All-Inclusive Resorts: AI for Cost-Per-Guest-Day Control
The wristband goes on at check-in, and at that instant the property's revenue from that guest is fully determined. Not estimated. Not forecast. Determined. Everything that follows - the seven breakfasts, the forty-one drinks, the catamaran excursion, the two spa treatments the guest booked and one they no-showed, the towel that never came back - is pure cost.
Traditional hotel management is built on the opposite assumption. The entire apparatus of revenue management, upselling, ancillary capture, and total revenue optimization exists because a transient hotel's revenue is open for the duration of the stay. A guest who eats in the restaurant generates incremental revenue. A guest who books the spa generates incremental revenue. Consumption and revenue move in the same direction.
At an all-inclusive, they move in opposite directions. Consumption is a cost event. And that single inversion invalidates most of the operating instincts a general manager brings from the branded select-service or full-service world.
The segment isn't a niche. Dataintelo, a market research firm, values the global all-inclusive resort market at $67.4 billion in 2025 and projects $134.8 billion by 2034, about 8% a year. The big brands are acting on it. Hyatt, which bought Playa Hotels & Resorts, ended 2025 with a record pipeline of about 148,000 rooms and says it is taking its all-inclusive offerings into new markets including Saudi Arabia, Southeast Asia and North Africa.
So more owners and operators are running assets where the one lever that really moves the P&L below the top line is cost per guest day. Plenty of them manage it with month-end spreadsheets and a chef's intuition.
The Metric That Actually Governs the Asset
Cost per guest day (CPGD) is the total variable and semi-variable cost of serving one guest for one 24-hour period on property. It is the all-inclusive equivalent of cost per occupied room, but broader, because at an all-inclusive nearly every department is a consumption department.
The reason CPGD matters more than any departmental cost percentage is arithmetic. If your average package rate is $340 per guest per night and your CPGD is $196, you are running a $144 contribution. Move CPGD to $178, a 9% improvement, and contribution goes to $162, a 12.5% increase in the number that actually pays debt service. No rate increase. No additional guest. No new market.
No standardized public benchmark for CPGD exists, which is part of the problem. What you can do is break it into the components each department controls and work out what drives each one.
| CPGD component | What moves it |
|---|---|
| Food cost | Production quantities versus actual covers; buffet waste |
| Beverage cost | Units poured per guest; over-pouring and shrinkage |
| F&B labor (allocated) | Schedule fit against covers by outlet and meal period |
| Activities, entertainment, excursions | Capacity staffed and provisioned versus guests who show up |
| Rooms, housekeeping, laundry | Occupants per room, length of stay, linen loss |
| Utilities, amenities, other variable | Occupancy pattern, amenity standards |
F&B dominates that list, and inside F&B the labor to produce and serve the food costs more than the food itself. CBRE reports that as a share of total F&B department expense, labor is the greatest at 59.4%, followed by cost of goods sold at 24% and other at 16.6%.
Plenty of cost-control programs get the implication backwards. Buying food better is a procurement exercise with a ceiling. Producing the right amount of food, at the right hours, with the right number of people on the clock, is a forecasting exercise with a much higher ceiling - because it moves food cost and labor cost simultaneously.
At a transient hotel, a guest who consumes more is worth more. At an all-inclusive, a guest who consumes more is worth less. Every operating instinct you imported from the branded world has to be re-derived from that one inversion.
Why Consumption Is Forecastable, and Why Resorts Rarely Forecast It
Experienced all-inclusive operators will tell you guest consumption is chaotic. It's less chaotic than it looks, because it's driven by a short list of variables the property already collects.
Consumption at an all-inclusive is a function of: guest mix (adults-only versus family, source market, repeat versus first stay), day of stay, arrival and departure times, weather, competing on-property activity, and package tier. Every one of those is either known at booking or knowable 24 to 72 hours in advance. What is missing is not data. What is missing is a model that turns those variables into a covers forecast by outlet by meal period, and a labor schedule derived from it.
The pattern most worth testing is the day-of-stay curve. There's no reason to expect guests to consume evenly across a stay. Arrival day is shaped by arrival times and a pull toward the buffet and the nearest bar. A guest who has learned the property by day two or three books the à la carte dinners and the excursions. Later in a long stay, novelty fades, guests sleep in, and some eat off property despite having paid not to. Whether your resort follows that shape, and how steeply, is something your own PMS and POS history can tell you.
