Hotel AI is only as good as the data it’s built upon

Hoteliers have told us for years that they have huge amounts of data, but in separate systems. The core issue is that each tool runs on its own data set and doesn't play nice with others. Your PMS is working with one set of numbers and your channel manager with another.

That's partly because connecting these systems is nobody's job. Every vendor solves their own puzzle, and inside the hotel no one is tasked with joining them up.

For a long time, connecting these systems was too complicated, but with AI and MCPs, seamlessly integrating solutions through a shared data layer is now possible.

On ownership, revenue managers live in the data and know which numbers make or break a strategy. The role is moving beyond adjusting rates towards a commercial architect who builds frameworks that optimize profit and balances automation with AI guardrails. The same person should define the data foundation and the terms the AI works under.

That said, the task of making data within different systems easy to connect to should sit with vendors. If it’s difficult to reach your own data, it's worth asking whether you've got the right tech partner.

The general malaise that the industry maintains when it comes to doing the hard work of technology and data structure.

This is the one initiative that is almost always pushed down the road. Largely because people hope someone else will do it. Secondly, because a secret hope exists that it will somehow magically sort itself out. We are in another one of those cycles now.

AI, like every other technology preceding it, is based upon data. As the old trope goes, garbage in - garbage out.

Who in the organisation should own it? The entire organisation, stemming from a realisation that all organisations are a body of information allowing it to operate and make informed decisions to move forward.

Who is responsible for it? The entire organisation and all aspects of the business contribute information.

Who should guide and manage the structure and quality of information? Technology. Data Architecture is a key responsibility of a technology function. It is one of the layers of overall technology architecture. The information in question for the most part resides in technology. The discipline requires an across organisation and across system view of a consolidated data footprint.

This is where the industry falls down.

I spoke about this topic on HEDNA's stage in Lisbon a couple of weeks back, on how AI is redrawing hotel distribution. Knowing that 79% of what AI knows about a hotel comes from OTAs, metasearch, and editorial sources — and not from the hotel itself, which only accounts for 7.5% — I believe we can agree that AI is filling the gap with third-party data. This is not a technology failure, it's an accountability one. Rate, inventory, and content live in systems that were never built to talk to each other: PMS, CRS, CMS, channel manager... and no single function owns whether that data is complete and machine-readable across all of them. IT owns pipes, marketing owns content, revenue owns rates, distribution owns channels. Nobody owns "AI-ready."

My view, stated plainly in Lisbon: this belongs with commercial leadership, because AI visibility is just another distribution channel needing the same rigor as OTAs (parity, structure, consistency) plus a real data steward to enforce it.

Hotels do not have a data problem. They have a "where does the answer live" problem. The rate rules are in the PMS, the dog policy is in the inbox, the early check-in answer is in the head of the most senior receptionist. AI stalls because no one has written that knowledge down in one place the systems can read.

The fix is not a bigger integration project. It is starting with one channel, letting real guest questions expose the gaps, and filling them one answer at a time. That is how a foundation gets built in weeks rather than years.

Who owns it: the front office manager. They already know which answers are wrong.

Asking what AI actually needs is also asking what the hotel implementing it needs. As various generative AI systems become increasingly more potent, concerns are also being raised about the data they have been built upon and possible copyright violations in that process.

In hospitality, accommodating our guests is the most important goal, and knowing about them and their preferences would enable the agentic AI systems to meet their needs. But to find out about guest preferences means that their data needs to be obtained in some way, and this poses a complex dilemma where the ideal AI system for catering to guests might require extensive access to data they would rather not share with hotels.

Although this might not stop hotels today, it is very important to take into account and make sure that ultimately guest should have a say about who uses their data. It would be unhospitable to have guests check out of hotels, but their data not being able to leave...

From an AI visibility perspective, AI is only as good at recommending your hotel as the data you feed it. Garbage in, garbage out (GIGO).

In SEO era, a major problem was it was hard to "steal" authority signals (PageRank) from OTAs as they do not allow outbound links to hotel sites. In AI era, it's no longer a problem to "steal" trust signals from them. Hotels can leverage OTAs’ authentic (verified), dense reviews to build a strong reputation for better AI visibility - an extension of “billboard effect.”

