Absorbing a $25 Wage Floor: Rebuilding the Labor Model Department by Department

Wage floors headed to $25 and beyond are now law or contract in Los Angeles, San Diego, New York and Waikiki. The hours have to come from somewhere: a forecast that writes the schedule, a router that sequences housekeeping, and an AI agent that takes the phone. Here is what each saves, what stays human, and how far it goes.

Look at the server in the photo above. She's carrying two side plates back from a breakfast buffet, and in a hotel inside Los Angeles city limits with 60 or more rooms, every minute she spends on that walk now costs at least $25.00 an hour, plus a health payment if the hotel doesn't provide coverage. The plates still need carrying. What changes is how many of her other minutes go to work that a system could do.

That's the whole argument of this piece.

The wage floors are real and they're scheduled. Los Angeles moved its hotel worker minimum to $25.00 on July 1, 2026, with steps to $25.50 in 2027, $28.50 in 2028 and $30.00 on January 1, 2030. San Diego's hospitality ordinance starts lower and climbs to $25.00 by 2030. New York's new union contract goes much further. Waikiki's union hotels added another raise on January 1. If you own or run a hotel in one of those markets, your 2027 budget is already carrying the increase whether you've modeled it or not.

The usual response is a hiring freeze and a lot of overtime. I'd rather rebuild the labor model one department at a time, with three specific systems doing the heavy lifting: AI demand forecasting that sets the schedule, task-based routing in housekeeping, and an AI voice and messaging agent that takes the front desk and reservations call volume. Below is where each one saves real hours, where it doesn't, and what the staffing model looks like afterward.

The Problem: The Floor Moves Faster Than the Rate

Start with the floors themselves, because they aren't the same rule in each market. Some are city ordinances with room-count thresholds. Others are union contracts that set the market rate whether or not your property is signatory.

New York is the outlier. The Hotel and Gaming Trades Council's eight-year agreement puts housekeepers on a path to $60 an hour, with average wage increases above 50 percent over the life of the contract and the employer health contribution rising from 27.25 to 30.25 percent of payroll. The Los Angeles numbers look modest next to that. They're still a $2.50 jump in one year on the hours that sit closest to the floor.

San Diego is the slow version. The city's hotel rate is $19.00 from July 2026, reaching $25.00 in July 2030. The city's own budget analyst found non-tipped hotel workers already paid anywhere from $17.25 to $30.00 an hour, with labor running 24 to 50 percent of hotel costs. That spread matters. A floor doesn't just lift the people below it; it squeezes everyone just above it, who will reasonably ask why the gap closed.

And the floors land on a cost base that was already moving the wrong way. HotelData.com's analysis of Actabl data from about 5,000 U.S. hotels found wage cost per occupied room rose 12.8 percent in 2025, to $48.32, with a 21.1 percent jump in the fourth quarter alone. Labor cost per hour was up 8 percent. Revenue didn't keep pace.

The Data: Hours Already Fell. The Easy Ones Are Gone.

Here's what makes this harder than it sounds. Hotels have been cutting hours since the pandemic, mostly by not refilling positions. CBRE's 2025 Trends survey found that since 2019 the hours worked at the typical hotel are down 7.4 percent while compensation dollars are up 22.1 percent. In 2024 alone, salaries, wages and benefits rose 4.8 percent against total revenue growth of 2.3 percent.

So the slack that existed in 2019 has mostly been taken. What's left is the structure of the work.

Actabl's own 2025 report shows operators still found efficiency. From January to September, hours per occupied room fell 13.5 percent in guest services, 7.1 percent in housekeeping and 14.6 percent in management, and minutes per occupied room were down 9 percent overall. Those are real gains. They're also the kind you get once, from tightening schedules and cross-training, and they don't repeat on their own.

Read that table as a map of where a floor hurts. A full-service hotel at 2.57 hours per occupied room feels a $2.50 increase nearly twice as hard per room as a select-service hotel at 1.44. A resort feels it three times as hard.

The Framework: Sort Every Hour Into Three Buckets

Before any software, go department by department and sort the scheduled hours into three buckets.

The first bucket is hours that exist because of demand you didn't predict. A server scheduled for a breakfast that ran at 60 covers instead of 110. A second front desk agent on a Tuesday afternoon when arrivals came in late. These are forecasting hours, and a better forecast removes them without anyone working harder.

The second is hours spent on routing and coordination: walking to the wrong floor, waiting on room status, calling the desk to ask what's next, radioing engineering. These are information hours. Better information removes them.

The third is hours of repetitive conversation. Most of the front desk phone, a large share of reservations, the same twenty questions on WhatsApp. These are the hours an AI agent can actually take.

Everything else is the work itself: cleaning the room, carrying the plates, fixing the leak, greeting the guest who walks in. I don't automate that bucket, and I'd be suspicious of any vendor who says they can at these wage levels.

The Mechanism: Three Systems, Wired Into One Schedule

1. AI demand forecasting that writes the schedule

The labor forecast in most hotels is a spreadsheet built from the revenue forecast, adjusted by a manager's memory. An AI forecasting model does the same job at a finer grain: it reads the PMS on-the-books by day and hour, the arrival and departure pattern, group pickup, outlet POS covers by daypart, flight and event data where it matters, and weather for resorts. It produces expected workload per department per hour, then converts that into required labor using your own standards: rooms per attendant, transactions per desk agent, covers per server.

The output is a draft schedule built to demand, published to staff phones, with a variance alert when actual occupancy or covers move. The manager still approves it. What changes is that the schedule starts from the forecast instead of last week's schedule.

Where it pays most is the edges of the day: the breakfast ramp, the late-afternoon arrival wave, the slow middle of a weekday. Those are exactly the shifts that get padded "just in case" when hours were cheap.

