When analysing a menu, should I focus on gross profit or actual contribution per dish?

With a selling price of MOP80 and direct variable costs of MOP40, the contribution per dish is MOP40, as this is the amount available to cover labour and rent. This is only an illustrative calculation. When comparing menus, you should use a consistent costing basis and should not treat contribution per dish as the restaurant’s net profit.

What data do I need to prepare before starting AI menu analysis?

Start by compiling data for four consecutive weeks, as daily sales volumes can be affected by holidays, weather and group bookings. At a minimum, prepare data on dish sales volumes, amounts actually collected, standard recipe costs and discounts, while separating dine-in business from takeaway business. Four weeks is only a starting point and does not necessarily reflect annual demand.

How much does it cost to start an AI menu-engineering project in a small restaurant?

To begin, you can use one spreadsheet alongside data exported from your existing POS system, as there is no need to purchase a new system immediately. The actual costs include time spent cleaning up data, calculating recipe costs and tool subscriptions. You should first confirm that the data is usable, then decide whether paying for an upgrade is worthwhile based on the time saved.

How can AI help me identify the dishes worth promoting most?

AI can categorise dishes into four groups based on sales volume and contribution per dish, as popularity does not necessarily indicate profitability. You can ask the AI to explain its classification criteria and identify missing data, after which the restaurant manager should verify the results. A dish with no cost data should not automatically be classified as a high-contribution item.

A dish sells very well but has low contribution per serving. Should I increase its price immediately?

For a popular dish with low contribution, you should first check three things: portion size, purchase cost and discounts, as raising the price is not the only way to improve performance. You could also test a bundled menu, but you need to calculate the contribution of the bundle as a whole. Change only one factor at a time so that it is easier to determine what caused any change in sales volume.

How can I improve sales of a high-contribution dish that few customers order?

For a high-contribution dish with low sales, you could start by testing one change, such as its name or position on the menu, as customers may simply have overlooked the product’s distinctive features. Keep the price and portion size consistent, then compare sales volumes and total contribution with those of comparable trading days. Avoid making several changes at once, as this makes it difficult to assess the results.

After adjusting the menu, how can I tell whether the changes have genuinely improved performance?

Compare three indicators each week: total contribution, number of servings sold and wastage, as higher turnover does not necessarily mean higher profit. When making comparisons, record holidays, promotions and the number of trading days. If the results cover only a short period, continue monitoring them rather than immediately treating them as a long-term improvement.

Should dine-in and takeaway costs be calculated separately when they use the same menu?

You should use separate costing bases for dine-in and takeaway sales of the same dish, as takeaway orders may involve additional packaging costs, platform commissions and discounts paid by the merchant. Calculate contribution using the amount actually collected and contractual charges for each channel, rather than using dine-in results to judge whether takeaway sales are worth promoting.

What operational risks should I be aware of when providing POS data to an AI analysis tool?

Before carrying out AI analysis, you should remove three unnecessary types of data: customer names, telephone numbers and payment identification data, as menu categorisation mainly relies on sales volumes and costs. You should also check the tool’s data-retention settings and spot-check the calculations. Pricing and delisting recommendations made by the AI must still be reviewed manually.

AI Menu Engineering for Macau Restaurants in 2026: From Identifying the Most Profitable Dishes to Continuous Optimisation — Macao Business Encyclopedia | CloudPipe AI

Macau restaurants are bustling, so why are they still finding it difficult to make a profit? This article examines practical applications of AI menu engineering in 2026. Starting with POS transaction data, ingredient costs and customer ordering behaviour, it builds three modules for calculating dish-level contribution margins, diagnosing slow-moving items and generating pairing recommendations. It also shows SME owners how to test pricing, adjust menu structures, track table-turnover rates and assess payback periods through transparent scenario calculations, establishing a process for continuous optimisation.

The truth about restaurant margins in Macao: strong customer numbers do not necessarily mean profit—menu structure is the key

Macau’s catering industry generated revenue of MOP 15.05 billion in 2024, while expenditure reached MOP 15.1 billion, because growth in turnover may not be enough to cover operating costs.According to the Statistics and Census Service’s 2024 Survey of the Food and Beverage Industry, revenue increased by 3.4 per cent year on year, yet the industry still recorded a loss. The survey excludes food and beverage outlets operated directly by hotels and gaming companies. For restaurant owners, a full house only indicates demand; they also need to consider what customers order and how much profit each dish generates.

