WhiteSpace  ·  AI Agents  ·  March 2026

What can AI agents actually do for hotel F&B?

Not theory. Not a pitch deck. These are real conversations with AI agents trained on an actual hotel's financial and guest data. Here's what they flag, what they recommend, and — just as important — what they can't do.

Every few months, a new AI tool promises to "transform hospitality." Most of them are chatbots with a hotel skin. They give generic advice, hallucinate numbers, and can't tell you anything about your property because they've never seen your data.

This is different. We built AI agents that are trained on a specific hotel's actual data — revenue by outlet, food cost by item, guest reviews by platform, social media engagement by post. They don't guess. They read the numbers and tell you what they mean.

Below are real conversations from a live hotel console. The property is a luxury resort in Thailand with five F&B outlets, ~155 covers per day, and ฿33.6M in F&B revenue over the past six months.

Three agents, three specialisms.

Each agent has a defined role and a clear boundary. They don't overlap, and they don't pretend to know things they don't.

Cost Controller

Watches food cost variances, flags red-zone items (FC% above 45%), catches margin drift before it compounds.

Menu Margin Advisor

Classifies every menu item using the Kasavana-Smith matrix (Star, Puzzle, Plough Horse, Dog) and recommends what to promote, reprice, or cut.

Vendor Alert

Connects guest reviews, social sentiment, and supplier costs. Finds where reputation and revenue intersect.

Conversation 1: "Is our food cost healthy?".

The most common first question. Here's what Cost Controller actually returns:

Cost Controller
Is our food cost healthy?
Your overall FC% is 34.0% against ฿12.48M costed revenue — within the 30-35% benchmark for luxury resorts. Gross profit is ฿8.24M (66% margin). However, you have 6 red-zone items above 45% FC that are dragging the average up: Lobster Extravaganza at 54.3%, Canadian Lobster at 53%, and Hokkaido Scallops at 49%. These items generated ฿1.18M in revenue but cost ฿755K — a combined FC% of 64%. Without them, your overall FC% drops to approximately 31%.

Notice what the agent does: it gives the headline number, benchmarks it, then immediately identifies what's pulling the average up and by how much. It doesn't say "your food cost is 34%." It says "34% is fine, but six items are hiding a problem."

That's the difference between data and intelligence.

Conversation 2: "What should we promote?".

Menu Margin Advisor uses the Kasavana-Smith matrix — the industry-standard framework that classifies items by popularity and contribution margin:

Menu Margin Advisor
What should we promote right now?
Your strongest promotion candidate is the Candlelight Set Dinner — classified as a Star with ฿1,024 contribution margin per cover, 200 covers in 6 months, and 36% FC. It's already popular and profitable. Feature it in sunset and romance packages to grow volume without discounting.

For high-upside items: In-Villa BBQ is a Puzzle — highest CM on your menu at ฿3,915 per cover but only 42 covers in 6 months. Possible causes: low awareness (it may not be on the restaurant menu), high price point (฿7,500), or guests don't know it's an option. Investigate whether front desk mentions it at check-in.

Avoid promoting Hokkaido Scallops — it's a Plough Horse with the worst CM (฿408) among popular items and 49% FC near the red zone.

The agent doesn't just say "promote the In-Villa BBQ." It flags that volume is low, considers why it might be low (awareness, price, visibility), and recommends a specific investigation before assuming you can simply "push" it. That's the kind of thinking that usually takes an F&B director an hour of spreadsheet analysis.

Conversation 3: "Why do guests say we're overpriced?".

This is where it gets interesting. Vendor Alert cross-references financial data with guest reviews:

Vendor Alert
Guests keep saying we're overpriced. Are we?
30% of guest reviews mention F&B pricing, and only 52% of those mentions are positive. But your overall FC% is 34% — a healthy margin. The gap is perceived value, not actual cost.

Cross-referencing: your red-zone items are premium seafood (lobster, scallops, Wagyu) priced at ฿1,800–฿2,800. Guests at a resort paying ฿7K-30K/night may benchmark against local Lamai restaurants where similar dishes cost 60% less. The competition isn't Four Seasons — it's the street.

Your strongest counter-signal: breakfast at Moon gets 88% positive sentiment, runs at 16% FC, and generates ฿3.84M. Guests feel breakfast delivers value. The disconnect is dinner, specifically at premium price points. Recommend reviewing portion presentation and plating on dishes above ฿1,500 — guests may need to see ฿2,475 on the plate.

No single dashboard connects TripAdvisor sentiment with food cost percentages with competitive pricing context. This agent did it in one response. The insight — "the competition isn't Four Seasons, it's the street" — is the kind of reframe that changes how you think about the problem.

"I asked the agent why guests complain about prices when our margins are normal. It told me something I hadn't considered in two years of running this restaurant."

F&B Director, luxury resort

Conversation 4: Going deeper.

Agents remember context within a conversation. You can ask follow-ups:

Menu Margin Advisor
Design a sunset set menu using our best-margin items
Based on your menu matrix, here's a sunset set menu targeting 30% FC:

"Sila Sunset Experience" — ฿2,200
Welcome: Sila Stone Cocktail (signature, high volume)
Starter: Kanom Jeen Set (Plough Horse, ฿452 CM, Thai identity — guests love it)
Main: Grilled Sea Bass (Plough Horse, ฿528 CM, 34% FC — reliable)
Dessert: Chef's selection

Estimated FC: ~28-30%. CM per cover: ~฿1,540. If you sell 8 per night on the rooftop, that's ฿12,320 in contribution margin per evening from a slot that currently underperforms Mon-Wed.

The agent pulled from the menu matrix, selected items with the right margin profile, referenced the sunset opportunity from social media data (highest engagement content), and designed something operationally realistic. Not a theoretical exercise — a menu you could print tomorrow.

Honest Limits

What the agents can't do.

They can't

Access data they haven't been given. If we don't have your supplier invoices, the agent will say "I don't have that data" — never guess.

They can't

Replace your operational judgement. They flag issues and recommend actions — you decide what to implement based on context they don't have (staffing, supplier relationships, events).

They can't

Predict the future. They analyse historical and current data. They don't forecast demand or guarantee results from their recommendations.

They will

Tell you when they're missing data. Every agent is trained to flag gaps explicitly — "I don't have labour cost data to answer this" — so you always know the limits of the analysis.

The agents are a starting point for your decision, not the decision itself. That's by design.

How is this different from ChatGPT?

You could paste your P&L into ChatGPT and ask it questions. Some people do. Here's why this is different:

Try These

Questions real hotel F&B teams ask their agents.

  • "Which items should I cut from the menu?"
  • "Why did food cost spike this month?"
  • "What do guests love most about our restaurants?"
  • "Design a set menu under 32% food cost"
  • "Are we overstaffed for the covers we do?"
  • "What would happen if we dropped our two worst sellers?"
  • "Compare our Google rating to our closest competitor"
  • "What's our most profitable outlet and why?"

The point isn't the technology.

Nobody in hotel F&B wakes up wanting an AI agent. They wake up wanting to know why food cost crept up, whether the new menu is working, and what guests actually think.

The agents are a means to an end. They take the data you already have — POS exports, cost sheets, review platforms — connect it, and give you answers you can act on today. Not next month when the report lands. Today.

That's what they can actually do.

See your own data through the agents.

Send us your POS export and cost data. We'll build your console with agents trained on your numbers. First console is free.

Get your free console