The Definitive Guide to Running Your Hotel with Claude and 24 AI Agents

How independent, boutique and small-group hotels can start using Claude. And why 2026 is the year you can no longer put it off.

---

🇬🇧 English version here

---

🖐️ A note upfront: this guide took longer to write than expected. Much longer. The reality inside a hotel is complex, and it varies enormously depending on the type and size of the property — so I had to simplify in places. Take it with a pinch of salt and adapt it intelligently to your actual context. And if you find alternative or more elegant solutions, let me know!

2026 is the year that decides who stays visible

Mike Coletta, Phocuswright analyst, published a global survey in January 2026 covering 178 travel executives: 88% say gen AI has already had a positive impact on their business, but only 12% feel genuinely ready to implement it.

The study shows that 61% of travel companies are already experimenting with or scaling agentic AI. 56% are aware of, exploring, or beginning to implement MCP and A2A interoperability protocols. 60% consider agentic commerce decisive or highly relevant over the next 3–5 years. These are numbers that reflect budgets already allocated.

And then there's the specific point for small hotels. Norm Rose, also a Phocuswright analyst, in his farewell report Will Agentic AI Disrupt Travel Distribution? (2025): hotels — especially small ones — ceded Web 2.0 to the OTAs. Every attempt to bring guests directly to their own sites has been, for fifteen years, a chase that never turned into a lead. Now a window is opening. He describes it like this: "there is an opportunity for individual properties and small chains to convince AI that they offer more value, provided they can implement their own agentic platforms."

Translation: if a conversational agent can compare you against four Marriotts in ten seconds, you don't win on branding. You win if your data is readable, your policies are clear, your experience is consistent, and if you have your own agents working for you 24 hours a day. Not to replace staff. To capture the revenue that walks out the door today without anyone noticing.

This guide tells you how. Not for the 8,400-hotel chain like Wyndham, which invested $425 million in AI and activated 250 dedicated agents (Wyndham Q3 2025 earnings call, November 2025). For your 12-, 30-, 80-room hotel. Or for the small group with three or five properties. The one where staff is never enough, the PMS doesn't talk to the CRM, and the Thursday-evening booking request in German sits unanswered until Monday.

---

The 7 pressures squeezing hotels today

Before we talk about Claude and agents, let's talk about what we're actually trying to solve. Every use case you'll find further on comes back to one of these seven pressure points.

1. Labour costs rising while staff doesn't show up. More than 50% of US properties report being moderately to seriously understaffed (AHLA Front Desk Feedback Survey, March 2026). Labour costs in 2026 grew roughly 5% compared to 2025 (Today's Hotelier, January 2026), and already account for around 50% of a hotel's total operating costs (Travel Outlook industry analysis, 2024). The problem isn't finding people. It's sustaining the model. Adam Glickman and Dmitry Koltunov, Phocuswright analysts, described it back in 2023 in Real-Time Revolution in Hotel Operations: patterns that today only operate in well-equipped chains, such as rolling rooms (not cleaning stayovers when the evening won't be sold out) and day-parts scheduling (scheduling staff across micro time slots). For an independent, these patterns are still theory — because there's no one to orchestrate them every morning.

2. OTA dependence. Booking.com takes 15–18% in most markets, Expedia 18–22%, and for independent hotels the real range goes up to 25% and beyond with preferred programmes (Cloudbeds OTA Commission Guide, December 2025). Booking Holdings in Q1 2026 reports that just over half of bookings come through directly. The rest is commission. Remember what Norm Rose said: small hotels ceded Web 2.0 to the OTAs. Agentic demand doesn't automatically solve the problem, because OTAs will be the first beneficiaries of agentic commerce thanks to their already-ready technical infrastructure. The window for independents only exists if they build their own agents.

3. The inbox is the real front desk. A booking enquiry that sits unanswered for four hours goes to the competitor. An enquiry in a language other than Italian or English often goes unanswered entirely. The inbox is the most neglected piece of hotel technology, and how well it converts directly determines the following week's occupancy. No revenue management dashboard captures this data, because the lost booking leaves no trace.

4. Reviews multiplying while responses can't keep up. Roughly 933,000 hotel reviews are published globally every day. The industry's average response rate is around 40%, and answering all of them would require approximately 37,350 hours of global labour per day (MARA Solutions analysis, May 2025). One extra star in the rating correlates with a revenue increase of between 5% and 9%, according to Harvard Business School research. Hotels that respond promptly to TripAdvisor reviews show 17% higher engagement and a 21% greater likelihood of a booking enquiry (Cornell Center for Hospitality Research). These numbers have been known for years. What changed in 2026 is that the tool to close the operational gap now exists.

5. AI search visibility being reallocated right now. In July 2025, Phocuswright published Chat, Plan, Book, measuring that 34% of travellers are already open to booking inside an AI platform when the functionality becomes available. Wyndham, as we've seen, has already activated integration with Claude, Gemini and ChatGPT. Mews in its 2026 Hospitality Industry Outlook (December 2025) cites Chris Hemmeter of Thayer Investment Partners saying it plainly: hotels that don't sort out their data will be invisible. The point isn't optimising for Google. It's being citable by a conversational agent that doesn't look at your landing page. I'm talking about GEO — if you've never heard this term, listen to this episode of my podcast.