Even when both dishes sell 100 orders, a MOP 10 difference in contribution per dish results in a total contribution gap of MOP 1,000, because sales volume is only meaningful when paired with costs.Suppose two dishes both sell for MOP 80 and, after deducting ingredients, packaging and commission per order, generate MOP 40 and MOP 30 respectively. These figures represent contribution towards wages, rent and other expenses, not net profit. The above is purely an illustrative calculation and does not reflect the actual performance of any business.

Menus can first be divided into four categories according to sales volume and contribution per dish, because best-selling dishes are not necessarily the most profitable.It is advisable to compile sales data from the POS system, standard recipe costs and discounts for four consecutive weeks, and then use AI to assist with classification: maintain the visibility of high-volume, high-contribution dishes; review portion sizes and set menus for high-volume, low-contribution dishes; test new names and menu positions for high-contribution, low-volume dishes; and assess replacement options for dishes that are low in both sales volume and contribution. Change only one variable at a time and compare total contribution and food waste each week.

Three local dining concepts can serve as starting points for menu research, because distinctive products help explain customer choices.Wong Chi Kei is renowned for bamboo-pressed noodles, as detailed on the relevant page; Tai Lung Fung is known for its Cantonese music tea house, while Mei Tan is famous for Indonesian cuisine, with further details available in the relevant guide. These are not examples of profitable businesses. For further information on restaurant management and menu optimisation, consult the “Macao Restaurant Management” feature.

AI Menu Engineering Framework: Three Modules for Dish Contribution Margin Analysis, Slow-Mover Clearance and Bundled Recommendations

In 2024, Macau’s food and beverage industry generated revenue of MOP 15.05 billion and recorded expenditure of MOP 15.1 billion, as revenue growth may not have been sufficient to offset cost pressures (source: Statistics and Census Service, 2024 Survey of the Food and Beverage Industry). AI menu engineering can be divided into the following three modules, sharing data on sales, recipe costs and inventory.

1. Dish Contribution Margin Analysis

For a dish selling for MOP 80 with an ingredient cost of MOP 30, traditional menu engineering calculates a contribution margin of MOP 50 by subtracting the ingredient cost from the selling price. This is an illustration rather than net profit. Actual decisions must also deduct variable costs such as packaging and platform commissions. Sales volumes and contribution margins should be classified daily, prioritising dishes with high sales volumes and high contribution margins for promotion.

2. Slow-Mover Clearance

It is recommended that slow-moving dishes first be identified using sales volumes from the past 28 days, as low sales on a single day may be affected by the day of the week and public holidays. AI can then combine shelf-life data, shared ingredients and stock-out records to recommend reducing preparation quantities, adjusting portion sizes or removing dishes from the menu. Dishes with low sales volumes but high contribution margins should first be tested in more prominent menu positions.

3. Bundled Recommendations

For an add-on selling for MOP 20 with a variable cost of MOP 8, each successful bundle adds MOP 12 in contribution because the incremental revenue still exceeds the incremental cost. This is an illustration. AI can recommend beverages or side dishes based on customers’ existing orders, while contribution margin per order can be compared weekly between the recommendation group and a control group, preventing discounts from eroding existing profits.

Three brands provide useful methodological references. Lightspeed excels in its menu engineering framework (see the related introduction); Toast is known for its menu analysis worksheet (see the worksheet page); and Tenzo focuses on interpreting contribution margins (see its dedicated analysis page). For further options, see the topics on “restaurant management systems” and “digital transformation for restaurants”.

Pricing and Bundles: How a Menu Engine Affects Average Order Value and Table Turnover

Macau’s food and beverage industry recorded revenue of MOP15.05 billion and expenditure of MOP15.1 billion in 2024, as growth in turnover may not be sufficient to cover rising costs. An AI menu engine should track average order value, contribution gross profit and table-occupancy time simultaneously. Source: the Statistics and Census Service’s 《2024 Survey of the Food and Beverage Industry》.

Pricing and Bundles: Calculate Incremental Gross Profit

Adding a drink can increase contribution gross profit by MOP12 per order, based on an assumed add-on price of MOP18 and an ingredient cost of MOP6; this is an estimate rather than an actual merchant result. The engine should prioritise high-margin, quick-to-serve pairings and deduct package discounts and platform commissions, avoiding a situation where more items are sold but less profit is earned.

Average Order Value Has Risen, but Table Turnover Should Also Be Checked

Average order value rose from MOP100 to MOP110, while table-occupancy time increased from 60 minutes to 75 minutes. As a result, revenue per seat per hour fell from MOP100 to MOP88 because the increase in dining time exceeded the increase in spending. This estimate assumes full occupancy and demand for waiting tables. During busy periods, drinks that can be served simultaneously with the main course should be promoted, while dessert upselling should be tested during off-peak periods.

