The Human Touch Has No Audit Trail

Open a hospitality feed this month, and you will find some version of the same reassurance, dressed in different house colors: artificial intelligence handles the logistics, so that people can handle the feeling. The machine informs. The human decides. Give colleagues back their time, and they will spend it on empathy.

It is a good sentence. It appears, in close paraphrase, on brand blogs, in investor decks, and on conference stages across the industry this year. It is also, on its own, true and increasingly insufficient — and the people who build the systems underneath it were the first to say so.

The industry’s own technologists moved first

At HITEC 2026 in San Antonio, the loudest theme on the show floor was not “AI frees humans for connection.” It was governance. With hotels running dozens, soon hundreds, of AI agents across revenue management, marketing, reservations, and guest communications, the operative questions had shifted: which model is doing this, what guest data can it reach, what is it authorized to act on, what happens when it fails, and can anyone actually produce an audit trail? A new software category — the Agent Management Platform — exists purely to answer those questions at scale. None of that discussion was about replacing judgment. All of it assumed judgment needs a record, not just an owner.

That is the gap in the “human touch” narrative as it is usually told: it names the safeguard — a person — without describing what makes that person’s judgment visible, checkable, or governed. “Trust our people” and “trust our controls” are not the same claim, and boards should stop accepting the first as proof of the second.

Where the sentiment runs out of runway

Three places this shows up most often:

• The number nobody can trace. Productivity claims in these pieces routinely lean on “recent research” without a named report, sample size, or methodology — a statistic that reads well and checks badly. If a brand is asking guests to trust it with their data, and won’t show its own readers where a headline number came from, that is a small but telling signal about how rigorously the larger “responsible AI” claim has actually been examined.
• Agency described only in benefit terms. Agentic AI is increasingly framed as a bigger, better insight engine — spotting patterns, informing decisions, freeing up capacity — with no mention of what happens when it is wrong, who reviews it, or what it is not permitted to do on its own. That is not a hypothetical risk. IBM’s 2026 Cost of a Data Breach Report puts the global average cost of a breach at $4.99 million, up 12 percent on the year; breaches involving AI tools were up 56 percent and cost roughly $1 million more than the average. “AI helps us know our guests better” gets considerably more expensive the moment nobody can say who is watching the AI that knows them.
• Personalization told as a service story, not a data story. The genre’s favorite anecdote — recognizing a guest’s preferences across three cities and two continents — is a compliance question in hospitality costume. Guest data moving cleanly between jurisdictions is precisely the assumption that Asia’s fragmenting digital trade rules are undoing in real time. “Responsibly” is doing an enormous amount of unexamined work in that sentence.

Judgment needs an operating model, not a headline

None of this argues against the underlying instinct. People genuinely are the differentiator in luxury hospitality, and no model deserves credit for a florist’s decades of noticing what a returning guest doesn’t say out loud. The argument is narrower: that instinct only survives contact with agentic AI if it is designed in, not assumed.

This is the operating premise behind the Human Experience Orchestrator model — HXO: the human, not the system, owns the moment, but that ownership is structured rather than sentimental. A defined Decision Envelope sets out what an AI agent may action unsupervised, what requires a colleague’s sign-off, and what should never be automated at all. An AI Override Rate — how often, and why, a colleague overrules the system — gets reported to the board with the same seriousness as occupancy or RevPAR. “Our people make it feel personal” stops being a slogan and becomes a control that can be audited, benchmarked, and defended to a regulator, an insurer, or a guest who asks how their data is actually being used.

The same discipline should apply to a word the industry treats as an unqualified virtue: seamless. Removing friction from a guest journey is usually presented as pure gain — fewer steps, less waiting, nothing for the guest to notice. But friction is often where the checkpoint lives: the disclosure before a rate changes, the confirmation before a preference gets acted on, the pause where someone could say no. A guest at a Manhattan hotel near Times Square learned this in 2026, when an automated rate change more than doubled his nightly cost mid-stay — and the detail that went viral was that staff had “no authorization to do an override” to honor what they had told him. A Decision Envelope should specify not only what an AI agent may do unsupervised, but also which points of friction are load-bearing and remain in place deliberately, even when removing them would make the journey feel smoother.

The sentence is worth defending

Pertlink’s own shorthand for this is that the intelligence may be artificial, but the experience is human. It is a sentence worth defending — but sentences don’t survive scrutiny; systems do. The industry doesn’t need to abandon the human-touch story it keeps telling. It needs to be able to show its working.

Terence Ronson

Terence Ronson is the Founder and Managing Director of Pertlink Limited, a boutique hospitality technology and AI advisory consultancy headquartered in Hong Kong since 2000, operating across the Philippines and Asia-Pacific. He chairs the AI Education & Training Subcommittee of the HFTP AI Collective and is an inductee of both the HFTP International Hospitality Technology Hall of Fame and the CHTA Hall of Fame.