If the curve is there, a resort running a fixed buffet production standard and a fixed labor schedule over-produces on arrival and departure days and runs short on the busiest middle days. With staggered arrivals the two effects partly cancel and disappear into the monthly P&L, which is how they can persist for years. They only show up when you forecast at the guest-day cohort level.
Waste is where the published results are strongest. Winnow, the AI food waste vendor, says kitchens that measure food waste typically cut it by more than half, and that its customers, Hilton, Accor, Marriott, Mandarin Oriental and IKEA among them, collectively save more than $100 million a year in food costs. Its hotel case studies include a 73% cut in waste by weight at Mandarin Oriental, Hong Kong and 67% at London Marriott Canary Wharf. Independent research points the same way: Cornell's 2026 Hotel Food Waste Report puts waste at about 8% of hotel food costs, and the Champions 12.3 study of hotels found $7 saved for every $1 invested in cutting food waste.
Those numbers matter at a transient hotel. At an all-inclusive, where nearly every meal is included and the revenue side can't absorb the miss, they matter more.
The Four Systems That Decide All-Inclusive Margin
1. Per-guest-day F&B production modeling
The goal is a covers forecast, by outlet, by meal period, 72 hours out, refreshed every 24 hours, with production quantities derived from it. The inputs are already in your stack: the PMS holds arrivals, departures, package tier, party composition, and length of stay; the POS holds historical capture by outlet; a weather API costs almost nothing.
What AI adds over a spreadsheet isn't magic. It can weight dozens of interacting variables that a human planner can't hold at once. And a buffet is where forecast error turns most directly into food cost, because over-production at a buffet has no salvage value.
2. Beverage consumption and variance control
Beverage is where all-inclusive resorts lose money most quietly. The package removes the transaction, and with it the natural control that a paid check imposes. Free pours drift. Inventory walks. And because there is no revenue line to reconcile against, the classic pour-cost percentage - the industry's default control metric - is structurally useless at an all-inclusive.
The discipline that works is unit variance: theoretical units consumed, derived from the guest-day forecast and historical per-guest drink counts, measured against actual units depleted from inventory. Inventory specialists are blunt that monitoring beverage cost percentage has proven ineffective at stopping shrinkage because it does not detect over-pouring or theft. Count drinks, not dollars.
3. Activity and capacity planning
Every excursion seat, spa slot, dive tank and à la carte cover carries a real marginal cost, and at an all-inclusive the guest pays nothing extra to reserve one and nothing to abandon it. Expect no-shows, on capacity that was staffed and provisioned in advance, and expect that nobody is measuring them.
A catamaran that sails at 60% of booked capacity has burned crew, fuel, and F&B provisioning against guests who are asleep. This is the one area where the all-inclusive model creates a cost problem that a paid-ancillary hotel simply does not have, and it is fixable with reservation-level no-show prediction, dynamic overbooking of activity inventory, and same-day re-release of abandoned slots. The mechanics are close to what we described in our work on AI for resort activity booking - inverted, because here the objective is cost avoidance rather than revenue capture.
4. Labor scheduling derived from the consumption forecast
Because labor is roughly 59% of F&B department expense, the consumption forecast is only half-monetized until it drives the schedule. The usual failure is a forecast that lives in the chef's office and a schedule built two weeks out from last year's pattern.
This matters more in a constrained labor market. AHLA's Front Desk Feedback survey found 65% of hotels reporting staffing shortages, with 71% carrying openings they could not fill and hotel employment still nearly 10% below pre-pandemic levels. CBRE notes that hours worked at the typical hotel in its 2024 sample were down 7.4% from 2019 while compensation dollars were up 22.1%, and that hotels are using more contract labor, at a significant premium. When every hour is expensive and scarce, placing hours accurately is worth more than negotiating the rate.
| F&B department metric | Value | Direction |
|---|---|---|
| Department profit margin, H1 2025 | 29.1% | Up from 28.7% prior year |
| Labor as share of department expense | 59.4% | Largest single line |
| Cost of goods sold as share of expense | 24.0% | Up 3.3%, year to date June 2025 |
| Other expense as share of department | 16.6% | Remainder of expense |
| Labor cost growth | 2.1% | Year to date June 2025 |
| Hotel hours worked, 2024 vs. 2019 | Down 7.4% | Compensation dollars up 22.1% |
Package-Level Profitability: The Analysis Nobody Runs
A typical all-inclusive prices packages by tier and channel, then measures profitability at the property level. That is one level too coarse. The question that decides whether a package should exist is not "did the resort make money this month" but "does this package, sold through this channel, to this source market, at this length of stay, produce positive contribution after the consumption it triggers."
