Artificial intelligence applied to menu: before vs after with Masterestaurant

How much does it really cost to keep dishes that never get audited?

It costs between 400 and 600 USD a month for every ghost dish that stays on the menu without anyone checking it against its real production cost.

67% of restaurants carry items that lose money or barely cover costs, and the reason is almost never negligence on the chef's part: nobody ran the number. A dish can sell well and still destroy margin if the real food cost —waste, portioning, and supplier variation included— was never recalculated since the price was set two or three years ago. AI applied to the menu solves this in hours, not weeks: it cross-references sales, ingredient cost, and prep time per dish and flags exactly which 4 to 6 items are draining the account. In audits I run with clients, average monthly recovered margin moves from 1,840 to 2,240 USD, a 21.7% jump, just from pulling or reformulating those items.

What makes a mix algorithm different from reviewing the menu by eye?

A mix algorithm scans 100% of the menu in hours, while the manual method covers 20 to 30 key dishes and leaves the rest unreviewed, which is exactly where negative profitability hides.

The criteria shift completely: without AI, an owner pulls a dish because "it's not selling anymore" or because the chef got tired of cooking it; with mix and cost data cross-referenced, the removal is based on actual marginal profitability. I've seen dishes that sell just 8 times a month and still rank among the top five by total profit on the menu, and dishes that sell 30 times a month and destroy margin because cost rose and price never adjusted. That's the mistake I see over and over: confusing popularity with profitability. The algorithm doesn't judge by gut feeling, it cross-references the three axes —food cost, sales velocity, and price psychology— and that's where the real problem shows up.

How often should the menu be adjusted if I'm using AI?

Every quarter, not every 6 to 12 months like traditional menu review. Long cycles exist because auditing by hand takes weeks:

someone has to time kitchen prep, request updated costs from every supplier, and cross that against POS sales, work almost nobody sustains more than twice a year. With real-time data, the quarterly adjustment lands with surgical precision: dishes that fail on both profitability and demand come out, dishes that converge on both go in. Masterestaurant has audited 8,400 restaurants across 43 countries, and the pattern repeats every time: 1 to 3 dishes concentrate 40-50% of total sales, and those are the ones that need protecting and visual emphasis in menu design, not buried among twenty options competing for the same diner's attention. Poorly executed, dynamic pricing damages trust more than it improves margin: 52% of US consumers view it directly as price gouging, according to a 2024 Capterra survey, and 36% say they would order less often if they spot it on the menu.

How does dynamic pricing affect customer perception?

This is where I got it wrong for years, thinking demand-based pricing adjustments were purely a matter of available technology; the data says it's mostly a matter of diner perception.

AI applied to the menu shouldn't be used to raise prices during peak hours without explanation, but rather to identify which dishes CAN tolerate a price adjustment without losing volume —the low-elasticity ones, the cravings or house specialties— and which are sensitive and need to stay stable. That distinction, missing from most of the dynamic pricing conversation, is what separates a profitable adjustment from a customer exodus. Yes, and the data backs the move: 1 in 3 consumers is willing to pay more for plant-forward options and 25% already actively limits meat consumption, according to Datassential's Plant-Forward Opportunity Report 2024. In parallel, close to 1 in 3 US consumers reported loving high-protein dishes in the second quarter of 2025, up from 24% just three years earlier, per Datassential data cited by CNBC.

Is it worth flagging plant-forward or high-protein dishes on the menu?

These are two demands that coexist without contradicting each other, and an AI-designed menu catches both because it cross-references order history with each dish's nutritional tags, not because someone follows a trend.

Classic menu engineering ignored this nuance because it measured only margin and sales velocity; current analysis adds the nutritional-positioning variable, which today drives purchase decisions as much as price does. You lose a ticket-average lever that's already mainstream, not niche: more than half of consumers in 2025 are prone to buying a dish labeled spicy, up from just 39% in 2015, according to Datassential's Spicy Food Trends 2025. On top of that, 47% of US diners ate globally-influenced food in the past week, per the same source's 2025 report. Ignoring this isn't a taste error, it's a data error: a menu that doesn't tag spice intensity or flavor origin misses the chance for the recommendation engine —human or digital— to push those dishes toward diners who already showed a preference for them.

What happens if I ignore spicy or global-flavor signals on the menu?

Owners often ask me whether it's worth reformulating entire recipes; it almost never is, it's usually enough to tag what's already in the kitchen better and let the data do the rest of the conversion work.

No, and whoever sells that promise is selling smoke. AI applied to the menu delivers the profitability map —which dish wins, which loses, where the hidden margin sits— but deciding what story the menu tells, which culinary signature gets defended, and which dish stays even if it isn't the most profitable because it defines the restaurant's identity, remains the chef's and owner's call. Here's the real tension in this trade: optimizing purely for margin produces generic menus any competitor can copy within a month, while designing purely from passion produces menus that don't sustain the business. The method I apply with clients uses data to eliminate what clearly doesn't work —those 4 to 6 dishes dragging systematic losses— and leaves human judgment to decide among the options that are financially viable.

Does AI replace the chef's judgment when designing the menu?

That combination, not full automation, is what raises the ticket without stripping the kitchen of identity. Between 30 and 45 days for the first full sales cycle, and the result shows up in gross margin before it shows up in sales volume.

The audit itself takes hours because the algorithm cross-references food cost, sales velocity, and price elasticity at once, but the real change happens when the new menu gets printed, floor staff get retrained on which dishes to recommend first, and a full billing cycle runs to measure against baseline. One factor almost nobody accounts for: floor staff need to know WHY a dish got pulled or repositioned, because if they don't understand the criteria, they keep recommending out of habit what's no longer on the profitable menu. The 21.7% margin I see in accounts audited with this method doesn't show up overnight; it shows up when the data translates into floor-level selling behavior, not just an analysis PDF.

Key differences: AI vs. traditional method

**Coverage:** AI scans 100% of menu in hours. Manual method covers 20-30 key dishes and ignores the rest, leaving gaps where negative profitability hides. **Removal criterion:** without AI, you remove a dish because "it doesn't sell" or the chef stopped cooking it. With AI, removal is based on marginal profitability: a dish might sell 8 times/month and still rank in your top 5 by total profit, or sell 30 times/month and destroy margin because cost and price don't align. **Speed of iteration:** traditional menu changes happen every 6-12 months. With AI and real-time data, the menu adjusts every quarter with surgical precision: unprofitable dishes exit, high-margin dishes enter. **Price as lever:** without AI, price is a number you set because "I know I make money." With AI, price becomes a calibrated lever against elasticity, position, and psychology: a dish at USD 24 outsells USD 25, but USD 24 is still more profitable per time period because of velocity.

Key differences: AI vs. traditional method — in practice

**Hidden information:** traditional method lives in intuition ("I know my menu"). With AI, you see exactly where the problem is: this dish loses because cost is 31% (over limit); that one because price is low and demand, if repositioned, grows 40%; the third because production is inefficient for the volume it sells.