AI Risks in Hotels: What Can Go Wrong, and What Good AI Looks Like
An honest look at the real risks of AI in hotel operations, from black-box pricing to alert fatigue, and what traceable, human-in-the-loop AI should look like.
Last month at HITEC in San Antonio (June 15–18, 2026), "The Dark Side of AI: Ethical Dilemmas and the Disasters Heading Our Way" was a popular session on the schedule. The fact that a session with that title is on the agenda at the industry's biggest tech conference tells you something: hoteliers are not naive about AI. They have questions. They have concerns. Some of them have already been burned.
Those concerns are legitimate. AI vendors who refuse to engage with them honestly are part of the problem. Below are the five most common ways AI goes wrong in hotels, and what the right response to each one looks like.
What Are the Real Risks of AI in Hotel Operations?
The risks of AI in hotel operations fall into five categories: recommendations acted on without visible reasoning, black-box pricing that cannot be explained to owners, staff losing institutional knowledge as AI takes over workflows, alert fatigue from systems that surface everything, and vendors overpromising capabilities that do not yet exist.
Each risk is real. Each one is avoidable.
Risk 1: AI Recommendations Acted On Without Anyone Understanding Why
This is the failure mode that costs hotels the most money.
A revenue manager gets an alert: “Drop your rate on Thursday by €15”. They are busy. The system has been right before. They approve it without reviewing the reasoning. Thursday turns out to be a compression night; a major event in the city that the AI missed because it was pulling from the wrong data source. The rate drops. Revenue is lost. When the owner asks why, nobody can explain it.
This is not a hypothetical. Versions of this happen in hotels regularly. An AI makes a recommendation. Someone acts on it. Nobody tracked the logic. Something goes wrong. Now nobody can reconstruct what happened or why.
The failure is not the AI making a mistake. The failure is a system that gave no visibility into its reasoning, so there was no moment where a human could have caught the error before it became a cost.
The antidote is not less AI. It is AI that shows its work.
Every Otel recommendation includes the source data it drew on, the reasoning behind the suggestion, and the available options, visible in plain language before you approve anything. The human stays in the loop not because we don’t trust the AI, but because the human has context the AI does not: the rugby match that was not in the data, the group that always cancels at 30 days, the owner's preference for rate integrity over occupancy.
Risk 2: Black-Box Pricing That Nobody Can Explain to an Owner
Here is a conversation that happens in hotels more often than anyone admits.
- An owner looks at the monthly numbers: why did we drop our rate on that weekend?
- The revenue manager says: the system recommended it.
- The owner asks: why?
- The revenue manager says: I'm not sure exactly, it takes a lot of factors into account.
That is not a defensible position. In a high-stakes ownership conversation, it is a career-limiting one.
AI pricing recommendations are only as useful as the team's ability to explain and defend them. If staff cannot articulate why a rate changed, what data drove the decision, what assumption the AI was making, or what alternatives it considered, that is not AI-assisted pricing. That is pricing by black box.
This risk is highest in multi-property environments and ownership structures where the revenue manager is not the final decision-maker. Every AI recommendation touching rate, inventory, or revenue strategy must be explainable in plain language to someone who was not in the room when it was made.
The fix is traceability built in from the start, not bolted on as a feature. Source, timestamp, reasoning, options, all visible on every recommendation. Not because operators will challenge every suggestion, but because the day they need to explain a decision to an owner, the answer should be one click away.
Risk 3: Staff Losing Institutional Knowledge as AI Takes Over Workflows
This risk is slower to emerge, but potentially more damaging than the others.
When AI starts handling the daily briefing, pickup analysis, and rate alerts, the routines revenue managers and GMs used to do manually, institutional knowledge quietly fades. The revenue manager who spent 45 minutes every morning pulling pickup reports developed, over years, an intuitive feel for what the data meant: what a 15% pace-down on a Tuesday looked like versus the same number on a Friday, which segments reliably washed, how the corporate calendar shifted the leisure mix. That knowledge was built through repetition.
When AI does the work instead, that expertise can atrophy. New team members never develop it. When the AI gets something wrong, or faces a situation it was not designed for, the human safety net is thinner than it was.
There is also a more immediate version of this risk: staff turnover. When a revenue manager leaves, they take their knowledge with them. If that knowledge exists only in their head, and not captured in the logic of how decisions were made, the hotel starts from scratch.
AI that documents its reasoning creates an institutional memory that does not walk out the door. The goal is a team that gets sharper over time, not one that becomes dependent on a system it no longer understands.
Risk 4: Alert Fatigue and the Recommendations Nobody Reads
One of our customers described receiving so many alerts from their existing systems that they’d started ignoring them by default.
"I get forty notifications a day. I stopped reading them. It's just noise."
This is a real and underappreciated failure mode. The promise of AI is that it surfaces what matters. If it surfaces everything, every minor variance, every small deviation from forecast, it becomes a different kind of noise, and operators tune it out.
An alert that cries wolf damages trust in all the alerts that follow, including the ones that matter.
AI that surfaces everything is not better than AI that surfaces nothing, it just distributes the problem differently. The right answer is alerts calibrated to what is genuinely decision-relevant, delivered with enough context that a GM can evaluate in ten seconds whether action is required.
Risk 5: Vendors Overpromising Capabilities That Do Not Exist Yet
Be sceptical of AI companies that promise outcomes they cannot demonstrate today.
The hospitality AI market is moving fast. Some claims, e.g., about automation rates, revenue uplift, time saved, are aspirational rather than evidenced. Not all of them are dishonest. Sometimes it is genuine enthusiasm about what the technology will eventually do. But hotels are being asked to make buying decisions now, and they deserve to know what is real today versus what is on a roadmap.
Our rule: present tense is for features that exist in the product today. Future tense is for things we are building toward.
The numbers we cite are specific, traceable, and named. That is the standard we hold ourselves to. If a vendor cannot point to outcomes like this, named, specific, and traceable, ask more questions before you sign.
What Good AI in Hotels Actually Looks Like
AI in hotels is not dangerous because it is powerful. It is risky when it is opaque: when recommendations have no visible reasoning, when alerts have no context, when decisions are made that nobody can explain or learn from.
The antidote is not less AI. Hotels that pull back from AI because of these risks will fall behind properties that figured out how to use it well. The RevPAR gap between operators with AI-assisted capacity and those without is already widening.
The antidote is traceable AI. AI that shows its work. AI that keeps humans in the loop not as a regulatory concession, but because human judgment, i.e., knowledge of the market, the guests, the team, is genuinely irreplaceable.
The AI handles the briefing. You make the call.
Continue reading
Related articles
Where Agentic AI is Actually Earning its ROI in Hotels
Not guest chatbots. Not autonomy for its own sake. Otel's Paul Ryan on where agentic AI is quietly earning its ROI in hotels right now.
Paul Ryan·15 Jul 2026
What I Saw at HITEC 2026 That Nobody Else Is Talking About
Most startups at HITEC were building guest-facing AI. The problem eating the first two hours of every GM's morning went almost unaddressed. Here's what Otel AI CEO Paul Ryan saw.
Paul Ryan·2 Jul 2026
Valuemaxxing. Hotels have always done it. They just called it Monday morning.
What does AI value look like when the margin for waste is zero? For hotel teams, the answer has always been the same: did the work get done?
Paul Ryan·17 Jun 2026