Every week, another hospitality technology vendor announces their new AI agent. The language is consistent: transformative, intelligent, purpose-built for hotels. The demos are compelling. The case studies are carefully selected. And the announcements are, in almost every case, telling only half the story.
NB: This is an article from Sirma Group
Subscribe to our weekly newsletter and stay up to date
The missing half is this: most of these agents live entirely inside the vendor’s own ecosystem. They read that vendor’s data, automate that vendor’s workflows, and speak exclusively to that vendor’s modules. The moment a hotel needs that agent to interact with a different system, a PMS from another provider, a revenue management platform, or an F&B solution, the intelligence stops. The agent hits a wall that it was never designed to cross.
I have spent more than a decade working across the hospitality technology landscape on the product side at Oracle Hospitality, then in commercial roles at NOR1 and Shiji, and then as part of Sciant before it was acquired by Sirma and became the foundation of what is now the Sirma Travel and Hospitality vertical. Across more than 900+ projects and every tier of the industry, I have sat on both sides of this table long enough to recognize a pattern when it is forming. And what is forming right now deserves a more honest conversation than the one we are currently having.
A familiar story wearing a new outfit
For two decades, the hospitality technology industry has operated on a mantra that sounds generous but is usually not: open APIs, open ecosystems, connect with anyone. And technically, that has always been true. You can absolutely connect. But someone has to build the connection. Someone has to maintain it. Someone has to absorb the cost of every API call that passes through it. The openness was always conditionally dependent on engineering capacity, budget, and the willingness of vendors to prioritise interoperability over lock-in.
AI agents have not changed that equation. They have made it more consequential.
When a hotel’s property management system had limited reporting capabilities, the cost was inconvenience. When a hotel’s AI agent cannot access operational data from the three other systems it needs to make a meaningful decision, the cost is the entire promise of AI. A siloed agent is not a partial solution. It is a sophisticated tool that cannot do the one thing hotels actually need: see the whole picture and act on it.
To be fair to the vendors building these products and I say this as someone whose business works alongside them every day the domain intelligence embedded in many of these agents is genuinely impressive. Companies that have spent years accumulating operational data in a specific area of hotel management are well-positioned to build agents that understand that domain deeply. That expertise is real, it has value, and it is not going anywhere. The problem is not what these agents know. The problem is the architecture that contains them and that is a problem vendors and their partners can solve together, if the industry is willing to have the honest conversation first.
Where the real value of AI actually sits
There is a second conversation the industry needs to have, and it runs parallel to the integration problem. Almost every AI application currently being marketed to hotels is guest-facing: intelligent chatbots, personalised recommendations, dynamic upselling at the point of booking. These are visible, demonstrable, and easy to put in a press release. They are not, in my assessment, where the most significant value of AI in hospitality will ultimately be found.
The back office is where AI has the potential to fundamentally change how hotels operate i.e. housekeeping optimisation, predictive maintenance, procurement intelligence, energy management, labour scheduling against demand forecasting. These are areas where AI-driven decisions compound over time, where the financial impact is direct and measurable, and where most hotels are still operating on instinct, spreadsheets, and experience. The unglamorous parts of hotel operations are, for precisely that reason, the most fertile ground for genuine transformation.
The irony is that unlocking this value requires solving the integration problem first. Back-office AI that cannot draw on data from across the property’s full technology stack will produce recommendations that are, at best, partially informed. At worst, they will be confidently wrong.
The model that actually works
I find it useful to think about a hotel’s AI future in terms of two circles. The outside circle is where hospitality has always lived and must continue to live: human-to-human interaction. The front desk agent reading a guest in thirty seconds and adjusting accordingly. The server who understands that the couple at table seven is celebrating something without being told. The housekeeper who has a brief, genuine exchange in the corridor that a guest remembers long after they have forgotten the thread count. This is not under threat from AI. This is, in fact, what hotels are selling. It cannot be automated and it should not be.
The inside circle is where AI orchestration becomes not just useful but necessary. The systems, the workflows, the operational decisions that guests never see this is where a properly architected AI layer can connect data flows across the entire property stack, manage specialised agents from multiple vendors, and produce decisions and automations that no single system could produce on its own.
Orchestration is the operative word. Not one agent. Not a collection of disconnected agents. A layer that sits above the vendor ecosystem, draws on APIs and emerging connectivity standards like MCP servers, and coordinates intelligence across the full technology landscape of a hotel. Critically, this is not a replacement for what vendors are building it is what makes what they are building actually work at the property level. Vendors bring the domain intelligence. Orchestration brings the connective tissue. Both are necessary. Neither is sufficient alone.
The risk nobody is discussing
There is one more dimension to this that the industry is almost entirely silent on: security and compliance. As hotels move guest data through AI pipelines at an accelerating rate, the questions of GDPR compliance, PCI DSS obligations, and data governance are becoming urgent. Larger hotel groups with dedicated legal and technology teams are beginning to ask these questions seriously. Most independent properties and mid-scale operators are not often because their technology vendors are not raising the issue with them.
This is a risk that will not remain theoretical for long. The regulatory environment around AI and data handling is tightening across every major market. Hotels that have made AI adoption decisions without a parallel conversation about data security and compliance are building on foundations that may become legally and operationally untenable.
The window is open, but not indefinitely
None of this is an argument against AI adoption in hospitality. It is an argument for AI adoption with understanding. The properties and groups that approach this as an infrastructure decision, investing in integration architecture, orchestration capability, data governance, and domain expertise alongside individual AI tools, will find that the returns compound in ways that late adopters will struggle to replicate.
The next phase of AI in hospitality will not be defined by experimentation. It will be defined by execution through purpose-built approaches designed to support real workflows across the full operational complexity of a hotel, not just the parts that look good in a product demonstration.
The vendors who will matter most in this next phase are not necessarily the ones with the most sophisticated individual agents. They are the ones with the domain knowledge and the integration depth to act as true enablers helping hotels connect what they have, secure what they move, and extract value from the full complexity of their technology ecosystem, not just one corner of it.