2. Task-based routing in housekeeping

We covered this in detail in yesterday's piece on housekeeping task routing. The short version: replace the floor-block sheet with a minute credit for each room based on its state (departure, stayover, long-stay checkout), a live sequence on each attendant's phone fed by PMS room status and the arrival list, and inspections aimed at risk instead of every room. The best-documented single result is a 7.4 percent cut in minutes per occupied room from splitting checkout and stayover credits, with quality unchanged. A realistic range for the full router is 5 to 15 percent.

Housekeeping is usually the largest block of hourly hours in a hotel, so even the low end matters more here than a big percentage anywhere else.

3. An AI agent on the phone and in messaging

This is the one that moved fastest in 2026, and the one with the best hard data. HelloShift published 15 months of calls answered by its AI voice agent across hotel properties: about 75 percent were handled without a transfer to a person, and roughly 72 percent were resolved. About one in four calls came outside 8am to 6pm. Another quarter of all inbound calls were hang-ups, wrong numbers and pocket dials that a human used to pick up anyway.

The topic mix is the useful part, because it tells you which hours come back to the desk.

On reservations, the larger resort operators have the clearest numbers. PolyAI reports that Golden Nugget automated 87 percent of PBX operator calls and 34 percent of reservation calls, averaging more than 300 completed reservations a week within two months of launch. That's vendor data and I'd treat it as the high end. It's still a useful bracket: PBX is mostly automatable, reservations about a third.

The mechanism matters more than the brand. The agent needs read and write access to the PMS for reservation changes, a live connection to the booking engine for new bookings, a ticket feed into housekeeping and engineering for in-stay requests, and a clean handoff to a person with the transcript attached. Without the write access, you've bought an answering machine that sounds nicer.

Department by Department: What Automates and What Doesn't

This is the table I'd want in front of me before a budget meeting. The hours-saved ranges are my planning estimates built from the sources above; confirm each against a month of your own data before you put it in a forecast.

The front office number looks big. It's big because the phone is the most automatable work in the building, and it's a smaller department. Housekeeping's percentage is lower and the hours are larger. F&B is where I'd lower expectations most: the gains come from not overstaffing a slow breakfast, and that's about forecasting, not machines.

The Model: One Full-Service Hotel Under a $25 Floor

To make this concrete, here's an illustrative model. It's a 250-room full-service hotel in Los Angeles running 80 percent occupancy, so about 73,000 occupied rooms a year. At Actabl's 2.57 full-service average that's roughly 187,600 labor hours, of which I've assumed 160,000 are hourly non-management hours. These are assumptions for the example, not a specific property.

Now price the floor against it. Assume 60 percent of those hourly hours sit at or near the old $22.50 rate and take the full $2.50 increase, and the remaining 40 percent get a $1.00 compression raise to keep some gap. That's about $304,000 in wages, or roughly $365,000 with 20 percent for payroll taxes and benefits. I've left out the $4.25 health payment on the assumption the hotel already provides coverage; if it doesn't, the number gets much bigger.

At the middle case, the redesign absorbs almost all of the July 2026 step. That's the realistic headline. It doesn't absorb the 2028 step to $28.50, and nothing here pays for the jump to $30.00 in 2030 on its own. Those will come out of rate, out of mix, or out of margin.

Thirteen thousand hours is about six and a half full-time positions. I wouldn't plan to lay anyone off to get them. Hotels lose people every year; the redesign is what lets you not refill some of those seats, and stop paying overtime to cover the gaps you couldn't fill anyway.

The Staffing Model That Results

The org chart after a redesign looks a little different, and it's worth being honest with the team about how.

The front desk gets smaller and more senior. Fewer agents answering phones, more time at the counter, and one person per shift who owns the escalations the AI agent hands over. I'd pay that person above the floor on purpose. When I ran sales strategy and operations at Starwood across more than 800 hotels, the properties that held service scores through staffing cuts were the ones that kept someone on every shift with the authority to fix a problem on the spot.

Housekeeping keeps its headcount close to flat and finishes the day more evenly. The supervisor role shifts from walking every room to working the exception list.

F&B schedules shrink at the edges of each meal period. The core service team stays.

A new job appears, usually part of someone's existing role: whoever owns the forecast and the agent. They review the weekly variance between forecast and actual, check the agent's transfer log, and fix the scripts. At our ten Collective Retreats locations, before the business was acquired in 2025, every system we rolled out without a named owner on site drifted within a quarter. That's still true.

Implementation: Six Months, In This Order

Month one is baseline. Pull a year of hours by department from payroll and occupied rooms from the PMS. Pull call volume and call reasons from the phone system, and if you can't get reasons, sample 200 calls by hand. Time a month of room cleans. Without this, you won't know whether anything worked.

Months two and three, the phone. It's the fastest payback and the easiest to measure. Start with PBX and after-hours, then add reservation changes once the PMS write access is tested. Track containment, transfers and complaints weekly, and listen to twenty recordings every Friday.

Months three and four, housekeeping credits and routing, following the sequence in the task routing piece.

Months four through six, forecasting into the schedule. This one depends on the first two, because the agent and the router generate the workload data the forecast needs. Run the AI-drafted schedule beside the manager's schedule for four weeks before you switch. Wage-floor ordinances also tend to come with scheduling and training rules (Los Angeles added a training requirement), so build compliance checks into the scheduling tool rather than a binder.

If the forecast is the piece you're missing, it's the core of our AI revenue optimization and forecasting work, since the same demand model that sets rates should be setting hours.

One caution. Don't announce this as a cost program. Tell the team the floor is coming, the hotel is paying it, and the systems are how it pays it without cutting their hours. In my experience that's also the truth.

Frequently Asked Questions