Menu Engineering and Profitability Optimization with AI

A restaurant’s revenue is shaped not only by how many guests walk through the door, but by what they order. Menu engineering addresses exactly this. It means evaluating every item on the menu across two dimensions: how frequently it is ordered and how much profit it generates per order. Following this evaluation, the item’s presentation on the menu, its price, and even its position on the page are optimized.

Items are placed on two axes:

Popularity (order frequency — how often it is ordered)
Profit margin (net profit remaining per unit sold)

The intersection of these two axes creates four groups:

1. Stars ⭐ (High popularity + High margin)

These are the heroes of the menu. They should occupy the most visible position on the menu and continue to be promoted. Items in this group are the most resistant to price increases, because customers already value them and order them.

2. Puzzles 🧩 (Low popularity + High margin)

Profitable but rarely ordered. The problems are usually: the item’s name or description is unappealing, it is in the wrong position on the menu, or staff are not recommending it to customers. The optimization path: renaming, adding a photo, training staff on recommendation strategies.

3. Plowhorses 🐎 (High popularity + Low margin)

Sold in high volumes but generates little profit. These are generally “price-sensitive” items — meaning customers stop buying them the moment the price rises. Strategies may include: revising portion size, running a combo campaign pairing it with a high-margin item, or reducing procurement costs.

4. Dogs 🐕 (Low popularity + Low margin)

There is no business case for these items being on the menu. Either transform them (rename, change presentation, revise the recipe) or remove them from the menu.

What does AI accelerate?

Dynamic classification

With the classic approach, you produce the matrix manually twice a year. AI keeps the same work updated every day. It builds different matrices based on factors like day of the week, time of day, weather, and season. For example, the matrix for a Sunday brunch menu is different from the one for a Tuesday business lunch menu.

Price elasticity testing

Suppose you want to know how sales will be affected if you raise an item’s price from one level to another. Testing this manually takes three to four weeks. AI can run micro-tests on a per-segment basis. It shows the new price to 10% of customers and the old price to the remaining 90%. Data is collected, and once a statistically significant difference emerges, the new price is rolled out across the entire menu.

Upsell and cross-sell recommendations

When a customer selects their main course, which dessert should be recommended to achieve the highest acceptance rate? AI analyzes combinations in past orders to extract patterns. Moreover, this recommendation is not one-directional — it generates personalized recommendations at the individual customer level.

On a physical menu, the “upper right corner” (where the eye lands at the end of the Z pattern) receives the highest attention. On a digital menu, items visible on the first screen have an advantage. AI shows with real data which item performs better in which position and makes recommendations accordingly.

The most powerful intersection emerges here. A few example scenarios:

Pushing a Puzzle

“Truffle risotto” is a Puzzle item. This week it is offered to loyalty members at a 20% discount.
The member tries the item for the first time thanks to the campaign. Because they are now familiar with it, they may continue to order it after the campaign ends.
Result: a shift from low popularity toward moderate popularity is achieved.

Protecting a Star

“Classic cheesecake” is a Star item, but its stock position is volatile. You can build a stock-linked campaign.
On days when stock is high, members receive a “cheesecake is ready today” reminder.
Result: the Star item’s revenue remains stable and stock waste is reduced.

Plowhorse combo

“Flat white” is a Plowhorse item — unprofitable on its own.
A loyalty member is offered this deal: “Flat white + croissant for [bundled price].” The croissant has a high margin.
Result: basket size grows and the combination becomes profitable.

Critical caveat: value perception

While AI is optimizing prices, it must not erode customers’ perceived value of an item. Once the perception of “this item got more expensive, the brand has declined” takes hold, loyalty loss begins. Recommendations from AI need to follow these rules:

They should be differentiated by segment — the same price increase should not be applied to everyone
They should be rolled out in a way that does not distort the overall menu perception
They should not change too frequently. Seeing prices change every week creates psychological risk.

Data requirements

The minimum data requirements for AI-assisted menu engineering are:

90 days of POS receipt line items (product, quantity, price)
Product cost information (preferably detailed recipe-level costs)
Campaign history (how long the effect of each campaign lasted)
Season labels (holidays, school breaks, weather)

Without a platform that collects and processes this data, menu engineering is done by hand. Done manually, it can only be updated two or three times a year. That pace is not in sync with the speed at which market conditions change.

Conclusion

When done correctly, menu engineering lifts a restaurant’s profit margin by 3 to 8 percentage points. For most restaurants, this translates to a 15% to 40% increase in total profit. Moving from manual analysis to data-driven continuous optimization is becoming the standard for restaurant operations in 2026.