into separate issues. Pull where exactly they're staying and the guest's current context. Send the AC and drain issues to engineering, send the towel request to housekeeping. Track whether those requests are accepted, so it can see that someone is on it, and when they're completed, it can see they've been completed. It keeps reception in the loop without them having to chase updates. And it has to be able to escalate when the situation crosses certain hotel rules.
Let's say engineering cannot resolve the AC inside a certain time frame. As one of the ways it can behave, that could trigger a manager review for this particular room. So the manager decides if they need to move these guests, or some kind of service recovery.
And what's really important is that AI should not invent that policy. It's not up to AI to decide that. The hotel defines it. The hotel sets the rules.
Before and after
| Reception coordinates | AI coordinates | |
|---|---|---|
| Guest request | Staff separate issues manually | AI identifies multiple intents |
| Context | Staff look up the stay | Relevant context is retrieved |
| Routing | Staff call or message departments | Work is routed under hotel rules |
| Status | Reception chases updates | Task state remains visible |
| Exceptions | Staff notice problems manually | Defined conditions trigger escalation |
And here it's very important: the whole point is not to remove people from this process, from this whole loop. The main idea is to stop requiring someone to manually move every single piece of information through different systems.
You set the rules. AI handles the rest.
Such an agentic system should not mean giving AI permission to run the whole property however it wants and just go berserk. It just means giving AI a certain level of autonomy to work through a certain set of decisions, bounded of course, and a certain set of actions, inside rules the hotel sets and controls.
AI could be really useful for understanding messy or unclear language. It could be understanding context. It could be finding certain information in the hotel's knowledge base. It could choose the right path, the way a certain request has to be routed right now. And what's more important, it can coordinate work across different hotel systems.
But we also need to remember that there has to be some deterministic logic. And it still has to own what has to behave predictably. It could be some kind of emergency escalation, certain financial limits, permissions, or a deadline. Or some kind of fixed policy.
And people need to stay responsible, of course, for serious complaints or compensation matters. It could be room moves, maybe safety issues, maybe certain guests become distressed. These situations also have to be handled by humans.
So AI can be used for the work that needs to be understood and coordinated. And people need to be kept where the guest needs judgment, where a situation is serious, and where it's important to keep a certain level of hospitality.
The real KPI: does the guest have to ask twice?
So in this whole picture, messages answered by AI is not the metric I would care about the most. It's not the primary metric. And it's also not about response time on its own.
The much better question is: did a certain request reach a real resolution without the guest having to ask twice?
And basically what it means is checking whether the request was routed correctly and properly, whether it was accepted, completed, communicated back to the guest, and actually resolved and closed.
Because let's say AI can answer the guest, but it cannot pull full live context, or create the right work, or use a certain hotel system, or monitor what happens next. In this situation the hotel just adds another channel for reception to deal with, and potentially clean up and manage.
And if AI in this situation can answer, but cannot move the work forward, can't actually do the work, the hotel just added a chatbot. And it has nothing to do with an operating layer that is actually performing actions that improve the hotel's operation.
The stress test is when the hotel gets busy
The real stress test for every hotel starts when the hotel becomes really busy. Let's say it's 6:00 p.m. and there's a group arrival, and they all hit the lobby at the same time. One guest asks about a crib. Another guest wants an update on his airport transfer. Another three rooms ask something about the restaurant. Someone says there's no hot water. And also, let's say, there's a VIP arriving in 20 minutes.
And of course AI cannot immediately create an engineer who is currently not on shift. In this situation, people still have to deliver the service.
But what it can do, it can stop every request from becoming another coordination problem the people working at reception have to keep in their heads. Basically the AI system takes several requests at once. It routes them and connects them to the right stays. It figures out what each request needs. It takes certain permitted actions, and it also tracks which requests remain open. And then it surfaces what it cannot handle right now, because there are always certain exceptions.
And that is basically how the same team that is currently on shift can handle more volume without the headcount increasing at the same rate.
AI also exposes weak operating rules
What's also important is that AI basically forces a hotel to answer questions staff previously handled informally. These questions could be: who has the breakfast hours information? What is currently urgent? Who is on call tonight? How long, let's say, can an AC issue remain unresolved? Who approves a late checkout? What does it mean that a task is actually resolved?
And all these questions are not AI questions. They are all operating questions. If nobody can answer them consistently and coherently, AI of course cannot fix the process. But what it can do, it can expose the confusion in the process. It can flag that a certain process was never clearly defined.
And that's what I mean by "You set the rules. AI handles the rest." It's not just positioning. It has to be built into the architecture, because that's the architecture itself.
Where I start
I don't think it's a good idea, and I do not begin with an autonomous AI concierge for the whole property. It starts where request volume is high, where the coordination becomes difficult, and where the consequences, of course, are manageable. For many hotels, in-stay guest requests are a very good candidate.
I look at the previous 30 to 90 days of calls, messages, unresolved tickets, maybe shift notes. And I find the requests guests make repeatedly, and very often a lot of them reception keeps chasing. These requests generate repeat follow-ups and repeat contact from the guests. And I also look at the places where information moves manually between the hotel's internal systems.
And then I build the AI operating layer around exactly those. I let AI understand and coordinate, and I configure what exactly it should understand and how it should coordinate. Certain things of course have to stay deterministic, and they need to enforce the hotel's fixed policies. People need to be kept on exceptions, and kept for consequential judgment.
And the measure in this situation has to be resolution, not something just novel.
Questions hotel operators ask
- Can the AI work inside their existing systems, the systems their staff already uses? This is exactly what it's supposed to do - work inside and with existing tools and systems. Because it doesn't need to create another inbox or dashboard. If that's all it does, it will just add more work instead of eliminating it.
- What happens when the AI is not certain, or unsure? In this situation there has to be a defined path to a person. And AI has to give them the relevant context.
- What should AI never decide on its own? There have to be rules that limit its autonomy. It has to be safety, and things like meaningful compensation. AI can manage and decide on a certain level of compensation, and it should do that, but meaningful compensation has to depend on human judgment. Certain sensitive guest situations or emergencies have to be handled by humans. And of course decisions that have financial consequences.
- What is the main metric a hotel owner should watch? I think it should be repeat requests and repeat contacts from the same guests. If the guest has to ask multiple times, it means the AI operating layer didn't close the loop.
The bigger opportunity
So what is the biggest opportunity? I think the most interesting opportunity in hotel AI is not another interface or guest-facing app. It's basically everything between the request and the outcome.
The guest asks only once, AI understands the request and what the guest needs and wants. The hotel's internal rules define what AI can do. The right systems and teams are kept in the loop and get the right context. The work keeps being done and it keeps moving. And finally, people step in where their judgment is needed and where hospitality has to be handled by humans.
And that is, in my opinion, where AI becomes a part of the hotel's day-to-day operation, instead of becoming just another piece of software sitting beside it.