The sources that AI learns a hotel are mostly certain. They’re Google Hotels, hotel websites, Tripadvisor, social media, OTAs, PR. The predictive weights of the sources are uncertain, due to the opaque nature of neural networks. Hotels thus need a multichannel strategy going beyond siloed revenue and profit optimization to build balanced data density across the sources. The concentration risk for discoverability is only getting bigger in AI era.

Hotels should understand that reviews are more than WOM in AI era. They're registered, timestamped, trusted CONTENT ready to be ingested by AI at inference. Your website can’t realistically and timely cover all micro intents travelers ask, but thousands of reviews can.

Start by defining what "AI" actually means. Either it's a helper (respond to this review) or it's a conversational system. And the dominant use case in hotel operations is the second: powering every conversation between a guest and a hotel. That conversation starts at booking, runs through stay and check-out, and continues into the marketing that brings the guest back.

The data problem is really a stack problem. Hotels still think in the old tech-stack categories, PMS here, CRM there, feedback somewhere else, and each one owns a fragment of the same conversation. What's needed instead is a stack that enables one seamless conversation from start to end, where outbound and inbound messages always meet in the same place, and where the memory of the conversation is always kept. We don't yet have a word for that system. But it's the system to strive for.

That's also the answer to ownership. Because it touches every silo in today's hotel org, from reservations and front office to marketing, revenue and IT, no single department can own it. It either sits with a genuinely cross-functional team, or it needs C-level or founder authority. Anything less, and the silos win by default.

Hotel AI stalls because most properties never agreed on what their own data means. A PMS, an RMS, a CRM and a POS each define key terms like ‘guest’ or ‘occupancy’ slightly differently. These then get stitched together with integrations that move data around without reconciling it. That's a fundamental language problem, and no AI model reasons its way past it. 

The fix is a semantic layer: a shared layer of meaning under every system, so the entire business runs on the same definitions and understanding. This is the core tech infrastructure of a hotel – built once and shared by every system and agent – not an afterthought bolted onto whichever tool was bought last. 

Ownership should sit with one accountable person, not scattered across IT, revenue and operations. This person is increasingly a technology leader with a mandate across the whole property, not just IT budgets. They are the one who defines and maintains the shared language, so everything works from the same picture. 

Get that right and AI stops guessing. It starts making the same calls as an experienced GM, and it frees people to spend time in areas that truly benefit from the human touch. 

The biggest obstacle is not the lack of data, but the lack of a structured, centralized and continuously updated way to manage it. Hotel information still lives across multiple systems — and often in people’s heads, spreadsheets or documents — making it difficult to maintain a reliable single source of truth.

But centralization cannot mean creating another system that hotel teams must constantly maintain. Technology should automatically search, collect, structure and update knowledge from the hotel’s different sources, involving hotel staff only when it cannot find or confidently determine the answer. This dramatically reduces the operational cost of keeping knowledge accurate.

When human input is needed, ownership should sit close to operations. In our experience, roughly half of guest questions are about the hotel itself: facilities, services, policies or local points of interest. Marketing, distribution or IT teams are often not close enough to the day-to-day operation to answer them accurately.

And the conversation is already moving beyond data. AI assistants and agents need access not only to accurate information, but also to capabilities: checking availability and rates, making, modifying or cancelling reservations, and eventually executing other hotel services.

The first AI challenge was making hotels understandable. The next is making them actionable.

Two things stop hotels.

The first is the vendors. Most hotel systems still won't give open API access which causes fragmented data, its not the hotel's fault. AI works well on a single system; it stalls when you try to connect PMS, revenue, payroll and F&B into one view. The industry's own leaders can't even agree who owns the data and this is a big part of the problem.

The second is capacity, budget and ownership.

Even if you solve these two there's still a problem nobody talks about enough: perfectly connected data doesn't get you 100% there. AI doesn't know you count no-shows in on-the-books revenue, or that last February's refurbishment moved you up the comp set. Context is the last mile, and only the hotel can supply it. Without it, you get generic AI slop.

Who should own it? If you have internal IT or CFO who oversees IT, it's them. What matters is it's one person, with a budget, and it's their job and not something the GM picks up when they've a spare hour.

Hotel AI is only as good as the data it is built upon. I agree, but good data is not the whole foundation.