Sometimes the answer is no, and the losers tend to follow a pattern. Premium packages that bundle unlimited top-shelf beverage and unlimited à la carte reservations attract the guests who will use them fully. Short stays carry the arrival-day and departure-day cost profile with fewer middle days to average it down. Wholesale and deep-discount channels add distribution cost to a package whose consumption looks the same as the direct guest's; in the illustration below, the wholesale package gives up $104 a night, 27% of its rate.
| Waterfall step | Direct, 7-night, standard | Wholesale, 4-night, premium |
|---|---|---|
| Package rate per guest night | $340 | $385 |
| Less: distribution and commission | −$24 | −$104 |
| Net package revenue | $316 | $281 |
| Less: food cost | −$49 | −$63 |
| Less: beverage cost | −$19 | −$37 |
| Less: F&B labor allocation | −$61 | −$74 |
| Less: activities and entertainment | −$17 | −$26 |
| Less: rooms and housekeeping | −$33 | −$41 |
| Less: utilities and other variable | −$20 | −$23 |
| Contribution per guest night | $117 | $17 |
The second column ends arguments in asset management meetings. A package carrying a higher headline rate delivers roughly one-seventh the contribution, because the distribution cost and the consumption profile move against it simultaneously. Neither factor is visible in a rate-and-occupancy report. Both are visible the moment you model contribution at package-channel-length-of-stay granularity.
Elsewhere in the industry, non-room spending is a growth story; a Starfleet Research report puts non-room sources at 30% to 40% of incremental hotel revenue growth. At an all-inclusive those categories are bundled rather than sold, so the same trend that creates revenue upside elsewhere creates margin exposure here. The bundle is only an advantage if you can model what it costs.
A package with a higher headline rate can deliver one-seventh the contribution of a cheaper one. Rate and occupancy reports will never show you this. Contribution modeling at the package-channel-length-of-stay level shows you nothing else.
What the Technology Actually Has to Do
All-inclusive properties sit on more usable operational data than almost any other lodging format, and integrate less of it. The wristband or room-charge posting creates a consumption record for nearly every guest interaction on property - a dataset a transient hotel would pay dearly for. It typically sits in three or four systems that do not speak to each other.
Platform vendors have noticed. Agilysys markets resort software that it says unifies PMS, POS, inventory, payments and guest experience tools across resort, dining, spa, golf and activities, with real-time reporting on capacity use. Its LMS platform lets guests pick package components at reservation, and its spa module posts room charges to external systems including Oracle Opera. The category is real and improving.
What almost none of it does out of the box is the specific thing this segment needs: a forecasting layer that consumes PMS arrivals, POS capture history, activity reservations, inventory depletion, and weather, and emits a production plan, a labor schedule, and a per-package contribution statement. That is an integration and modeling problem, not a licensing problem - which argues for building a thin analytical layer on top of the platforms you already own rather than replacing them.
Resorts beginning this work often find the constraint is plumbing rather than intelligence - getting PMS, POS, activity, and inventory data into one place where a model can see all of it at once. Our Custom AI Integrations & Automations practice → exists for exactly that gap.
A Sequenced 90-Day Path
Try to optimize the whole property in one initiative and you'll likely stall on data quality and ship nothing. Sequence it so each phase pays for the next.
| Phase | Focus | Data required | Output |
|---|---|---|---|
| Days 1 to 30 | Establish CPGD baseline by day, tier, and cohort | PMS arrivals/departures, GL by department, headcount | Measurement only |
| Days 15 to 45 | Buffet waste measurement at disposal points | Waste capture by station and meal period | Waste baseline by station |
| Days 30 to 60 | 72-hour covers forecast by outlet and meal period | PMS + POS capture history + weather | Production plan |
| Days 45 to 75 | Labor schedule derived from covers forecast | Forecast output + scheduling system + productivity standards | Published schedule |
| Days 60 to 90 | Beverage unit variance and activity no-show prediction | Inventory depletion, activity reservations, attendance | Variance and no-show reports |
| Day 90+ | Package contribution model, channel and LOS decisions | All of the above, joined at reservation level | Repricing decisions |
Two design rules keep this on track. First, the baseline comes before the model. A resort that cannot state its CPGD by day and cohort today cannot evaluate whether a forecasting system improved anything, and will end up arguing about attribution instead of banking savings. Second, every forecast must terminate in an action owned by a named person - a production quantity signed by the chef, a schedule published by the F&B director. A forecast that ends in a dashboard saves nothing.
