AI Menu Engineering for Hotels: How Data-Driven F&B Decisions Add $15+ to Guest Spend

Most hotel owners wouldn't let a 2.5% RevPAR move pass without a meeting. They will sign off on a food cost percentage nobody has re-checked since the last menu print. F&B is historically the industry's second largest revenue source after rooms, per CBRE, and it gets a fraction of the scrutiny.

For two decades hotel F&B has been run as a guest amenity that roughly breaks even. 2025 made that harder to defend. CBRE's H1 2025 analysis of U.S. full-service, resort, and convention hotels shows F&B revenue per occupied room up 3.8% year over year, ahead of the 3.0% growth in total hotel revenue, with department margin at 29.1%. Banquets were mixed: revenue fell 7.3% at convention hotels and rose 8.7% at resorts. The operators doing well here are winning on math.

That math has a name. Menu engineering - the discipline of analyzing each item's contribution margin and popularity, then redesigning the menu, the pricing, and the prep around what the data says - has been a restaurant fundamental since Michael Kasavana and Donald Smith developed their matrix at Michigan State University in 1982. What changed is that the matrix no longer has to be a quarterly Excel exercise. It can run as a continuous loop on top of your POS, inventory system, reservations data, and supplier feeds. Jengu, which sells AI tools to hotels, tells operators to expect 1.5 to 3.5 points of food-cost reduction within 6 to 12 months and 8% to 15% F&B revenue uplift. That's vendor guidance, not an independent benchmark. The $15 per occupied room in the headline is a target to size against your own numbers, and the economics section below shows what it would be worth.

This is written for the GM or owner who knows F&B is leaking money and can't say whether the leak is pricing, mix, waste, labor, or all four. Usually it's all four, in different proportions on different days, which is why a spreadsheet can't keep up.

The strategic case: why F&B is hospitality's most under-engineered revenue center

Three things have made F&B economically interesting again. The first is mix. In Mandelbaum & Grigg's read of the CBRE data, food revenue rose 5.2% in hotel venues and 4.0% at banquets in H1 2025, while beverage was flat in venues and down 2.0% at banquets. The second is margin: F&B department profit margin went from 28.7% to 29.1% in twelve months. Modest, but moving the right way. The third is cost structure. In the same CBRE data, labor was 59.4% of F&B department expenses, cost of goods sold 24.0%, and other expenses 16.6%.

Put together, small percentage moves in F&B now show up in EBITDA. Take a hypothetical property with $4M of F&B revenue and $5.7M of rooms revenue. Two points of food cost is $80,000, about what a 1.4% RevPAR gain would add in rooms revenue. Ask which of those two gets more time in the owner's meeting.

Why has this gap persisted? Because hotel F&B has been operating on a stack - printed menus refreshed twice a year, recipe costing done in a binder, weekly inventory counts on clipboards, item-level profitability reviewed by one chef in their head - that was designed for a world where labor was cheap, supply chains were stable, and guest expectations were uniform. None of those three holds in 2026. When labor is three-fifths of the department's expense base, a binder and a clipboard are expensive controls.

The four problems AI menu engineering actually solves

Be precise about what "AI menu engineering" is and isn't. It is not a chatbot writing menu descriptions. It is not a glossy dashboard summarizing yesterday's covers. It is a closed-loop system that ingests POS-line, recipe, inventory, supplier, reservation, and external-demand data and answers four specific operational questions every day.

1. What should each item cost?

Recipe costing is the foundation. Supy's 2026 menu engineering piece makes the point plainly: many operators unknowingly run menus on last month's costs, and quarterly reviews can't keep pace with supplier price changes. AI systems pull supplier invoices via OCR or EDI, recalculate every recipe nightly, and flag the items where actual margin has slipped below threshold. The useful output reads less like "your beef tenderloin is now expensive" and more like "this dish, which sells 47 covers per week, is now contributing $4.10 instead of $6.80; here are three plate-cost adjustments that restore margin without changing portion or perceived value."

2. What should each item cost the guest?

Dynamic menu pricing in hotels should be a controlled discipline. The version worth considering is bracketed: AI sets the price inside an owner-approved corridor based on demand, day-part, channel, and even weather. The published evidence is thin. Vendor blogs from Loman and Checkmate describe the benefits in general terms, and the one hard number Loman offers elsewhere, up to 40% more sales in peak hours, comes with no source. Hotels should ignore the stunt versions (surge pricing on a burger at lunch) and focus on the controlled versions (room-service breakfast tray pricing tied to occupancy, group function menu pricing tied to attendee count, banquet bar pricing tied to historical pour ratios).

