Use AI to Plan a Weekly Specials Menu That Makes Money

To use AI to plan a weekly specials menu, give it four things it cannot guess: what is in your fridge and when it expires, how much prep time your team has, what your base menu already buys, and the food-cost percentage you need to hit. Then ask for a short table of options, not a recipe essay. Without those inputs you get a pleasant, generic list. With them you get specials that use up stock and still earn their place on the board.

Why generic AI specials fail in a real kitchen

A request like "give me five specials for next week" produces five dishes that sound right. None of them know that you have eleven kilos of squash to move before Thursday, that the second cook is off Tuesday, or that the braising pot is the only one big enough for the soup.

Specials exist to do three jobs at once: move stock before it turns, give regulars a reason to come back, and carry a healthy margin. Every one of those depends on facts that live in your head or your stock sheet. The assistant is a fast thinking partner. It is not a person standing in your kitchen.

Build the kitchen reality sheet

Keep one short note and update it each week. Paste it above every request.

• Use first. Ingredients you hold, with quantity and use-by date.

• Prep capacity. Hours of prep available per day, and which days are short-staffed.

• Equipment limits. One oven, one fryer, one burner free, whatever binds you.

• Base menu ingredients. The things you already order every week, so specials can borrow from them instead of adding new lines to your delivery.

• Your costs. Per-unit prices you actually pay. The assistant does not know them, and if you leave this out it will invent plausible numbers.

• Target food cost. The percentage of menu price you aim to spend on ingredients. You choose this number, based on your own books.

The prompt

You are helping plan next week's specials for a small restaurant.
Use only the facts below. If you need a price I have not given you,
ask me instead of guessing.

USE FIRST: [item, quantity, use-by date] ...
PREP: [hours per day, short-staffed days] ...
EQUIPMENT LIMITS: ...
BASE MENU INGREDIENTS WE ALREADY BUY: ...
MY COSTS PER UNIT: ...
TARGET FOOD COST: [percent]

Give me 5 specials as a table with these columns:
dish, main ingredients, which USE FIRST items it uses,
extra ingredients I would have to buy (should be few),
prep minutes, plate cost from MY COSTS, menu price needed to hit
my target food cost.

Flag any dish that needs a day I am short-staffed.
Do not state allergen information. I will handle that myself.

Two instructions in there are deliberate. "Ask me instead of guessing" stops invented prices. "Do not state allergen information" avoids having a confident-sounding line on your board that nobody verified.

The costing formula, in one line

The assistant can do this sum, but you should be able to check it:

menu price = plate cost divided by target food cost

An illustrative example, with made-up numbers: if a squash and sage risotto costs 3.60 to plate and you aim for 30 percent food cost, the price is 3.60 / 0.30 = 12.00. If the table says 9.50, either the plate cost it used is wrong or the margin is thinner than you intended. Spot-check one row by hand every week. If one is wrong, ask it to show its working for all of them.

Read the table like an owner

When the five options come back, sort them by what they do for you, not by how appealing the name is.

Pick three, not five. A board with too many specials slows the kitchen and blurs the choice for the customer.

The part a human must do: allergens

Allergen information is a legal and safety matter in most countries, and it is yours to get right, from the actual recipe and the actual supplier labels. An AI can easily overlook a hidden ingredient or assume a recipe that differs from yours. In the UK, the government publishes guidance for food businesses on providing allergen information and handling allergens. If you trade elsewhere, look up the rules that apply to you.

A simple routine: after you choose the three specials, write the final recipe, check each ingredient label against your allergen list yourself, then have a second person check it before the board is written.

Close the loop at the end of the week

Specials only improve if you feed the results back. On Sunday, note for each special how many sold, how much was left, and what was wasted. Paste that into the next week's request with one line: "These sold, these did not. Plan accordingly." Over a month you build a record that is worth more than any suggestion engine, because it is about your customers.

Related reading on the same theme: using AI to plan stock for a busy week pairs well with this routine, planning a seasonal promotion with AI covers the bigger campaigns, and a QR code menu for a restaurant makes it easy to update the board digitally. The wider picture is in AI for small business.

If a key ingredient arrives late, chasing a supplier with AI covers the messages to send.

Questions people ask

Can AI write my restaurant specials menu?

It can draft options quickly, but it only works well when you give it your stock, prep limits and costs. Treat the output as a shortlist to choose from, and finish the recipes and allergen checks yourself.

How do I price a special?

Divide the plate cost by your target food-cost percentage. A plate that costs 3.60 to make at a 30 percent target needs a menu price of 12.00. Use your own real ingredient prices.

Should I let AI list allergens on the menu?

No. Allergen information needs checking against the real recipe and supplier labels by a person who is accountable for it. Ask the assistant to leave it out, and do it yourself.

How many specials should a small restaurant run at once?

Two or three is a common choice for a small kitchen. More can slow service and make the board harder to read.