What Changes When This Works
A resort that gets cost-per-guest-day control right is managing a different business from the one next door, even with identical buildings. It knows before the week begins which cohorts are arriving and roughly what those cohorts will consume. It produces to that number and staffs to it. It knows which packages and channels earn their place in the mix, and retires the ones that don't rather than discounting them further.
Meanwhile the resort next door is running the same buffet quantity every Tuesday it has run for six years, staffing to a schedule built from memory, and discovering in the month-end package that food cost moved 180 basis points - with no reliable way to say which day, which outlet, or which guest cohort caused it.
The all-inclusive model puts all of the operating risk on one side of the ledger. Managed with transient-hotel instincts, that's a structural disadvantage. Managed the other way, a business whose revenue is fixed at check-in has a margin you can actually control, provided you can see the consumption coming.
Frequently Asked Questions
What exactly is cost per guest day, and how is it different from cost per occupied room?
Cost per guest day (CPGD) is the total variable and semi-variable cost of serving one guest for one 24-hour period on property - food, beverage, the labor to produce and serve it, activities and entertainment, housekeeping and laundry, utilities, and amenities. Cost per occupied room (CPOR) is measured per room rather than per guest and, at a transient hotel, typically excludes most F&B because F&B is a revenue department there. The distinction matters for two reasons. First, all-inclusive rooms usually hold two or more guests, so a per-room metric conceals the fact that a room with three occupants costs far more to serve than a room with one at the same rate. Second, at an all-inclusive, F&B is not a revenue department - it is the largest cost department. Any metric that excludes it is measuring the wrong thing.
Is guest consumption at an all-inclusive really predictable enough to forecast?
Usually more than operators assume. Consumption is driven by a small set of observable variables (guest mix and party composition, source market, day of stay, arrival and departure times, weather, competing on-property activity and package tier), nearly all of which are known at booking or 24 to 72 hours ahead. Day of stay is a good place to start: check your own history for lighter arrival and departure days and a mid-stay peak. The barrier is rarely predictability. The data sits in four disconnected systems and nobody has joined it.
Why is pour cost percentage the wrong beverage control metric at an all-inclusive?
Pour cost is beverage cost divided by beverage revenue. At an all-inclusive there is no beverage revenue line - the drink is bundled into the package - so the ratio either cannot be computed or gets computed against an arbitrary allocated revenue figure that moves with rate rather than with consumption. Worse, the metric is weak even where it does apply: industry specialists note that monitoring beverage cost percentage has proven ineffective at stopping shrinkage because it does not detect over-pouring or theft, The metric that works at an all-inclusive is unit variance - theoretical units consumed, derived from the guest-day forecast and historical per-guest drink counts, measured against actual units depleted from inventory. Count drinks, not dollars.
How much can a resort realistically save, and how quickly?
It depends on where you start, and no published benchmark gives a CPGD figure for the segment, so set your own baseline first. Waste measurement and covers forecasting tend to deliver the first gains; labor scheduling is the bigger prize because labor is about 59% of F&B department expense, per CBRE. The published waste evidence is encouraging: Winnow reports hotel case studies with waste cut by 67% to 73%, Cornell puts food waste at about 8% of hotel food costs, and the Champions 12.3 study of hotels found $7 saved for every $1 invested in reducing food waste. Savings from labor only show up when the consumption forecast actually drives the schedule; a forecast that stops at a dashboard produces nothing.
Do we need to replace our PMS and POS to do this?
Almost never, and starting there is a good way to stall. Resort platforms have improved; vendors now market unified PMS, POS, inventory, payments and package management for multi-amenity properties. But the specific capability this segment needs is a forecasting and contribution layer that reads across all of those systems at once and emits a production plan, a labor schedule, and a per-package contribution statement. That is an integration and modeling problem, not a licensing one. The practical path is to leave the systems of record in place, extract their data into one analytical layer, and build the forecasting on top. It is faster, materially cheaper, and it does not put the operating platform at risk in the middle of a season.
About the author. Peter Mack is a hospitality technology strategist and founder of HospitalityOS, helping independent hotels and resorts implement AI systems that drive revenue and reduce operational costs. With 25 years in hospitality operations and technology, he has worked with properties of all types and in every region as both a General Manager, Founder, Operator, Asset Manager, and Owner.