For an independent hotel, the first question is not which AI tool to buy. Ask instead: where does guest information get lost? Are reservations, guest preferences, and team observations kept in separate places? Who checks accuracy and acts on it? AI scales confusion, not hospitality.

This is why I start with readiness. A hotel needs infrastructure that connects the right information, people who trust and understand it, and a culture that gives the team room to use it. Data should help a front-desk colleague recognize what matters to a guest, not reduce that guest to a profile. The team and customers must remain part of the conversation.

I see AI as a cultural transformation for hotels. Independent properties can take back lost ground and find overlooked opportunities. That is the Blue Ocean opportunity: not copying another hotel's technology stack, but using what makes your property distinct to create hospitality guests cannot get elsewhere.

Clean data matters. Knowing what to do with it, and bringing your people along, matters just as much.

What stops them? Technical debt in vendors’ systems, overcomplicated hotel setups and insufficient staff training.

Who should own it? The hotel’s technology department, alongside operational teams.

Ownership starts with understanding the full workflow and questioning whether its complexity is necessary. Does the hotel need hundreds of rate codes? Have staff been trained to use the PMS properly?

But internal ownership does not make hotels responsible for every underlying problem. We, as vendors, got us here.

Too often, sales and new features have taken priority over architecture. Even connected systems can leave gaps: a reservation may reach the PMS without the travel agency information needed for reporting.

We call these “legacy” products and integrations, but that label can obscure years of accumulated technical debt. Vendors have built workarounds, passing the cost on to customers and partners. Some have modernised successfully. Where they have not, hotels are paying the price.

Hotels must simplify workflows and train their teams. Vendors must build reliable, robust systems that integrate easily and keep data consistent. Without both, a cleanup is temporary.

When we ask whether hotel data is ready for AI, should we also ask whether our systems were built to keep it ready?

Hotels do have a data problem, but I don't think the answer is necessarily to create one enormous source of truth before AI can deliver value.

What AI actually needs is access to the right context at the moment a decision is made.

Take a reservation inquiry. To give a genuinely good answer, an AI agent might need live availability and rates from the PMS, the guest's previous stays and preferences, the content of the current conversation, hotel policies, and perhaps information about packages, restaurants or spa availability.

Today, that information often sits in different systems.

The biggest obstacle is therefore not just poor data quality. It is that hotel technology has historically been designed as a collection of isolated applications rather than as an environment where intelligence can operate between them.

I believe the next step is to make hotel data accessible, structured and actionable in real time, while keeping clear permissions and ownership around each system.

And ownership can't only be with IT. If AI is making decisions that affect conversion, pricing, guest experience and operations, the data foundation becomes a business responsibility too.

What stops hotels from building the data foundation AI actually needs is the assumption that a data foundation means replacing systems. When treated as a major capital program, the effort waits for a budget cycle, deferring AI progress along with it. Hotels that move faster connect and activate the systems they already have. 

The business owns it, but the work can't sit on one side of the house. Put it solely with IT and the integration may meet the technical spec but not the daily reality of the people using it. Leave it solely with the commercial side and the vision may never become a record other teams can recognize, trust, and use. The business defines what a unified guest record means; tech delivers the integration across systems. Neither can deliver the outcome independently.

The hotels I've seen get this right share this. They established shared business and tech ownership, connected what they had, and started with the guest record. One trusted, usable record that teams could actually act on. Everything else followed.

Fragmentation remains partly because it pays! Hotels bought technology one application at a time, and plenty of vendors found it convenient to keep the hotel's data inside their own walls. Connecting those systems helps, but AI needs more than connected data; it needs a shared understanding of how the hotel works.

So I'd aim less for a single source of truth and more for a single source of meaning. The ontology: an operating model covering the property's data and relationships, plus its rules (including which ones win in a conflict), decision rights, allowable actions, and the agents handling each task. It has to carry both brand standards and the property's own SOPs.

With brands, operators, and owners each holding a piece of the P&L, no one party should own all of it. Brands define their standards, while operators and owners add layers they keep when contracts end (the guest data stays with the hotel). AI that can reason across two models also makes switching partners far less painful.

Technology leadership owns the architecture; GMs and department heads define how their property runs. And don't wait for the whole thing. Start with one decision that crosses departments.