3. What should we be selling more of?

This is the classic menu engineering question, reframed as a continuous AI task. The Stars/Plowhorses/Puzzles/Dogs matrix from Kasavana and Smith still works - but it works far better when an algorithm runs it across 180 menu items on a rolling 90-day window, segments by day-part and channel, and adjusts the position, photo, description, and pricing of each item based on its current quadrant. Toast's guide to the matrix lays out the quadrants and the classic moves. Hotels arguably have more to gain from running it continuously than a standalone restaurant does. Their menus are longer, their day-parts more varied, and their channels more numerous.

4. What are we throwing away?

Food waste is the most underrated lever in hotel F&B. UK hospitality throws away around 920,000 tonnes of food a year, per WRAP figures compiled by Waste Direct. AI-powered waste tracking is usually a camera and scale at the back-of-house bin, with a model that recognizes what's being thrown out. The best public evidence on the economics predates most of the AI tools: in the Champions 12.3 review of 42 hotels in 15 countries, the average site cut kitchen food waste 21% by weight within a year, more than 70% of sites recovered their investment inside the first year, and the average hotel saved $7 for every $1 invested. WRAP also publishes a case on Fairmont Jakarta cutting food waste 25% and saving US$216,000.

The data foundation: what AI actually needs from your stack

An AI menu engineering program is only as good as its data feeds, and in a lot of hotels those feeds are missing, inconsistent, or too slow. Before any vendor demo, an honest stack inventory is worth more than a strategy deck. The minimum feeds are:

Data feed	Source system	Refresh cadence	Why it matters
POS line items (item, modifier, time, channel, server)	Toast, Micros, Lightspeed, Square	Real time	The atomic unit of menu engineering
Recipe and modifier costs	Recipe management (meez, Apicbase, MarginEdge)	Nightly	Contribution margin denominator
Supplier invoices and unit prices	AP system, supplier portals, EDI/OCR	On receipt	Keeps recipe cost current as supplier prices move
Inventory counts and theoretical-vs-actual usage	Inventory module of recipe system	Weekly minimum, daily ideal	Surfaces shrinkage and waste
Occupancy, group block, and arrival forecasts	PMS (Opera, Mews, Cloudbeds)	Hourly	Demand signal for prep, staffing, dynamic pricing
External signals: weather, events, flights	OpenWeather, PredictHQ, FlightAware	Daily	Captures non-occupancy demand variance

Implement them in the order shown. POS lines and recipe costs come first because everything else depends on them; external signals come last because they only help once the rest is clean. Revfine's F&B analytics review makes the related point that PMS, POS, and finance systems are often disconnected, and that the goal is a single source of truth. A hotel that buys an external-signals demand engine before its POS-to-recipe mapping is clean is buying a Ferrari with no road.

The AI menu engineering matrix, modernized

The original Kasavana-Smith matrix lives on, but the way operators act on it has shifted. The classical four-quadrant logic - Stars, Plowhorses, Puzzles, Dogs - still defines the operating moves. What AI changes is the granularity (per-day-part, per-channel, per-segment), the cadence (continuous, not quarterly), and the depth of the recommended action.

Quadrant	Profitability	Popularity	Classic action	AI-augmented action
Star	High	High	Feature prominently	Test small price increases inside elasticity band; protect supply chain
Plowhorse	Low	High	Reduce cost or raise price	Identify 2 to 3 cost drivers (protein, garnish, plating labor); A/B test sub-recipes
Puzzle	High	Low	Reposition or rename	Optimize menu position, description, photography; route to upsell prompts
Dog	Low	Low	Remove	Test reformulation with shared ingredients before removal; check seasonal pattern

Two refinements matter for hotels specifically. First, hotels run menus across many channels (restaurant, room service, banquet, mini-bar, retail), and the same item may sit in different quadrants on different channels. A burger that is a Plowhorse on the lobby restaurant menu can be a Star on the room-service menu where convenience premium applies. Software can track quadrant assignment per channel. A chef with a spreadsheet won't. Second, hotels carry significant seasonality and group-mix variation. An item that looks like a Dog in March can be a Star in October. AI rolling-window analysis prevents premature delisting; static quarterly analysis does not.

The economics: what $15 per occupied room would be worth

There is no independent benchmark behind a figure like $15 per occupied room, so treat it as a target to test against your own numbers. Here is where it would have to come from.

Lever	Mechanism	P&L side	What to measure
Dynamic pricing inside owner-approved corridor	Price set within an elasticity band by day-part and channel	Revenue	Average check and covers on piloted day-parts against a control period
Menu mix optimization (Stars/Puzzles routing)	Layout, description, and prompt-based steering toward high-margin items	Revenue and margin	Contribution margin per cover
Personalized pre-arrival and in-stay upsell	AI-recommended F&B add-ons in confirmation, mobile, and in-room channels	Revenue	Offer conversion and F&B revenue per occupied room
Recipe-cost discipline and supplier optimization	Continuous reconciliation of supplier prices against recipe cost	Cost	Theoretical versus actual food cost
Waste reduction	AI bin-tracking and demand-based prep	Cost	Tracked waste by weight and value against baseline

The levers overlap, so the gains won't simply add up. Waste reduction captures some of the same kitchen labor as recipe discipline, and mix optimization works some of the same elasticity as dynamic pricing.