Test for two weeks and change only one recommendation at a time, as adjusting prices and packages simultaneously makes it difficult to identify the impact. Compare add-on purchase rates, contribution gross profit per customer, food preparation and service times, and table-occupancy times for the same weekdays and time periods before deciding which changes to retain or withdraw.

Three merchants can be used as examples of dish positioning: Tai Lei Kei is renowned for pork chop buns, Luk Kee Congee & Noodles for lo mein with shrimp roe, and Sam Bun Cheong for hand-pulled beef noodles. Detailed introductions are available on the Tourism Board’s relevant merchant page; these are not examples of AI applications. For more similar food and beverage options, see the “Cha Chaan Teng／Noodle Shops／Coffee Shops” feature.

Clearing Slow-Moving Items: How AI Can Identify Which Dishes Should Be Removed and Which Are Simply Misplaced

Macau’s food and beverage industry generated MOP15.05 billion in revenue and recorded MOP15.1 billion in expenditure in 2024, indicating that clearing slow-moving dishes must balance revenue and costs. The figures come from the Statistics and Census Service’s 2024 Survey of Food and Beverage Establishments; removing dishes solely because of low sales may mistakenly eliminate high-margin options that customers have not noticed.

First, Distinguish Between Items Customers Never See and Items They See but Rarely Order

It is recommended to first integrate four types of data: impressions, orders, contribution margin and wastage, because low sales may be caused by placement, stockouts or insufficient demand. AI should group results by dine-in versus takeaway, time period and language version; electronic menus can measure impressions, while paper menus require controlled version-change tests. Impressions cannot be inferred from point-of-sale sales alone.

Reposition first: Where impressions are low but contribution per dish is high, change the position or image first, altering only one variable at a time.
Improve first: Where impressions are sufficient but few customers order the dish, review its name, price and description.
Consider removal: Where demand remains low after testing and the dish has high wastage, uses unique ingredients or slows down food preparation, seek the manager’s approval before removing it.

When orders increase from 20 to 35 per 1,000 impressions, the conversion rate rises from 2% to 3.5%; this is a hypothetical calculation, not an actual merchant result. It is recommended to compare similar weekdays, time periods and stock availability conditions, and to check whether gross margin per order has increased. If the sample is insufficient, extend the observation period to avoid mistaking the transfer of orders from other dishes for growth.

Brand Value Must Also Be Preserved

Three brand examples can remind operators to maintain the menu’s distinctiveness: Wong Chi Kei is renowned for bamboo-riser noodles, with further details available on the Wong Chi Kei page; Lord Stow’s Bakery and Margaret’s Café e Nataka are each renowned for Portuguese egg tarts, with further details available in the Macau Government Tourist Office’s Macau Delights guide. These are brand examples only and are not examples of the brands’ AI applications. For further information on food and beverage operations and menu management, refer to our dedicated topic on restaurant operations.

Implementation: From POS Order Data to Menu Revisions – A Complete End-to-End Run

Macau’s food and beverage industry recorded revenue of MOP15.05 billion and expenditure of MOP15.1 billion in 2024. Menu revisions should therefore be evaluated using contribution margin, as an increase in turnover does not necessarily mean higher profitability. See the Census and Statistics Bureau’s 2024 Food and Beverage Industry Survey.

Reconcile the figures first, then hand them over to AI

It is recommended to use order data from the most recent eight weeks as a baseline, as this should cover several ordinary weekdays and weekends. This is an operational recommendation rather than a statistical threshold. Export dish codes, quantities, revenue after discounts, refunds, time slots and sales channels, then reconcile these against recipe costs, packaging, platform fees and waste. Separately flag out-of-stock days and remove customer personal data before uploading the data.

Make small-scale changes and retain a rollback version

Each round should test only three dishes and change one variable, as altering prices, photographs and menu positions simultaneously makes it difficult to identify what produced the result. AI should propose hypotheses, which the store manager should then verify. For example, if the selling price is MOP88 and the ingredient and other variable costs are MOP38, the contribution margin per portion is MOP50 before deducting fixed costs such as rent. Menu position could be tested first, after which a decision can be made on whether to adjust the price.

Tool routing and continuous review

The three brands can be assessed by functionality: Lightspeed specialises in menu item popularity reports, as detailed on its menu reporting page; Oracle Simphony focuses on restaurant operations reporting, as explained in its reporting overview; and Possier offers menu engineering reports, as detailed on its analytics features page.

