The number that should
make you uncomfortable
The Stanford HAI AI Index Report for 2025 contains a figure that every hospitality executive should sit with: 88% of organisations have now adopted AI in at least one business function. It is the highest number the Index has ever recorded. It is also, on closer inspection, almost entirely misleading as a measure of progress.
Because the same report finds that agent deployment — AI systems that take autonomous action rather than simply generating text for a human to evaluate — remains in single digits across nearly every business function measured. Marketing, operations, customer service, finance. Less than 10% in each.
The gap between those two numbers is the most expensive misconception in enterprise AI right now. And in hospitality, where the pressure to show AI adoption is coming from ownership groups, OTA partners, and consulting firms simultaneously, the gap is wider than almost anywhere else.
Adoption, in the way the HAI Index measures it, means an organisation has deployed at least one AI-enabled tool or system somewhere in its operations. It does not mean that tool is producing measurable results. It does not mean the organisation has changed how it works. It does not mean anyone is accountable for the outcome. It means a contract was signed and a system is running.
The 88% figure tells you about procurement. It tells you almost nothing about value.
What jagged intelligence
means for your team
One of the more useful frameworks to emerge from recent AI research is the concept of “jagged intelligence” — the observation that AI systems are dramatically capable in some tasks and dramatically incapable in others, with no obvious pattern that maps to human intuitions about difficulty.
A model that can draft a compelling hotel description in seconds may fail completely at counting the number of rooms in a property description it has just written. A system that can synthesise a revenue briefing from a complex data export may hallucinate a competitor’s rate that was never in the data. The capability profile is jagged: peaks and troughs that don’t follow the gradient you’d expect.
The capability profile is jagged — peaks and troughs that don’t follow the gradient you’d expect.
This matters enormously for hospitality operations, where the consequences of AI errors vary wildly by function. A badly drafted email to a guest is recoverable. A mispriced rate loaded into the PMS because an AI misread a competitor scrape is not. A maintenance request routed to the wrong team because a classification model got confused is a guest experience failure. An incorrect allergy flag in a F&B system is a liability event.
The jagged intelligence problem means you cannot evaluate an AI system in isolation from the operational context it is working in. You have to map the capability profile against the failure tolerance of each function. Where the troughs in the profile meet the high-consequence functions, you need human oversight in the loop. This is not a temporary workaround until models get better. It is the correct architecture, right now.
Why the agent gap
is the real story
The distinction between AI tools and AI agents is not semantic. It is the difference between a system that assists a human decision and a system that makes or executes a decision autonomously. It is the difference between a revenue briefing that a manager reads and a pricing adjustment that fires without anyone reviewing it.
The HAI Index’s finding that agent deployment remains below 10% across business functions is not a sign that organisations are being slow. In most cases it is a sign that they are being appropriately cautious. Agent deployment requires three things that most hospitality organisations do not yet have: clean data architecture, clear accountability structures, and explicit failure protocols.
Without clean data, an agent reasons against incomplete or corrupted inputs and produces actions that are confidently wrong. Without clear accountability, nobody knows who is responsible when the agent does something it shouldn’t. Without explicit failure protocols, the system has no defined behaviour for the edge cases it will inevitably encounter — and in hospitality, edge cases are not edge cases, they are Tuesday.
The 88% adoption figure obscures this completely. An organisation that has deployed a chatbot on its website, a copywriting assistant for its marketing team, and an AI scheduling tool in HR has “adopted AI in three functions.” It has also done almost nothing to prepare for the agentic layer that will define competitive advantage over the next five years.
The governance gap
nobody is talking about
The HAI Index also surfaces a finding that receives far less attention than the adoption numbers: the majority of organisations that have deployed AI have not yet established formal governance frameworks for it. No defined policies for what AI can and cannot decide autonomously. No audit trails for AI-influenced decisions. No clear process for when a human overrides the system and what that override gets logged as.
In hospitality, this governance gap has specific consequences. Revenue management decisions influenced by AI that turn out to be wrong have no clear attribution — was it the model, the data, the parameters someone set six months ago, or the manager who approved the recommendation without checking? Guest communications generated by AI that cause problems leave no accountability trail. Predictive maintenance flags that were ignored because nobody trusted the system leave no record of the decision not to act.
Adoption without governance is not progress. It is liability without accountability.
Governance is not the same as restriction. It is not about slowing down AI deployment or adding bureaucratic layers to every decision. It is about knowing, at any moment, what the AI is doing, why it is doing it, and who is accountable for the outcome. Without that, adoption produces liability without accountability — which is a worse position than not having adopted at all.
What the Index means
for hospitality specifically
Hospitality has three characteristics that make the 88% gap particularly significant.
High human contact density. The value proposition of a hotel is a human experience. Every AI application that touches the guest interface needs to be evaluated not just for efficiency but for what it does to the quality of that experience. The jagged capability problem matters most here: the tasks where AI performs poorly — genuine empathy, reading an ambiguous situation, handling the unexpected — are often the tasks that determine whether a guest returns.
Fragmented data infrastructure. Most hotel operations run on five to eight systems that do not share a common data layer. AI applied to any one of those systems produces a local optimisation against a partial view of the business. The agent deployment gap is partly explained by this fragmentation: you cannot deploy a revenue management agent that reasons across channel, rate, and inventory data simultaneously when those three things live in three different systems that don’t speak to each other.
Compressed decision cycles. Revenue management decisions in a hotel happen daily. Housekeeping assignment happens hourly. Guest communication happens in real time. These compressed cycles mean that AI errors propagate faster than in most business contexts, and the window for human correction is narrow. Governance frameworks need to be designed for this operational tempo, not imported from enterprise contexts where decisions move more slowly.
The question worth
asking now
The right question for a hotel group to ask in 2026 is not “have we adopted AI?” Almost certainly you have. The right questions are more specific and more demanding.
Which AI applications are producing measurable outcomes, and what are those outcomes? Which are running without a clear accountability structure? Where in the operation is AI operating in functions with low failure tolerance and no human review loop? What would it take to move from the tool layer to the agent layer in one specific function — and do you have the data architecture and governance framework to do it safely?
The 88% number will keep rising. It is not the number that matters. The number that matters is how many of those deployments are producing results that show up somewhere other than a slide about AI adoption.
If this was useful,
the next conversation is here.
Studio Oriente works with hotel groups and tourism companies at the point where AI strategy meets real operations. No pitch deck required.
Let’s talk →