Hotels can’t show what AI actually earned them
Hyatt and peers shift from counting hours saved by AI to finding it on the profit-and-loss sheet.
By The Route · · 5 min read
Hotel companies can tell you exactly what artificial intelligence costs them this quarter. None of them can yet point to a line on the profit-and-loss statement and say: this is what it earned back. That gap, not the technology itself, was the subject at Destination AI in Washington this week, where executives admitted the industry has spent two years measuring the wrong thing.
A front-desk manager's hours saved don't pay her salary
Picture the actual trade a hotel is making. A property installs an AI tool that drafts guest-service replies, routes maintenance tickets, or flags no-show risk before check-in. The general manager reports back: staff save four hours a week. That number goes into a slide deck. It does not go into the budget, because "hours saved" isn't a line item accountants recognize — it has to convert into fewer shifts, higher occupancy, or fewer refunds, and that conversion is the part nobody has finished proving. Pat Nestor, Hyatt's senior vice president of data and AI, put the problem plainly at the Washington event: the early metric was personal efficiency, and personal efficiency "doesn't live on a P&L sheet." A $200-a-night room in Chicago or a $90-a-night room in Tulsa still has to show the saving as either lower cost per occupied room or higher revenue per available room. If it doesn't, the tool was a convenience, not an investment.
From "did we deploy it" to "did we absorb it"
The hotel industry's AI spending followed the same arc as its earlier bets on booking engines and loyalty apps: a land grab first, accounting second. Through 2023 and 2024, chains including Hyatt, Marriott and Choice Hotels rolled out generative-AI pilots for guest messaging, revenue management and back-office scheduling, largely funded as innovation-budget experiments rather than tracked against specific cost centers. The pitch to franchisees and owners was speed — these tools would compress response times and cut the hours staff spent on repetitive tasks, at a moment when hotel labor costs were already running ahead of pre-pandemic levels in most US markets.
What changed by late 2025 was not the technology but the audience asking about it. Franchise owners who fund renovations and payroll out of their own margins started asking corporate parents for a number, not a demo. Nestor described Hyatt's internal shift as a move from "adoption" — did staff start using the tool — to "absorption" — is the company extracting value from it that shows up somewhere auditable. That's a much harder question, because it requires isolating AI's effect from everything else moving a hotel's numbers in the same quarter: occupancy swings, wage inflation, insurance costs after storm seasons, and renovation cycles that Host Hotels and others have been funding out of restoration budgets rather than growth capital.
The catch: nobody has agreed on what "payoff" means
The honest problem is definitional, not technical. A tool that cuts a front-desk shift by one hour a day saves perhaps $15 to $20 in US labor cost depending on market and union status — real money, but too small per property to show up against a hotel's other cost pressures unless it's aggregated across hundreds of properties and held constant over multiple quarters, which almost no chain has done publicly. Executives at the Washington event conceded that most current AI metrics — adoption rate, query volume, hours logged — measure usage, not value, and usage is cheap to report because it's generated automatically by the software itself. Actual savings require someone to run a controlled comparison: this property with the tool, that one without, same season, same market, same staffing baseline. That comparison is expensive, slow, and frequently undermined by the fact that AI tools get updated mid-test. Nestor's prediction of "more scrutiny" next year is really an admission that this year's numbers weren't scrutinized enough to survive a franchisee's spreadsheet.
Who actually pays if the savings don't materialize
The risk sits lowest down the ownership chain. Corporate brands like Hyatt and Marriott license or build the AI tools and can spread development cost across thousands of properties; individual franchise owners pay licensing or subscription fees per property regardless of whether their specific hotel sees a measurable return. If a 150-room property in a secondary US market pays $300 to $800 a month for an AI guest-messaging platform and can't show it reduced staffing need or improved review scores enough to justify the fee, that owner is absorbing a cost the brand's aggregate data might still call a success. The tag on this story lists Hyatt, Marriott and Choice Hotels together for a reason: all three have public-facing AI initiatives and private, unresolved conversations with the owners who actually fund them.
The promise that can't be kept twice
The open question the industry hasn't answered — and avoided answering directly in Washington — is whether AI can deliver both of the things hotel companies have promised simultaneously: that it won't cost jobs, and that it will show up as profit. Those two claims pull against each other by definition. If an AI tool genuinely replaces labor hours, the saving shows up as reduced headcount, which is the job-cut outcome executives have been reluctant to confirm publicly while unions and franchise associations are watching. If it doesn't replace labor hours, the saving has to come from somewhere else — higher conversion on direct bookings, better rate optimization, fewer comped nights from service failures — and those effects are smaller, slower, and harder to attribute to the software alone. No major chain has published a multi-quarter study isolating either effect. Until one does, "more scrutiny next year" is a forecast, not a finding.
What would actually settle this
A chain would need to run the same AI tool across a matched set of properties against a comparable control group for at least two full booking cycles, then publish the labor-hour and revenue delta separately — something no major hotel company has done in public filings or investor calls through 2025.
Which hotel brands are furthest along
Hyatt has been the most specific publicly about shifting from usage metrics to profit metrics; Marriott and Choice Hotels have discussed AI tools in investor materials without yet publishing comparable before-and-after cost data.