98% of Hoteliers Use AI, Yet 30% of Guests Check In at Kiosks

Ninety-eight percent of hoteliers used AI in operations during the past six months, according to Mews's Hotelier Survey 2026, which covered more than 500 properties between December 2025 and March 2026. Yet at US hotels with Mews kiosks installed, only 30% of reservations are checked in at the kiosk, against 20% globally, with another 10.5% of US reservations arriving through the online guest portal.

Those figures anchor a new Pertlink Viewpoint paper, "Front Desk Manning v2.0," which argues that hotel staffing guides still count transactions while the transactions quietly leave the desk. The author, who acknowledges the numbers are one vendor's view from self-selected properties, draws a sharp conclusion: stated preference is not behaviour. Seventy percent of US travellers say they would check in by app or kiosk; roughly 30% actually do at Mews kiosk hotels.

What does the guest actually want?

The paper assembles three separate Mews surveys — US travellers in 2025, 3,250 travellers across eight markets in 2026, and the 500-plus-property hotelier study — and concedes they are not like-for-like. Together they point one way: guests want to skip the queue, not the person. In the July 2026 traveller survey, 89% wanted either a hybrid stay or AI working quietly in the background, and almost nobody wanted an AI-led stay. Sixty-six percent preferred a person for the front-desk welcome.

Two cautions follow. Some guests cannot use digital check-in at all, for reasons of age, sight, dexterity or language; for them the desk is access, not preference. And 67% of guests would worry about staff losing jobs to AI. Stripping the desk, the author writes, is a brand decision, not merely a labour one.

Who pays, and what replaces what?

The labour backdrop is bleak. AHLA's February 2026 survey of 246 hoteliers found more than half reporting understaffed properties despite 70% offering higher wages. Japan's 2026 tourism white paper found 72.2% of 522 lodging facilities short of labour. In the Philippines, the Philippine Hotel Owners Association estimates more than 55,000 direct jobs will be needed for 158 pipeline accommodation projects.

On the technology side, Accor launched ALL Concierge across app, website, WhatsApp and iMessage in September, after a pilot drawing on more than a million guest conversations, with a staff-facing companion, The Butler. Yanolja is rolling out an AI concierge in Asia built on eight specialised agents. Canary Technologies analysed more than 10 million AI-handled guest calls, texts and webchats across thousands of hotels in over 100 countries — a vendor's view of its own platform, as the author notes.

The economics of keeping the hours

The paper's worked example is a 250-room hotel at 80% occupancy with 100 daily arrivals and 30% self-check-in. Freed agent time totals roughly 150 minutes a day, about 900 hours a year, arriving in five-minute pieces rather than removable shifts. Keeping those hours on the floor costs roughly US$20,000 a year at a loaded wage of US$22 an hour. Against a Cornell-linked estimate that a 1% reputation gain associates with up to 1.42% higher RevPAR — worth about US$124,000 on US$8.76 million in rooms revenue — break-even needs only a fraction of a percent.

Every caveat is stated plainly. The Cornell figures are associations from 2012 and 2006 studies, not guarantees. The wage is a placeholder. The chain from freed hours to reputation is an illustration, not a forecast.

How the desk would change

The prescription is concrete. Hire for five skills the paper groups under "HXO potential" — behavioural intelligence, interpretive AI literacy, situational awareness, discretion engineering, narrative crafting — using structured role-plays scored against behavioural anchors. Give desk agents a written Decision Envelope, set by the GM and Finance, covering room moves, late check-outs to 14:00, and recovery gestures up to 15% of the night's rate. Log every AI override as an AI Override Rate. Mews found 41% of hoteliers have no formal AI policy; where one exists, 92% report strong trust in their AI tools, against 49% where none does.

Schools share the load. Research from Regent's University London found 80% of hospitality recruiters had passed over graduates who were not work-ready. Cornell, Glion, PolyU and Les Roches now teach AI courses, mostly to managers rather than the front line.

The paper's closing test is whether guests begin a sentence the machine cannot finish: "there was this person, and they just—" Whether operators reinvest freed hours before cutting them, as the author urges, will decide if that sentence survives the decade.