Illustrative arithmetic, not a benchmark: a 200-room property at 70% annual occupancy has about 51,100 occupied room-nights a year. At $15 per occupied room, that is about $766,500 of incremental annual F&B revenue. Whether $15 is reachable depends on where you start. At the 8% to 15% uplift Jengu quotes, you would need $100 to $190 of F&B revenue per occupied room today to get there. Implementation costs vary too much by vendor and property to quote a fair range, so get quotes and run the payback against your own baseline.

Implementation: a 90-day rollout that actually works

The trap most hotels fall into is treating AI menu engineering as one project. It is not. It is four overlapping disciplines - costing, pricing, mix, waste - that each need their own owner, data feed, and metric. The 90-day rollout below assumes a property with a functioning POS and recipe management system. If those are missing or broken, fix them first; no AI layer rescues bad foundational data.

Phase	Weeks	Workstream	Owner	Exit criterion
0. Baseline	1 to 2	POS export, recipe audit, supplier price file pull, current matrix run on last 90 days	F&B Director + Controller	Single source of truth for last quarter's item-level P&L
1. Costing discipline	3 to 6	Recipe management deployment; supplier EDI/OCR; nightly recipe re-cost	Executive Chef + F&B Director	Theoretical-vs-actual food cost tracked weekly, inside a drift tolerance you set
2. Mix & menu redesign	4 to 8	AI matrix per day-part and channel; menu redesign for top 3 venues; description and layout tests	F&B Director + Outlet Managers	Measured gross profit change on the top 30 items
3. Dynamic pricing (controlled)	7 to 10	Bracketed pricing on room-service and banquet first; restaurant last	F&B Director + Revenue Manager	Documented elasticity bands; measured result on piloted day-parts
4. Personalized upsell	8 to 11	Pre-arrival F&B prompts via CRM; in-room mobile prompts; check-in upsell	Director of Marketing + F&B Director	Baseline and measured conversion on F&B upsell offers
5. Waste tracking	9 to 12	Bin scales + camera vision; prep adjustment loop; weekly waste-cost report	Executive Chef + Sous Chefs	Tracked waste falling against baseline (the Champions 12.3 hotels averaged 21% in year one)
6. Steady state	13+	Weekly menu engineering review; monthly recipe re-cost audit; quarterly elasticity refresh	F&B Director	Continuous improvement against quarterly targets

Three rollout failure modes are worth flagging because they are common. First, leaving the executive chef out of the costing workstream - chefs who don't own the recipe data won't trust the recommendations, and the program stalls. Second, deploying dynamic pricing on the public restaurant menu first - guest-facing surprise drives complaints and brand harm; pilot in non-guest-facing or low-elasticity channels (room service breakfast, banquet bars). Third, treating waste tracking as an environmental program rather than a financial one - when it is run by sustainability, it produces reports; when it is run by F&B with a P&L impact target, it produces savings.

Vendor categories: where the tools actually live

Hotels evaluating AI menu engineering technology face a fragmented market. There is no single vendor that does everything well. The practical move is to assemble the categories below and use middleware (or the modern PMS data hub) to integrate them.

Category	What it does	Representative vendors
Recipe management + costing	Recipe database, allergen tracking, costing, prep yields	meez, Apicbase, MarginEdge, Crunchtime
POS analytics + menu engineering	Item-level P&L, mix matrix, A/B testing	Toast Analytics, Restaurant365, Tastewise
Dynamic pricing	Demand-based, channel-aware price corridors	Few hotel-specific tools; usually a POS or ordering platform feature or a custom build
Waste tracking (vision + scale)	Camera + scale at bin; ML item recognition	Winnow, Leanpath, KITRO, Orbisk
Pre-arrival & in-stay upsell	Personalized F&B offers via CRM/PMS	Revinate, Oaky, Canary, Cendyn

Two evaluation criteria matter more than the others. First, integration depth: a vendor that "supports Opera" but requires a nightly CSV export is materially worse than one with a real-time API. Second, data ownership: ensure your contract guarantees you can export item-level history if you switch vendors. The dirty secret of this category is that switching costs are mostly about data, not workflow.

What can go wrong

Three failure modes are worth planning for.