6. Fragmented software with no common semantic layer. PMS, channel manager, booking engine, CRM, OTA extranets, review system, admin management system, F&B. Seven or eight systems producing data that don't talk to each other. Wouter Geerts of Mews called it the missing semantic layer (HotelYearbook, January 2026), and in the Phocuswright January 2026 survey, 50% of travel executives cite integration complexity as the main obstacle to adopting new technologies, versus 45% citing a skills gap and 40% citing legacy systems. The main block isn't ideas or budget. It's systems that don't talk to each other.

7. Personalisation. "You will be remembered, you will receive offers tailored to your behaviour." Every hotel CRM vendor has been repeating this line since 2014. Meanwhile the average guest receives the same automated welcome email as the guest who's stayed five times. Personalisation never arrived because it requires an orchestration that the standard PMS doesn't do, and that until recently was too expensive to build from scratch.

---

The camera flip

The gap between the 88% who see a positive impact from gen AI and the 12% who feel ready to implement it isn't a story of immature technology. The technology is mature. It's a story of workflow design.

The front desk that doesn't answer a request in German isn't a language problem. It's a workflow that assumes every email passes through a human. Occupancy that suffers isn't a pricing problem. It's revenue management that runs once a week instead of once a day. The one-star review with no response isn't a reputation problem. It's a workflow that depends on the manager remembering to open TripAdvisor on Tuesday.

When the workflow is the problem, the agent can be the solution. Not additional staff. Not another vendor. The agent running while you're having breakfast.

And here's where Greg Abbott of DataArt's observation comes back in: the difference between companies that use AI to do the same thing with 10% less and those that use it to do more with what they already have. The first ones cut. The second ones grow.

---

The first 15 minutes: basic setup

Before we talk about 24 agents, let's talk about what you can do this afternoon. Three micro-steps, five minutes each, and by the end you have Claude responding in your brand voice, pulling from the documents you've given it, and producing output that sounds like you. It's not an agent yet. It's the foundation every agent runs on.

Prerequisite. You need at least Claude Pro (roughly $20 a month): the free version doesn't include the Opus model or Extended Thinking, which are the two ingredients that make the difference between a pretty answer and a useful one in a hospitality context. For an independent hotel, Pro is enough. Only if you manage multiple properties or want to activate the business Connectors (Slack, HubSpot) is it worth moving to Max or Team.

---

① Setup — Create your hotel-profile.md

Open Claude in the browser. Select the Opus 4.7 model with Extended Thinking active (the option is in the selector at the top). Time: 5 minutes.

Paste this prompt:

Interview me using AskUserQuestion to build my hotel-profile.md
file. Cover: hotel name and location, number and types of rooms,
target guests and main origin markets, brand voice (formal, warm,
informal), the three things we're known for, the three things we
are NOT (e.g. we are not a budget chain), key policies
(cancellation, pets, children, smoking), director's name for
signatures. Keep the final file under 2,000 words.

Answer Claude's questions one by one. At the end, save the output as hotel-profile.md on your desktop. This is the document Claude will read before every task from here on.

---

② Context — Create the Cowork folder

Download the Claude desktop app (claude.ai/download). Time: 5 minutes.

On your computer, create a folder named exactly Claude. Inside it, move the hotel-profile.md file you just created. Open the desktop app, go to the **Cowork** tab, and select that folder as your working folder.

Go to Settings → Cowork → Global Instructions and paste:

Read hotel-profile.md before every task.
Save all outputs to this folder.

From this moment on, Claude remembers who you are every time you open a conversation, and everything it produces stays in the hotel folder — not scattered across different chats.

---

③ Output — The first real task

Time: 5 minutes. Pick a task you do every week. The simplest and highest-return one: responding to a booking enquiry that arrived by email in a language other than your own.

Open a new chat in Cowork and paste this minimal prompt:

I want to respond to this booking request [paste the guest email].
Read my folder. Use AskUserQuestion before starting to check if
anything is missing.

Claude will ask you one or two targeted questions (are the dates confirmed? which rate to propose? is there availability?), then generates the response. You read it. If it sounds like you, send it. If it doesn't sound like you, add what's missing to the hotel-profile.md file and run the task again.

This is the core loop: task → output → profile correction → better output. In fifteen minutes you've built the foundation, and from tomorrow every request in German, French or Japanese that lands in your inbox can be processed with this same flow.

---

Full setup to get to all 24 agents

The quick-start above gives you the operational base with which Claude works as a personal assistant that remembers who you are. What follows is the structure on which to build all 24 agents in the guide, and you do it once only. It takes roughly ninety minutes the first time; after that the system is up and running.

Claude functions like an operating system, not a chat. The practical difference is this: a chat forgets everything. An operating system remembers contracts, brand voice, rate sheets, SOPs, OTA contracts, historical reviews, and applies them every time. Setting it up properly once means every agent that comes after starts with the right context.

Step 1. Create a Project for your hotel.

Go to claude.ai, create a new Project. Name it after the property: Boutique Hotel Della Pieve, Gruppo Tirreno Properties, whatever it is. One Project per property if you're independent. One Project per brand plus sub-Projects for each property if you're a small group. Don't mix Projects. The Project is the unit of memory.

Step 2. Upload the key documents.

Brand book, operations manual, rate sheet, contracts with tour operators and OTAs, guest FAQ in two or three languages, the concierge guide to selected vendors (restaurants, transfers, experiences), the playbook of...