Increase revenue per order: The restaurant menu engineering playbook

Every restaurant has a menu engineering problem. Most owners know which items sell well and which don't. Very few know which items are actually profitable, and even fewer optimize their menu based on that data more than once a year.

Traditional menu engineering is a spreadsheet exercise. Pull sales data, calculate food costs, plot items on the Stars/Plowhorses/Puzzles/Dogs matrix, make some decisions, repeat in six months. It works, but it's slow and it captures a snapshot, not a trend.

AI menu engineering does the same analysis continuously, on every day's sales data, and generates specific recommendations your team can act on immediately.

According to the National Restaurant Association's 2025 Restaurant Operations Data Abstract, food and beverage costs came in at a median of 32.4% of sales in 2024 - a number that looks stable but masks enormous variation between high-performers and the rest. High-volume restaurants ($2M+ annual sales) ran food costs at 31.0%; lower-volume operators ran 33.7%. That 2.7-point gap is almost entirely menu engineering and purchasing discipline.

The menu engineering foundation (and why AI changes it)

Menu engineering has been around since 1982, when Cornell professors Michael Kasavana and Donald Smith published the framework that every restaurant school has taught since. The framework is simple:

Stars are high popularity, high margin. Feature them prominently and don't discount them. Plowhorses are high popularity but low margin. Guests love them but they hurt your food cost. Consider raising prices slightly or finding cheaper ingredients. Puzzles are low popularity, high margin. Profitable when ordered, but guests don't order them enough. Better descriptions, placement, or server recommendations can move these. Dogs are low popularity, low margin. Cut them. They take up space and prep time.

The problem with doing this manually: it takes hours to gather and analyze the data, so it happens quarterly at best. By the time you act on the analysis, the data is stale. Seasonal items have come and gone. Your food costs have changed. A supplier issue moved a key ingredient's cost by 20%.

AI menu engineering runs this analysis daily (or hourly during service) and surfaces specific recommendations your team can act on the same shift.

What AI menu engineering actually does

1. Continuous item classification

The AI classifies every menu item against your current data, not last quarter's data. An item that was a Star in summer might be a Plowhorse in winter if demand drops but food costs stay constant. You see this in real time.

More importantly: the AI identifies trends before they're obvious. If a dessert's order rate has dropped 15% over six weeks while margins held, that's a Puzzle getting worse, not a Dog yet, but heading there.

2. Menu placement optimization

On physical menus, the top-right corner of each page and the first and last items in each section get the most visual attention (the "Golden Triangle" in menu design). On digital menus, the first items in a category, the featured items section, and the items shown in upsell prompts drive a disproportionate share of sales.

AI optimization for placement: match your highest-margin profitable items (Stars and high-margin Puzzles) to the high-attention positions. This is not a one-time fix, it changes as item performance changes.

One fast-casual chain moved three high-margin items from mid-section positions to the "featured" placement in their kiosk ordering system. Average check size increased 8% within two weeks. No price changes, no new items.

3. Description copy optimization

"Chocolate cake" and "Warm Belgian chocolate lava cake with Tahitian vanilla bean ice cream and housemade caramel" are the same item priced at different levels, and the second one actually sells more.

Menu language is one of the most cost-effective levers in restaurant marketing. Sensory words (crispy, warm, rich), origin descriptors (Belgian, housemade, locally-sourced), and texture descriptors (creamy, flaky, crunchy) consistently increase order rate for the items they describe.

AI systems can A/B test description variations on digital menus and identify which language drives the best combination of order rate and margin contribution. The winning description gets promoted automatically.

4. Personalized digital menus

This is where AI menu engineering moves beyond traditional methods.

Physical menus are static. Everyone sees the same menu. Digital menus (kiosks, QR-code ordering, apps) can show different featured items based on:

• Time of day: feature breakfast items before 11am, lunch specials during midday rush, cocktails and appetizers from 4-6pm

• Weather: hot soup on cold days, salads and lighter fare on warm evenings

• Customer history: return customers see items they've ordered before and items similar to their preferences

• Party composition: tables with children get family-friendly items featured more prominently

Domino's uses ordering history personalization and reports that personalized customers order 10-15% more frequently than non-personalized users. At a restaurant doing $2M in annual revenue, that's $200K-300K in additional revenue from existing customers.