Testing should cover at least two complete weeks, as comparisons must include the same weekdays and service periods. Extend the period if there are insufficient orders. At the same time, monitor contribution margin per 100 orders, waste and fulfilment times, and flag public holidays and promotions. Roll out the changes on a wider scale only once improvements have been confirmed; otherwise, roll back. Update costs and rerun the analysis each month. For further options, see the dedicated topics on restaurant POS systems and digital restaurant operations.

Implementation pitfalls: why popular dishes may need to be cut, and how to reconcile AI recommendations with the boss’s intuition

Macau’s catering industry recorded revenue growth of 3.4% to MOP15.05 billion in 2024, while expenditure reached MOP15.1 billion. This shows that selling more does not necessarily mean earning more, as revenue growth may be offset by costs. The figures are taken from the Statistics and Census Service’s “2024 Survey of the Food and Beverage Industry”.

Start by reducing the prominence of popular dishes; they do not necessarily have to be removed

Suppose a dish sells for MOP80 and leaves MOP20 after deducting variable costs such as ingredients, packaging and platform fees. If it occupies a key stove position for 10 minutes, it contributes MOP2 per minute. Another dish leaves MOP18 and takes only three minutes to prepare, contributing MOP6 per minute. This example illustrates why capacity should also be considered during busy periods. However, if a popular dish drives additional drink purchases, removing it outright could also result in a loss.

Turn intuition into a testable hypothesis

It is recommended to begin by offering the dish for a limited period for two weeks or reducing its prominence, as small, incremental changes are easier to compare. The two-week period is an operational recommendation, not a statistical threshold. The boss should record the judgement that “regular customers come here for this dish”, while the AI sets out the relevant cost evidence. The two can then compare whole-order gross profit, serving times and complaints across the same day of the week, time slot and sales channel. Public holidays should be marked separately. Decision criteria should be established in advance and reviewed by the restaurant manager; the aim should not be to maximise gross margin on individual dishes alone.

Three businesses offer further reading for readers seeking different market positions: the Army Club, known for Portuguese cuisine, is covered on its dedicated page; Wong Chi Kei, known for congee and noodles, has a related page; and Yee Shun, famous for double-skin milk, also has a dedicated page. These are not examples of businesses adopting AI or removing dishes from their menus.

For further information on restaurant management and selecting restaurants in Macau, see the “Macau Restaurants” and “Digital Transformation for the Food and Beverage Industry” features.

Cost-effectiveness: How long does menu optimisation usually take to pay back, and the common pitfalls

Macau’s food and beverage industry generated revenue of MOP15.05 billion in 2024, while expenditure reached MOP15.1 billion. This means that payback should be assessed on the basis of incremental cash income, as increased turnover may still be offset by costs. Sources: the Statistics and Census Service’s “2024 Survey of the Food and Beverage Industry”; the survey excludes food and beverage outlets directly operated by hotels and gaming operators.

How long does it take to pay back? Calculate the monthly net gain first

Assuming a one-off investment of MOP6,000 and a monthly net gain of MOP2,000, the payback period is three months, because the number of months to break even equals the one-off investment divided by the monthly incremental net cash income. This is an illustrative calculation, not the Macau industry average; the official statistics cited above do not provide the payback period for AI menu optimisation.

An additional MOP3,000 in contribution, less MOP1,000 in software and maintenance costs, leaves a net gain of MOP2,000. Upfront costs should include data preparation, design, printing and training. Time saved on food preparation should only be included in the cash return if it actually reduces overtime or expenditure.

Test two menu items first and avoid three pitfalls

It is advisable to test two menu items and monitor performance continuously for four weeks, because changing prices, portion sizes and running promotions simultaneously makes it difficult to identify what is driving the results. Compare the same weekdays and time periods, and record contribution per customer, wastage and refunds. Avoid three common pitfalls: looking only at turnover, overlooking platform commissions, and attributing increases in customer traffic during festivals to the AI.

Choose tools that work with your existing systems

Excel, Power BI and Tableau provide three options for analysis, allowing restaurants to select tools according to their data volumes and in-house maintenance capabilities. Excel excels at spreadsheet analysis; see the product page for details. Power BI excels at report visualisation; see the relevant page for details. Tableau excels at interactive visual analytics; see the product overview for details. For more tool options, refer to the “Digital Transformation for the Food and Beverage Industry” feature.

AI Menu Engineering for Macau Restaurants in 2026: From Identifying the Most Profitable Dishes to Continuous Optimisation