AI in a hotel, without a new PMS: where to start

Most hotels that want AI don't need a new PMS first. The data they need is already there, in the PMS, the channel manager, the review sites and the stock records. It's just in ten places, and the morning report is still put together by hand. This is the order we'd start in, from our work building blurple hospitality OS with a hotel group.

1. Start with a question, not a tool

Before choosing anything, find where hours and money leak. In most hotels the same few come up:

• the morning report, assembled by hand from several systems;
• commission paid on channels, and rates that don't match from one channel to another;
• stock that runs out, or gets ordered in a rush at the wrong price;
• reviews that go unanswered, and complaints nobody connects to the department behind them.

Pick the one that costs you most. That is your first use case, and also how you'll measure whether AI was worth it.

2. Get the data out before anything else

AI can only be as useful as the data it can read.

• The PMS: cloud PMSs such as Oracle OPERA Cloud offer APIs. Older on-premise systems usually don't, but they can export scheduled reports, and that is enough to start.
• The channel manager: systems like HotelRunner hold bookings and rates per channel.
• Reviews: Google, Booking and TripAdvisor, where most guests speak.
• Stock and accounting: often a spreadsheet or an export. That's fine.

Bring them into one hotel data model, so a delay in housekeeping shows at the front desk and a banquet order shows in the store room. Each new system becomes a new adapter, not a new project.

3. Begin with tools that read, not tools that act

The first things worth switching on change nothing in the hotel's systems:

• A daily brief for management every morning: occupancy, revenue, problems and what to do today, on WhatsApp or by email.
• A concierge for the team: questions in plain words, such as "What did we pay in Booking commission last month?", answered from the hotel's own data, with the source.
• Alerts: a rate error, an overbooking risk, a sudden wave of cancellations.

They are low-risk, they're used every day, and they build the trust the next steps need.

4. Then one module with money in it

Once the data flows and people trust the numbers, add one module where the gain can be counted:

• Revenue: checks on commission, rate mismatches and rate parity, then price suggestions by season, day and live demand.
• Procurement: consumption learned per occupied room and per event, a warning before stock runs out, a suggested order.
• Guest voice: reviews sorted by department and topic, and reply drafts.

One module at a time. A hotel that switches on ten at once can't tell which one worked.

5. A person approves anything that touches money, guests or orders

AI should suggest; people should decide. Orders, price changes and messages to guests arrive as suggestions, and the system records who approved them. As trust grows, a hotel can let some of them run on their own, but that is a decision to take later, deliberately.

6. Every number shows where it came from

Hoteliers trust numbers they can check. Make every figure clickable down to the record it came from. A brief that can't show its source gets ignored by the second week.

7. Protect guest data from the start

• Mask identity, contact and payment details by default; run analysis on aggregated or anonymous data.
• Sign a data processing agreement that says where the data is kept. In Türkiye that means KVKK; for European guests, the GDPR too.
• Keep a record of who saw and changed what.

8. Go where the team already is

Managers work at a desk; housekeeping and maintenance don't. Give the office team a web panel, and let the rest receive and close their tasks on WhatsApp, or by voice. A tool that asks a housekeeper to log in somewhere won't be used.

9. Prove it in a working hotel

Try each module first as a simple prototype in a real hotel, and measure four things:

• hours saved on reports and checks, before and after;
• money found in commission gaps, rate mismatches and rushed purchases;
• forecast accuracy, where there's a forecast;
• use: whether the GM actually opens the brief, and how many questions the team asks.

Keep what proves its worth, change what nearly does, and drop the rest. Only then build it properly.

What we'd avoid

• Replacing the PMS to "get ready for AI." It's the most disruptive change a hotel can make, and it isn't needed to start.
• A guest chatbot first. It's visible, but it touches guests before the hotel trusts its own data.
• Dashboards nobody opens. If the insight doesn't arrive where the team works, it doesn't exist.

Where we come in

• blurple hospitality OS: one AI layer for the whole hotel, on top of the PMS you already run. In development, with early access open.
• Applied AI: copilots and agents inside your own products, designed so people can check the answers and approve the actions.

We're building this with a hotel group, module by module. If you run a hotel or a group and want to compare notes, we'd be glad to.