How to use AI to cost and optimize your menu

Why AI doesn't replace your costing method — but speeds it up several times over.?

AI is a calculator with natural language, not an accountant with judgment. What it does well:

take a recipe dictated by voice or written by hand and turn it into a structured recipe card in under 90 seconds, compared to the 12-18 minutes it takes an operator without a tool. What it cannot do: decide whether your 34% food cost is acceptable given your average check, your market, and your concept. That judgment remains yours. At Masterestaurant we call it the 32% rule: no dish goes on the menu with a food cost above 32% of the sale price, and AI only tells you whether you hit that target — you decide what to change if you don't.

Step 1 — Build your ingredient database with current prices before turning on AI

The mistake I see over and over is owners asking AI to "calculate the dish cost" without feeding it real prices. The result is a phantom number. Before opening any AI tool, export your latest supplier invoice to CSV or photograph it; with a short prompt the AI converts it into an ingredient table with price per gram in under 2 minutes. Updating them manually takes 4-6 hours per month; with AI the same process drops to 35-50 minutes. Diego F. Parra recommends locking in this routine every first Monday of the month.

Step 2 — Generate recipe cards with AI and validate them with your head chef in 15 minutes

An AI-generated recipe card has three internal steps: the model receives ingredients and quantities, calculates food cost per portion, and generates the expected yield after cooking. The critical point is yield: AI uses bibliographic averages, but your kitchen may operate differently depending on equipment and cut. That is why the Masterestaurant protocol requires the head chef to cook the dish once, weigh the actual results, and correct the yield factor in the card before approving it.

Step 3 — Use AI to classify your menu using menu engineering (Stars, Plowhorses, Puzzles, Dogs)

The Miller and Kasavana menu engineering framework (1982) divides dishes into four quadrants by popularity and contribution margin. Applying it manually to a 40-item menu takes around 3 hours with POS data; with AI and a monthly sales export, the same analysis takes 8 minutes. The correct contribution margin is sale price minus food cost — do not subtract payroll, rent, or utilities from the dish; those belong in the global breakeven calculation. The average casual dining restaurant has between 6 and 9 "Dogs" consuming inventory without generating cash. Eliminating or redesigning those dishes frees up capital tied up in inventory, in Diego F. Parra's experience advising restaurants.

Step 4 — Optimize prices with AI without breaking perceived value

Once dishes are classified, AI can simulate pricing scenarios in seconds. Give it three variables: your current food cost, your target contribution margin, and the current sale price, then ask for the minimum, suggested, and ceiling prices for each dish. If your average falls below that band, AI flags it in seconds. The limit: AI does not know how much your customer's wallet can handle or what your competitor two blocks away is charging. That requires field benchmarking, not a language model.

AI recipe cards vs. manual cards: what actually changes in real operations

AI-generated recipe cards are not more accurate by default — they are faster and more consistent in format. Accuracy depends on input quality: an ingredient with an ambiguous name («white onion» with no weight or trim factor) produces a cost just as inaccurate as one done by eye. Where AI wins unambiguously is at scale: a restaurant launching a new menu of 22 dishes can have all preliminary recipe cards in 35 minutes, versus 2-3 days of administrative work. That shortens the menu development cycle, cutting the average launch time from days down to a fraction using an AI-assisted workflow.

How to integrate AI into your weekly costing routine without becoming dependent on it?

The most expensive trap is delegating judgment, not just the task. The workflow that holds up in operations of 1 to 5 locations: Monday, update supplier prices with AI (35 min);

Wednesday, review food cost variances on high-volume dishes (the top sellers represent most of the sales in most casual restaurants); Friday, run the menu engineering analysis if any prices changed. AI executes; you interpret. If the lomo saltado food cost climbed from 29% to 34% in two weeks, AI flags it — but you are the one who calls the beef supplier or adjusts the portion size. That decision is worth more than any algorithm.

The only number that matters: total contribution margin, not food cost in isolation

A low food cost does not guarantee profitability. For example, a dish with a lower food cost percentage sold a few times a day can generate less cash than one with a higher percentage sold many times over. That is why Diego F. Parra insists that AI must always calculate the total contribution margin of the sales mix, not just the unit food cost. The math is simple: add (sale price − food cost) × units sold per dish. The restaurant with the best mix is not the one with the cheapest dishes to produce, but the one selling the highest volume of dishes with the greatest absolute contribution margin. With AI this is calculated in 3 minutes from a weekly sales report; without AI, the same analysis takes between 90 minutes and half a day.