Only a Tenth of Hospitality Leaders Feel Ready for AI. What the Rest Should Do First.

The survey figure doing the rounds in hospitality this year is that around one in ten industry leaders feels prepared for AI. Nine in ten do not.

That number is usually presented as alarming. It is more useful read as accurate: most leaders have correctly noticed that they do not know what to do first. Here is what to do first.

Why the readiness gap is not technical

The technology is largely solved. A language-model assistant that understands any phrasing, answers in any language, and escalates what it cannot handle is a product you can buy this afternoon. Nobody at a hotel needs to build it.

What leaders are unprepared for is everything around it:

• Who owns what the assistant says, and keeps it current
• Where an escalated request goes, on which channel, and what happens if nobody responds
• What the assistant must never answer automatically
• What to measure, and against what baseline

Those are organisational questions. A vendor demo does not answer them, and a leader who has only seen demos is right to feel unready - they have seen the easy part.

Start by counting

The most useful thing any property can do before buying anything takes two weeks and a sheet of paper.

Record every inbound guest question at the desk: the question in the guest's words, the channel, roughly how long it took, and whether a briefed new starter could have answered it. At the end, sort by frequency.

The result is almost always the same shape. A small number of questions account for most of the volume - studies put roughly 40% of hotel phone inquiries as repetitive basic questions about hours, prices and availability - and nearly all of them fall in the no judgement needed column.

That count is the readiness. It tells you what to automate first, and it is the baseline every later result is measured against.

Automate one thing

Not the whole operation. The single most frequent question from the count.

Handle it properly, end to end, including what happens when the answer is no and when the guest follows up. Then count that question again a month later.

Starting narrow does three things. It is provable - one number before, one after. It is fixable - if the tone is wrong you find out on one question, not forty. And it builds the team's trust, because staff who watched one thing work are more receptive to the second than staff handed a system.

Name the owner

An assistant with no owner drifts out of date within a quarter. The property changes - a new shuttle schedule, a closed pool, a price rise - and nobody thinks to tell the assistant, because no single change feels significant.

The owner is someone close to guests, usually front office, with authority to change an answer without convening anyone. Not IT, who own the plumbing but not the answers. Not marketing, who own the tone but not the operation. At a small property it is one supervisor with an hour a week. The important part is that it is a name, not an assumption.

Decide what it must not say

Before go-live, list the topics the assistant should never answer automatically regardless of confidence. Complaints. Accessibility requests. Anything with legal or financial weight. Medical questions.

The assistant routes those to a person with context. This list is a management decision, it takes twenty minutes, and it is the thing that prevents the one incident that ends a deployment.

Being late is not the problem

The properties that rushed in 2024 mostly bought decision-tree chatbots, watched guests learn to avoid them, and switched them off. Arriving now, with the current generation and a clear idea of what to measure, is not late. It is skipping the expensive mistake.

What is a problem is arriving with no baseline, no owner and no boundaries - because then the assistant works and nobody can tell, or it drifts and nobody notices.

The sequence

Count for two weeks. Automate the top question. Name the owner. List the boundaries. Measure at ninety days against the count.

That is what ready looks like. It is not a technology roadmap and it does not need one. The nine in ten who feel unready are mostly waiting for a plan that turns out to be a fortnight of tallying at the front desk.