What I said
at FITUR TechY
In late January I joined Bruno Hallé of Cushman & Wakefield on the opening stage of FITUR TechY in Madrid, as part of the TechYnegocio programme organised by FITUR in collaboration with the Instituto Tecnológico Hotelero — marking the forum's 20th anniversary. The brief: make the case for practical AI in hotel management, not the promotional version.
The argument I made is the same one I keep making in client conversations, and the one this blog keeps returning to. The technology is not the problem. The architecture underneath it is.
That might sound like a technical observation. It isn't. It's a strategic one — and getting it wrong is why most hotel AI projects produce conference presentations instead of results.
Five to eight systems
that don't talk
A typical hotel operation runs five to eight systems that don't speak to each other. PMS, CRS, RMS, POS, accounting. Each holds a fragment of the truth about the business. What the channel mix looked like last Tuesday. What a guest spent at the restaurant versus what they booked the room for. What the cost per occupied room was in February versus March.
AI applied to any one of those fragments produces a local optimisation at best, and a confident hallucination at worst. The model reasons against whatever it can see — and what it can see is partial, inconsistent, and often stale.
The instinct when this becomes visible is to buy a new integration layer. Connect the systems. Build the API bridge. That's usually the wrong move, or at least the wrong first move. What the data actually needs is not another tool on top of the stack. It's a clean, standardised environment underneath it: a data layer that pulls from every system, normalises the schema, and becomes the single surface any AI model can reason against without fighting the fragmentation.
Adding AI to fragmented data doesn't fix the fragmentation. It amplifies it.
Marriott and Hilton understood this early. They didn't replace their PMS — replacing a PMS in a large hotel operation is a multi-year programme with significant operational risk. They built a parallel environment where data flows cleanly from every source and any analytical or generative layer can be applied without fighting the underlying structure. That's the model. Not the tools on top of it, but the architecture underneath it.
The lesson for independent hotels and smaller groups isn't that they need to replicate what Marriott spent eight figures to build. It's that the sequencing principle holds regardless of scale: clean data first, intelligence layer second.
The wrong place
to start
There's a persistent bias in hospitality AI toward customer-facing applications. Chatbots. Concierge agents. Personalisation engines at the booking funnel. These are visible, they're easy to demo, and they produce the kind of content that looks good in an innovation report.
They're mostly the wrong place to start.
The returns in back-of-house are larger and faster: demand forecasting, shift planning, room assignment logic, maintenance scheduling, revenue reporting. These are high-volume, rules-adjacent tasks where AI creates genuine operational leverage — freeing the people who currently do them manually to do the things only people can do well.
The front of the house is a hospitality business. It should be run by people who are good at hospitality — at reading a guest, at managing a difficult situation, at creating the kind of experience that generates a repeat visit. The back of the house is an information processing business. Reservations, schedules, pricing signals, cost structures. That's where AI earns its place first, and earns trust with the teams who will eventually extend it further.
The chatbot that handles check-in questions is fine. The revenue manager who can ask a plain-English question about channel mix and get a correct, sourced answer in thirty seconds — that changes how the morning starts. Start there.
One question
before the pilot
The question I kept returning to on stage, and the one I'd ask any hotel group right now, is this: do you have a single, clean, queryable record of your reservation history, your market segmentation, your channel performance, and your cost structure? Not spread across exports and reports that someone stitches together in a spreadsheet each Monday morning. In one place, in one schema, accessible without a systems team involved every time someone needs to look at it differently.
If the answer is no, that's the project. Not the AI pilot. Not the vendor evaluation. Not the chatbot RFP. The data architecture first — then the intelligence layer on top of it.
This is uncomfortable advice for an industry that has been sold AI as something you buy and deploy. A product you implement in Q3, report on in Q4, and move on from in Q1. But the implementations that produce measurable results share a common foundation: someone did the unglamorous work of getting the data right before anyone wrote a single prompt.
The technology is an accelerator. It makes good processes faster. It doesn't fix broken ones.
Studio Oriente · Keynote, FITUR TechY 2026
That's the argument I made in Madrid. It didn't require a new tool, a new vendor, or a new budget line. It required a different order of operations — and the discipline to follow it even when the demo looks impressive and the pilot timeline looks short.