How to Use AI for Menu Engineering in Restaurants

Most menus carry a few dishes that quietly lose money and a few that could earn far more. Menu engineering is the method that finds them. It sorts every dish by two things, how often it sells and how much profit it makes, then tells you what to do with each one.

The maths is not hard, but doing it by hand across 60 dishes with costs that change every week is where owners give up. That is the part AI now handles. It pulls your sales and recipe costs from the POS and rebuilds the picture on its own, so the analysis is never out of date.

This guide covers the method in full, the two numbers behind it, a worked example you can copy, and how AI does the heavy lifting. Across the restaurants we work with, this is one of the clearest paths to more profit without a single new customer.

Key Takeaways

• Menu engineering ranks each dish by popularity and profit, then labels it a star, plowhorse, puzzle or dog.
• The two numbers you need are menu mix percentage and contribution margin.
• AI pulls both from POS sales and recipe costs, so the matrix updates on its own.
• Each category has one action: promote stars, re-work plowhorses, reposition puzzles, cut dogs.

What Is Menu Engineering?

Menu engineering is a data method for making a menu more profitable, first introduced in 1982 by Michael Kasavana and Donald Smith (Menu engineering, Wikipedia). It studies each dish on two axes and groups it into one of four types.

The method itself has not changed in 40 years. What changed is the effort. Doing it well means fresh sales counts and an up-to-date food cost for every dish. That is a lot of manual work, and margins in Indian restaurants stay tight.

The food services industry was valued at ₹5,69,487 crore in FY24 (NRAI India Food Services Report 2024). Yet a lot of that revenue passes through kitchens that still run on gut feel about which dishes actually earn.

AI removes that grind. It reads what sold from your POS and the cost of each recipe, then keeps the matrix current without anyone opening a spreadsheet. For the broader picture of where this fits, see our overview on AI for restaurants.

The Four Menu Categories

Every dish lands in one of four boxes, set by whether its popularity and profit are high or low. This is the whole model in one view.

The names tell you the plan. Stars sell well and earn well. Plowhorses sell well but earn little. Puzzles earn well but sell slowly. Dogs do neither. Each box needs a different move, which we cover further down.

The Two Numbers Behind the Matrix

To place a dish, you need two figures for it. Both come straight from data you already hold.

The first is menu mix percentage: how big a share of total dishes sold this one dish is. If you sold 1,000 plates and 220 were paneer butter masala, its menu mix is 22%. A dish counts as popular if its share beats a cut-off, usually 70% of an equal share. The menu mix percentage figure is the popularity axis.

The second is contribution margin: the selling price minus the food cost of that dish. A ₹240 dish that costs ₹96 in ingredients has a contribution margin of ₹144. This is where a costed recipe matters, so it helps to first know your food cost per item.

A dish is “high profit” if its contribution margin sits above the average across the menu. Popular plus high-margin puts it in the star box. The two cut-offs, the popularity line and the average margin, are the only thresholds the whole method rests on.

A Worked Example You Can Copy

Take a small cafe in Koramangala selling five items over a week (an illustrative example, not a real outlet). Here is the same maths AI runs, shown by hand.

With five items, an equal share is 20%, so the popularity line sits at 14% (70% of 20%). The average contribution margin is ₹87. Now read each row against those two cut-offs.

The biryani sells below 14% but beats ₹87 in margin, so it is a puzzle. The masala dosa sells hugely but earns under ₹87, which makes it a plowhorse. The gulab jamun fails both tests, so it is a dog. The same logic scales to a 200-dish menu; only the row count grows.

How AI Does This Faster Than a Spreadsheet

The example above took five rows. A real menu has sixty, with add-ons, variants and half-plates, and ingredient costs that move every week. That is where the by-hand version dies. AI keeps it alive.

It reads the units sold from your POS, so the menu mix is always live. It reads the food cost from each costed recipe, so the contribution margin updates the moment a supplier price changes. Nobody re-types a number. Our guide on how to use your POS sales report shows the raw feed this runs on.

The real gain is the alert, not the table. When tomato prices jump and a star quietly slips into plowhorse territory, AI flags the shift the week it happens, not at the next quarterly review.

A dish can also flip the good way. A cheaper supplier or a smaller portion can lift a plowhorse back into star range, and the tool shows that the moment the new cost lands. You can size a single dish yourself first with a recipe costing calculator to see the effect.

What to Do With Each Category

Sorting the menu is only useful if it changes what you do next. Each box has a clear move.

• Stars: protect and push them. Put them where the eye lands first on the menu, keep quality steady, and never discount what already sells and earns.
• Plowhorses: these are popular, so tread carefully. Trim the food cost through portion or supplier changes, or nudge the price up in small steps. Our note on menu pricing strategies covers how to raise a price without losing the crowd.
• Puzzles: they earn well but hide. Rename them, move them higher on the card, train staff to suggest them, or bundle them with a star.
• Dogs: they do neither job. Rework the recipe once, and if it still fails, cut it to free up prep space and stock.

One caution. Do not touch a plowhorse and a puzzle in the same week and expect to read the result. Change one lever, watch the next fortnight, then move again. A menu is a system, so a price rise on one dish can push guests towards another, and you want to see that shift cleanly before the next change.

How to Start Menu Engineering With AI

You do not need a project plan. You need clean data and one first pass.

1. Cost your recipes. Every dish needs an accurate food cost, since the whole matrix leans on it. This is the step most menus skip.
2. Pull a sales period. Take at least two to four weeks of POS sales so the menu mix is not skewed by one odd day.
3. Let the tool sort the menu. With costs and sales in place, the matrix builds itself, and you get four lists instead of a wall of dishes.
4. Act on one box first. Start with plowhorses or dogs, where the money leaks fastest, and leave the rest for the next round.

A POS that already holds your sales and recipe costs makes this a report rather than a project. That is the case for a platform like Petpooja POSS. Its sales report gives you the menu mix for each dish, and recipe costing gives the food cost, so both axes of the matrix come from one place.

Petpooja’s AI features keep that food-cost number honest. On the buying side, Best Match and live price comparison track ingredient rates as they move. A dish’s contribution margin then updates on its own, rather than going stale between reviews. That is what turns menu engineering from a quarterly chore into a live view, and the P&L from your POS shows the result in your final numbers.

Conclusion

Menu engineering is one of the few profit levers that costs nothing to pull. You are not buying ads or cutting staff, only reading your own menu with clear eyes and acting on what it says.

The method is 40 years old and still works. What AI adds is speed and freshness, turning a quarterly spreadsheet chore into a live view that flags a slipping dish the week it slips. Start by costing your recipes, pull a few weeks of sales, and let the matrix show you where the money is hiding.

Frequently Asked Questions