5. Demand forecasting for waste reduction

Menu engineering isn't just about revenue, it's also about food cost. AI demand forecasting predicts order volumes for each item by day, time, and weather conditions.

This enables you to prep the right quantity of mise en place per shift, order ingredients matched to forecasted demand instead of guessed averages, reduce food waste (typically 4-10% of restaurant revenue), and avoid 86'ing popular items mid-service.

A mid-size restaurant group running 8 locations implemented AI demand forecasting and reduced food waste by 23% in the first year, saving $180K annually. The system cost $40K to build and connects to their existing POS.

Food waste isn't a small problem. ReFED's 2025 Foodservice Sector analysis found U.S. foodservice operations generated 12.4 million tons of surplus food in 2024, valued at $157 billion - equivalent to 14% of total foodservice sales. Of that waste, 78% went straight to landfill. The National Restaurant Association estimates that for every $1 invested in food waste reduction, restaurants realize approximately $8 in cost savings - one of the highest ROI interventions in restaurant operations. Demand forecasting is one of the few interventions that addresses the problem at the source, not after the food is already wasted.

Dynamic pricing: Viable, but handle with care

Dynamic pricing, raising prices during peak demand, made headlines in 2024 when Wendy's announced and then quickly reversed plans to implement it. The backlash was predictable: customers don't want to feel penalized for wanting dinner on a Friday night.

The version that works: frame it as discounts, not premiums. Instead of "Friday dinner rates are 15% higher," run "Happy Hour: 15% off all entrees from 3-5pm Monday-Thursday." Same math, completely different customer perception.

Technically, dynamic pricing for restaurants costs $15K-30K to build (rate rules engine, POS integration, digital menu update API). The bigger challenge is the communication strategy, not the technology.

The data you already have (and what you're not using)

Most restaurant operators underestimate how much usable data they already have:

Your POS already exports item-level sales by day, time, and server. That data includes order size and combination patterns, discount and comp rates, and table-level sales for fine dining. Your accounting system holds ingredient costs and invoices, actual food cost vs. theoretical, and variance tracking.

What you don't have yet but can get: customer identity data from your loyalty program or QR-code ordering, demographic patterns by order type, and real-time competitor pricing from third-party data services.

Three months of POS data is enough to start running meaningful menu engineering analysis. Six months shows seasonality patterns. One year shows year-over-year trends.

What this costs to build vs. buy

SaaS menu engineering tools (Avero, Craftable, MarketMan) run $500-1,500/month. They cover standard analysis and food cost management but have limited personalization and no custom integrations.

A custom AI menu engineering system runs $25K-60K. It connects directly to your POS and inventory system, builds the recommendation layer to your specific menu structure and margin targets, enables personalization on your kiosk or app, and leaves you owning the model and the data.

The SaaS path makes sense if you want to start quickly and your operation is standard. Custom makes sense when you have a digital ordering channel (app, kiosk, website), significant order volume (100+ covers/day), and specific margin optimization requirements.

Where to start

If you're running a restaurant group and haven't done menu engineering analysis in the last 90 days:

1. Export 90 days of POS data with item-level sales and prices
2. Build or buy your recipe cost data (actual ingredient cost per item)
3. Run the classic four-quadrant analysis, identify your Dogs and cut them
4. Move your best Puzzles into higher-attention menu positions
5. Rewrite descriptions for your top Puzzles using sensory language

You don't need AI to start. You need data and the discipline to act on it.

Where AI adds value: continuous analysis as conditions change, personalization at scale (impossible manually), and demand forecasting for waste reduction.

"The biggest missed opportunity in restaurant technology is that operators have years of transaction data and do almost nothing with it. Menu engineering used to require a consultant and a spreadsheet. Now it requires a data export and 48 hours of processing." - Andrew Freeman, Restaurant Consultant and Founder, Andrew Freeman & Co.

RaftLabs has found that POS integration is where most restaurant AI projects take longer than expected. The data has been collecting for years in Toast or Square or Lightspeed, but it's rarely in a clean format ready for analysis. Getting the data pipeline right takes about a week. Once that's done, building the recommendation engine is the straightforward part.

Talk to a founder if you're building a restaurant tech product or want to add AI menu optimization to an existing platform.

Further reading: AI in Hospitality, the full picture of how AI is changing hotel and restaurant operations. Voice AI for Restaurant Phone Orders, how AI is handling the phone ordering channel.