AI in Hospitality: What Actually Works on the Floor

I've been brought into three AI implementation projects in the past two years. Two were recovering from failed first attempts. The third was trying to avoid the same fate.

The failed ones had something in common: they started with the technology.

The Typical Arc

A GM or Director of Operations comes back from a conference excited about AI. They've seen a demo — a chatbot that handles guest requests, a revenue tool that reprices rooms, a kitchen display system that reduces ticket times. It looks impressive. They contact the vendor. A pilot is approved.

Three months later, the chatbot is answering about 20% of what guests actually ask. The revenue tool is generating recommendations that the reservations team overrides 70% of the time because "the system doesn't understand our market." The kitchen display is technically running, but two of the five chefs have figured out how to bypass it.

The technology wasn't the problem. The implementation was.

What Gets Skipped

AI tools in hospitality need clean data inputs and consistent human behaviors to work. Most hotel operations have neither.

If your room type descriptions aren't standardized, your revenue AI can't learn effective pricing patterns. If your front desk team uses six different ways to log a guest preference in the PMS, your personalization tool will generate nonsense. If your kitchen team's workflow isn't documented, a display system will just digitize the existing chaos.

Before any AI tool, there's foundational work that nobody wants to pay for because it's not exciting: data hygiene, process standardization, and change management for the team that will actually use it.

This work takes weeks. It's not glamorous. But without it, the AI has nothing solid to stand on.

What I've Seen Actually Work

Automated pre-arrival communication — when it's connected to a real CRM with actual guest history. A hotel that knows a returning guest always requests a high floor and a foam pillow can send a pre-arrival message that feels genuinely personal. The AI drafts it; a human reviews it in thirty seconds. This works because the data foundation (a clean CRM with real history) was already there.

AI-assisted maintenance dispatch — in properties with a solid maintenance ticketing habit. When technicians consistently log their work, an AI tool can genuinely predict which equipment is likely to fail next and help prioritize preventive work. I've seen this reduce emergency calls by about 20% in one property over six months. But it required eighteen months of consistent logging data before the predictions were worth trusting.

Revenue management tools with human override built in — not as an afterthought but as a design principle. The best implementations I've seen treat the AI as a first draft that a revenue manager reviews every morning. The manager learns from it; the tool learns from the overrides. After six months, the gap between AI recommendation and manager decision narrows significantly.

The Honest Assessment

Most properties are not ready for AI in the way vendors promise it will work. That's not a criticism of the technology. It's a description of the gap between where most operations actually are and where they need to be for AI to add value.

The properties that benefit earliest are the ones that already have their house in order: clean data, consistent processes, and teams that are comfortable with digital tools. For them, AI genuinely accelerates what they're already doing well.

For everyone else, the most valuable investment isn't an AI tool. It's the operational foundation that makes AI possible later.

That's less exciting to sell. But it's honest.

If you're considering an AI project and you're not sure whether your operation is ready, that uncertainty is useful information. It usually means the foundation work comes first.