Guest backlash on dynamic pricing. Guests accept rate variability in rooms; they accept it less in food. The safer rule is to keep it invisible at the point of sale. Prices update overnight, never during a meal; the published menu shows the day's price; nobody hears "this item is $2 more because it is 7pm." The cautionary tale is Wendy's. After its CEO said on a February 2024 earnings call that the chain would test dynamic pricing on new digital menu boards, coverage framed it as surge pricing, and the company had to say publicly it would not raise prices when customers visit most. Loman's pricing guide recommends a fixed maximum and minimum price for the same reason.

Algorithmic monoculture. If every property in a comp set uses the same vendor and the same elasticity model, prices converge and competitive advantage evaporates. The defensible play is to layer property-specific guest data (loyalty profiles, in-stay behavior, F&B history) on top of the vendor's base model - which is where clean CRM integration earns its keep.

Recipe drift. AI margin calculations are only as good as the recipes underneath them. Kitchens that quietly change a recipe - adjusted portion, substituted ingredient, modified prep - without updating the recipe database create a slow-motion data poisoning. The discipline is monthly recipe-versus-actual audits; the technology is sub-recipe versioning.

Where this goes next

Two shifts look likely to change AI menu engineering inside hotels. The first is the convergence of revenue management for rooms and F&B, usually called total revenue management: pricing rooms, F&B, spa, and ancillaries as one optimization problem. The math is hard and the prize is large. If a guest profile signals a 60% chance of a $200 restaurant spend, that can justify a room rate a rooms-only model would turn down.

The second is recipe-level generative AI used for cost-constrained recipe development, which is more useful than another tool for writing menu copy. Given a target gross profit, target plate cost, available ingredients, and prep equipment, the next generation of tools will propose three viable recipes, each costed and tested in simulation. Hotels with strong chef talent will use this to expand the chef's range; hotels without will use it to reduce reliance on chef tenure. Either way, the chef still signs off on the plate.

The owner's question

The framing question for any owner reading this far is simple: what is the size of the F&B opportunity at my property, and what is the smallest set of investments to capture half of it inside twelve months? The answer is property-specific, but the methodology is not. Pull last year's item-level POS data. Run the Kasavana-Smith matrix on it. Compute the food cost percentage and compare it to your own trailing twelve months. Quantify the room-service and banquet attach rates. Inventory the recipe and supplier feeds you have today. The gap between today's F&B contribution and what disciplined AI menu engineering would produce may well be larger than the next RevPAR project on your capex list.

Hotels starting this work often benefit from a structured baseline assessment before selecting any vendor - a vendor demo without an honest read of your current item-level economics is a recipe for buying the wrong tool. Our AI Revenue Optimization & Forecasting engagement → is built around exactly that sequencing: baseline the F&B P&L, identify the levers worth pulling first, select and integrate the right tooling, and operate the program for the first two quarters until it is internally self-sustaining.

FAQs

How quickly can a hotel see results from AI menu engineering?

Recipe costing and menu mix usually move first, because neither needs a guest-facing change. Dynamic pricing and waste tracking take longer to settle. For waste, the best public benchmark is the Champions 12.3 hotel review: an average 21% cut in kitchen food waste within the first year, with most sites paying back inside that year. Properties that try to roll out all four disciplines at once tend to destabilize their kitchens.

Will dynamic menu pricing alienate guests?

It can, which is why the guardrails matter: bracketed prices inside owner-approved corridors, daily rather than real-time updates, no prices shifting mid-meal, and a carve-out for loyalty-member-facing menus where consistency matters more than incremental margin. The Wendy's episode in 2024 showed how fast "dynamic pricing" gets heard as "surge pricing."

Do small or boutique hotels benefit, or is this only for large operators?

Boutique properties often benefit disproportionately. Their menus are shorter, their data is often cleaner, and their chefs sit closer to the P&L. A 60-room boutique with one signature restaurant can often act on the recommendations faster than a large operator, because fewer people have to agree. The constraint at small properties is bandwidth, not capability.

Where does this overlap with revenue management for rooms?

Increasingly, in deliberate ways. Modern total revenue management views a guest's expected F&B contribution as part of the room-rate decision. If your AI menu engineering stack produces reliable per-guest F&B contribution forecasts, those forecasts should flow back to your rooms revenue management system. Of all the integrations in this piece, that one has the most upside.

What is the most common implementation mistake?

Deploying dynamic pricing before the recipe costing layer is clean. Without accurate per-item cost, dynamic pricing is optimizing the wrong objective function - you are setting prices to maximize a margin number that is itself wrong. The correct sequence is: recipe management and costing first; menu mix optimization second; dynamic pricing third; upsell and waste tracking in parallel with steps two and three. Skip step one to chase pricing and you are optimizing against a cost number that's wrong.