Hotels Must Meter AI Costs Per Guest, Hospitality Veteran Warns

Hotel AI costs scale with every guest interaction, and most 2027 budgets are not built for that. Terence Ronson, Founder and Managing Director of Pertlink Limited, says the industry's familiar budgeting logic — licenses, renewals, predictable capital cycles — no longer holds.

Ronson, who chairs the HFTP AI Collective's AI Education & Training Subcommittee and has spent more than 25 years opening hotels across Asia, laid out his framework in a recent interview. His central claim is blunt. AI does not behave like software a hotel owns. It behaves like a utility that bills by consumption.

What changes in the 2027 budget?

Ronson proposes a metric he developed called Token Cost Per Guest (TCPG). It measures the AI cost of serving one guest across the entire journey, from the first pre-arrival question to the reply to a post-stay review.

"AI doesn't come with a subscription fee. It comes with a meter," he says. "And that meter runs every time a guest says hello. If you can't tell your owner what AI costs per guest, you don't have an AI budget. You have a hope."

He also argues annual budgeting is too slow. Payroll, capital projects, and contracts stay stable across twelve months. AI consumption does not. Usage can rise with guest volumes, adoption rates, and vendor pricing changes an annual plan will miss.

"Keep the annual budget for payroll and capex. Put AI on a rolling monthly or quarterly forecast, with a clear trigger for when usage, and cost, outruns the plan," Ronson says. "Flexibility without a trigger is just drift."

His prescriptions for hotel leaders break into five points:

• Budget AI as a metered operating cost, tracked per guest.
• Fund AI literacy as seriously as the technology itself.
• Move AI spend to a rolling forecast, not an annual one.
• Judge technology on efficiency and freed staff time, not adoption counts.
• Treat governance and data quality as competitive differentiators.

Why does the workforce line matter?

The largest barrier to AI adoption is people, not technology, according to Ronson. Many organizations still treat training, role-specific guidance, and change management as optional line items. Revenue managers, housekeeping supervisors, engineers, HR teams, and general managers will all work with AI-enabled workflows.

He recommends ring-fencing a workforce line inside every AI budget rather than funding training from whatever is left over.

"Buying AI without training your people is like installing a grand piano in a hotel with no pianist," he says. "It looks impressive in the lobby. Nobody gets any music."

External forces belong in the 2027 budget too. Guests increasingly begin trips inside AI assistants that shortlist hotels before anyone clicks. Properties that budget only for traditional search and OTA visibility risk becoming invisible. Across Asia-Pacific, fragmenting data-localization rules also shape where guest data can be stored and which AI vendors a hotel group can use. Ronson frames these as budget decisions as much as legal ones.

"The booking used to be the battle. Now the battle is the context before the booking," he says.

How should operators measure value?

Technology projects are typically judged on implementation milestones, adoption rates, or features delivered. Ronson calls those measures insufficient.

"A busy chatbot is not a profitable chatbot," he says. "The most common ROI mistake I see is counting the conversations and never counting the cost of each one, or what happened to the guest afterwards."

His Human Experience Orchestrator (HXO) model asks two questions. Does the investment improve efficiency or reduce the cost of serving guests? And does it create space for meaningful human interaction?

He urges leaders to separate three outcomes that business cases often blur: efficiency, meaning the same work done faster; cost saving, meaning spend removed; and value creation, meaning revenue or loyalty that did not exist before. Each needs its own measure.

On sequencing, his advice is to start behind the scenes, where data is cleaner, guest risk is lower, and returns are easier to prove. "Boring AI pays for exciting AI," he says. "Get the back of house working first, then use what it saves to fund what guests will actually see."

What does governance require?

Ronson points to two court rulings as evidence that governance can no longer sit in an afterthought. In Moffatt v Air Canada, the airline was held liable for incorrect refund guidance from its website chatbot. A Munich court ruled on inaccurate AI-generated information about a hotel.

He recommends two tools that belong in the budget, not just the policy manual. A Decision Envelope documents what an AI system may decide alone, what it must escalate, and what it may never do. An AI Override Rate tracks how often staff correct or reverse AI decisions.

"If nobody ever overrides the AI, either it's perfect or nobody's watching," he says.

Data quality sits underneath all of it. Fragmented or inconsistent data undermines every AI investment built on top of it. What the framework does not provide is benchmark figures; Ronson offers no target TCPG range or industry cost baseline, so operators will need to build their own baselines.

"None of these are glamorous," he concedes of his three budget priorities. "All of them are the difference between an AI strategy and an AI expense."

By 2030, he expects access to AI to be standardized — similar models, similar tools across competitors. What will not be commoditized is organizational judgment, operational discipline, and guest experience. As he puts it: "The intelligence may be artificial. But the experience is human."