AI-narrated version of this post using a synthetic voice. Great for accessibility or listening while busy.
Menu Engineering Is Not Just Graphic Design
Most restaurant owners hear “menu engineering” and think it means making the PDF look nicer. It does not. Menu engineering is the practice of analyzing dish profitability against popularity, then strategically positioning items to push margins higher. Cornell University hospitality researchers codified it in the 1980s. The basic matrix – stars, plowhorses, puzzles, dogs – is still used by chain operators spending thousands on consultants. Independent operators rarely touch it because they assume it requires expensive software or an MBA.
AI tools have quietly changed that equation. Not perfectly, not magically, but enough that a solo owner running a 40-seat neighbourhood spot in Hamilton or Victoria can do meaningful menu analysis in an afternoon. This post covers the specific workflow, the tools worth using, and the ones that overpromise.
The Core Workflow: Four Stages Where AI Actually Helps
Stage 1 – Profitability Classification
You need two numbers per dish: contribution margin (selling price minus food cost) and a rough popularity rank from your POS data. If your POS exports a CSV, you are ready. If not, a two-week manual tally on a spreadsheet works fine.
Feed that data into ChatGPT-4o or Claude 3.5 Sonnet with a structured prompt. Ask it to classify each item into the four quadrants, flag items where your food cost percentage is above 32 percent, and suggest which “puzzle” items might be worth a price test versus a quiet removal. Both models handle this competently. Claude tends to give slightly more conservative recommendations, which is actually useful – you do not want an AI cheerfully telling you to axe your chef’s signature dish without surfacing the loyalty risk.
One honest limitation: neither tool knows your local food cost reality. Beef prices in Calgary differ from beef prices in Toronto, and neither model has your supplier invoices. You must provide accurate food cost numbers yourself. If you feed it guesses, you get confidently wrong quadrant placements.
Stage 2 – Menu Description Rewriting
This is where AI earns its keep most clearly. Weak menu copy costs you money on every single transaction. “Grilled salmon with vegetables” is a dead description. It communicates nothing about texture, sourcing, or differentiation. Research consistently shows that sensory and origin-specific language increases perceived value and actual order rates.
The workflow here is simple but specific. Give the model: the dish name, the actual ingredients, your restaurant’s tone (casual, upscale, comfort-focused), and one or two origin or preparation details your kitchen is genuinely proud of. Ask for three description variants at different word counts – short for tight layouts, medium for standard menus, long for digital or QR menus where space is less constrained.
Jasper markets itself for this use case but is overpriced for independent operators and its restaurant-specific templates are shallow. You get better results with a well-constructed Claude or ChatGPT prompt at a fraction of the cost. Copy.ai has a similar problem – decent output, questionable value when the underlying models are accessible directly.
What actually works: a reusable prompt template you build once and reuse per menu cycle. Something like – “You are writing menu descriptions for [restaurant name], a [cuisine type] restaurant with a [tone adjective] voice. Our guests are [brief demographic note]. Rewrite the following dish description to emphasize sensory detail and origin. Avoid cliches like ‘decadent’ or ‘mouth-watering’. Dish details: [paste here].” That structure consistently outperforms generic AI writing tool interfaces.
Stage 3 – Pricing Strategy and Psychological Positioning
Pricing is where operators are most nervous and where AI assistance is most underused. The classic menu psychology literature – anchor pricing, charm pricing, removing dollar signs, price clustering – is well-documented but hard to apply when you are staring at 30 line items trying to hit margin targets.
Use a spreadsheet alongside your AI conversation. Paste your current prices, your target food cost percentages, and your three highest-margin items. Ask the model to identify whether your current pricing structure has a clear anchor item (the deliberately expensive option that makes mid-range items feel reasonable), whether your price spread is creating psychological friction, and whether any high-margin items are underpriced relative to their perceived value category.
This is genuinely useful analysis. A useful catch from a real test session: a pasta dish priced at $18 sitting next to a steak at $28 is working as an anchor. Move that pasta to $19 or $20 and you erode the anchor effect while barely moving the steak’s appeal. An AI prompt focused on price architecture catches that kind of structural issue that owners miss when they look at margins line by line.
Important caveat for Canadian operators: menu pricing analysis must account for HST or GST/PST treatment on food versus alcohol, and how that affects your stated versus effective margins. AI models will not automatically apply Canadian tax logic. You need to flag whether your prices are pre-tax or tax-included, and whether your margin calculations are pre or post-tax. This sounds basic but it is a consistent source of error when operators try to use these tools without flagging the Canadian tax context explicitly.
Stage 4 – Seasonal Menu Planning and Substitution Logic
This stage is less about optimization and more about operational efficiency. AI is genuinely helpful for generating seasonal menu variation candidates based on your existing dish architecture. Give it your current menu structure, flag which items are fixed (because of equipment, staff skill, or guest loyalty), and ask for five seasonal substitution suggestions per category that use overlapping prep components to minimize kitchen complexity.
This is where Perplexity adds value that pure chat models do not – it can pull current seasonal produce information by region, which matters if you are trying to build genuinely local sourcing claims into your menu. Ask Perplexity what is in peak season in Ontario in October, then take that list into Claude or ChatGPT for the actual menu development conversation.
Tools That Are Not Worth Your Time for This Workflow
Several tools specifically market to restaurants and hospitality. Most are disappointing for menu engineering specifically.
- Popmenu AI – good for digital menu management and SEO, weak for actual engineering analysis. Its AI features are mostly description generation wrappers around basic models with heavy markup on the subscription.
- MenuDrive – an ordering platform, not an engineering tool, despite occasional AI marketing language on their site.
- Lightspeed’s AI features – useful for sales reporting but the AI layer does not do quadrant analysis or pricing architecture work in any meaningful way as of this writing.
The honest answer is that no dedicated restaurant AI tool currently does menu engineering as well as a thoughtful prompt workflow using Claude or GPT-4o with your own data. That may change. For now, the vertical SaaS layer is adding cost without adding capability.
What This Workflow Cannot Do
Be honest with yourself about the limits. AI menu engineering assistance cannot replace actual sales data. If you are a new restaurant without six months of POS history, the quadrant analysis is speculative. Run the analysis with the data you have, but hold conclusions loosely.
It also cannot account for the intangible loyalty value of certain dishes. Some items are dogs by the numbers and irreplaceable by the culture of your restaurant. Your grandmother’s pierogi special may have a 38 percent food cost and sell 12 covers a week. No AI should tell you to remove it, and a good prompt will not – but you need to tell the model that some items are off the table for removal consideration.
Finally, none of this replaces tasting, service observation, or talking to your regulars. The workflow is an analytical layer, not a replacement for hospitality instinct.
Getting Started Without Overthinking It
If you want to run this for your restaurant this week, start with Stage 2 only. Pick your five lowest-performing dishes by sales – items guests order least. Rewrite their descriptions using the prompt structure above. Update your menu. Run it for 30 days and track whether order frequency shifts. That single experiment will tell you more about whether AI-assisted menu work is worth your time than any amount of reading about it.
The full four-stage workflow is a half-day project. It does not require a consultant, a hospitality degree, or expensive software. It requires your POS data, honest food cost numbers, and a willingness to treat your menu as a strategic document rather than a list of what your kitchen makes.
Related Auburn AI Products
Building content or automations around AI? Auburn AI has production-tested kits: