AI menu engineering can help a restaurant interpret item sales, recipe costs, modifiers and channel differences. The objective is a better reviewed decision about availability, placement, portion, price or recipe. It is not an automatic instruction to raise the price of every popular item.
Menu analysis becomes misleading when recipe costs are stale, modifiers are ignored or gross margin is confused with contribution after channel fees and discounts. Establish definitions before asking a model for recommendations.
This guide describes Binarify’s proposed implementation approach, not a client result. See our AI consultancy for restaurants and food businesses for the wider assessment.
Define the economic view
Choose the measures that match the decision. Ingredient cost and gross margin may support recipe review, while incremental contribution can be more useful when comparing channels or promotions. Allocated rent and head-office costs serve a different purpose and should not be mixed into an item recommendation without explanation.
Build an item-level view with:
- quantity, net sales and discounts;
- modifier and bundle behaviour;
- active recipe, yield and latest applicable ingredient costs;
- packaging and identifiable channel costs;
- voids, refunds and complimentary items;
- availability and stockout periods;
- effective dates for menu, recipe and price changes.
Show which inputs are complete. An item with unknown recipe cost should not be ranked as highly profitable because its cost field is blank.
Reconcile menu identities across channels
The same dish may have different names or prices at the till, on the website and through delivery platforms. Create a stable internal item and modifier mapping. Preserve channel-specific versions where packaging, portion or fee treatment differs.
Use AI to propose mappings from inconsistent descriptions, then have an owner confirm them. Do not combine two items solely because their names are similar. Record when mappings begin and end so prior periods remain interpretable.
Check the existing reporting capability
Toast’s menu-report documentation describes item and modifier reporting and can show cost, gross profit and gross margin when cost data is configured. Other POS products provide similar reports. Confirm whether the existing report answers the question before creating a new analytical layer.
If a restaurant cannot trust recipe costs or channel mappings, the priority is data maintenance. A language model summarising an unreliable margin report only makes the mistake easier to read.
Move from observation to decision
1. Establish a stable comparison period
Account for openings, closures, promotions and limited availability. Low sales during a long stockout do not demonstrate low demand. New items need sufficient exposure before comparisons are useful.
2. Segment by location, daypart and channel
An item can contribute differently by location or channel because price, tax treatment, packaging or commission differs. Aggregate reporting may hide the decision that a manager can actually change.
3. Generate reviewable options
AI can prepare questions and scenarios: update a description, improve placement, test a portion, review a recipe, remove an item from one channel or trial a price. Show the supporting figures and uncertainties. Culinary feasibility and brand positioning remain part of the review.
4. Approve and version the change
Record the decision, owner, effective date and systems that need updating. Allergen and ingredient records need their governed update process. Do not let generated copy introduce an ingredient or dietary claim that the approved recipe does not support.
5. Measure the response
Compare a suitable before and after period, or use a controlled location test where practical. Check item contribution, product mix, total order value, substitution, waste and customer feedback. A higher margin percentage can still produce lower total contribution.
Example: a popular delivery item
Illustrative scenario: a menu item has strong delivery-platform sales but appears weak in the restaurant’s headline gross-margin report. The review finds that its delivery price and packaging were excluded from one data source while a modifier cost was missing from another.
The team corrects the mapping before considering a price change. It then tests a channel-specific bundle and measures contribution per order, attachment rate, complaints and preparation time. The original automated recommendation is retained so the effect of corrected data is visible.
Measures that keep the analysis honest
Track:
- percentage of item sales with a current, complete recipe cost;
- net contribution by item, location, daypart and channel;
- product mix and modifier attachment;
- stockout-adjusted availability;
- waste associated with the item;
- recommendation acceptance and the reason for rejection;
- change in total contribution, not only item margin percentage.
Avoid presenting a classification such as “star” or “dog” without the period, threshold and economic definition. The label is a prompt for investigation, not the decision.
Connect item demand to the restaurant forecasting workflow and ingredient movement to inventory and purchasing. Use the restaurant AI ROI guide to include data-maintenance and implementation costs. Book a conversation to examine one menu decision.