Increase Average Order Value on Autopilot
Three modes from co-purchase to AI-ranked to per-customer personalization. Pick the level that fits your venue. No staff effort. No guessing.
What Are Smart Suggestions?
Every time a customer adds an item to their cart, the system checks what other customers ordered alongside it. If people who order a burger usually add fries and a drink, those items appear as suggestions. The data comes from your own order history — not assumptions.
Three layers, opt-in per venue: co-purchase rules (free, always on), AI-ranked re-ordering with localized reason copy that adapts to the cart context, and per-customer personalization that uses each regular's order history to tailor suggestions. Connect your own Claude key for unlimited AI usage. Suggestions appear as a post-add popup, on the cart page, and at the checkout review step.
What This Does for Your Business
Higher Average Order Value
Customers see extras that match what they already chose. A well-timed suggestion turns a single main into a full meal. The increase adds up across every order.
Zero Staff Effort
No training staff to upsell. No scripts. The system handles it automatically on every order, every shift, every day.
Based on Real Orders
Suggestions come from what your actual customers buy together — not from generic rules or manual configuration. The patterns are specific to your venue.
Keeps Up With Your Menu
The engine rebuilds daily. When you add new items or seasonal dishes, the system picks up new buying patterns automatically.
How It Works
Layer 1 — Co-Purchase Rules
Always-on baseline. The engine tracks which products appear together in completed orders and surfaces the most-frequent combos as suggestions. Free, fast, runs on every venue.
Layer 2 — AI-Ranked Suggestions
Claude Haiku 4.5 re-orders the top rule candidates based on the full cart, time of day, and order type. Returns localized one-line reasons for each pick — in your customer's language. Warm-cache latency is sub-200 ms; automatic fallback to Layer 1 on any timeout or error.
Layer 3 — Per-Customer Personalization
For repeat customers (≥ 3 completed orders), the system computes a taste signature from their order history and finds products semantically close to their preferences. The Layer 2 LLM then re-ranks this personalized pool. Anonymous carts and first-time customers stay on Layer 2.
Three Surfaces, One Engine
Suggestions appear as a post-add popup on the menu, a section on the cart page, and one last 'forgot a drink?' row at checkout. All three call the same engine, so analytics and A/B results are consistent across surfaces.
How Suggestions Work
Three-Layer Engine
Pick the level that fits your venue. Co-purchase rules (free) are always on. AI-ranked layer adds cart-context awareness and localized reasoning. Personalized layer tailors picks to each repeat customer's history.
- Built from your own order data, not generic rules
- Layer 2 generates EN/BG/EL reason copy in one round-trip
- Layer 3 uses vector embeddings of the customer's last 20 orders
- Each layer falls back deterministically to the layer below it
Bring Your Own Claude Key
Connect your own Anthropic API key in admin settings to bypass the platform's monthly AI quota. Unlimited suggestions, you pay Anthropic directly at their wholesale rate. Drop the key in once — every cart from that point on uses it.
- Unlimited AI requests when BYOK is connected
- Encrypted at rest in VenueIntegration storage
- Separate from Layer 1 — rules path is always free
- Per-venue spend cap as a safety net
Cost & Quota Controls
Built-in safeguards so AI costs never surprise you. Shared monthly quota with the AI Assistant + Restaurant Grader; per-venue cap; hourly burst guard against runaway carts; one-click kill-switch to flip back to free Layer 1 instantly during incidents.
- Live quota + spend dashboard at /admin/ai-usage
- Email alerts at 90 % and 100 % of monthly cap
- Per-feature cost breakdown (chat / upsells / grader / menu-import)
- Operator kill-switch on the Features page
Honest A/B Measurement
50 / 50 cookie-based split between Layer 1 (control) and the higher layer (test), sticky per customer for 30 days. The dashboard reports session-level conversion, acceptance rate, and AOV per arm with a 95 % significance gate — no winner declared until ≥ 1 000 sessions per arm.
- Sticky cookie split — same customer sees the same arm
- Shown / accepted / converted funnel tracked per row
- Significance test prevents premature claims
- Source column distinguishes rule / llm / personalized contributions
Where Suggestions Make a Difference
Burger + Sides + Drink
A customer orders a burger. The popup shows fries and a soda — the two items most commonly ordered with burgers at your venue. A single item becomes a combo.
Pizza + Dessert
People who order pizza at your place often add a tiramisu. The cart page shows it before checkout. A suggestion the customer did not think of but is happy to add.
Coffee Shop Morning Rush
Espresso orders almost always come with a pastry. The popup shows your top-selling croissant right after the coffee is added to cart.
Lunch Set Completion
A customer adds a soup. Your data shows most soup orders also include bread and a salad. The suggestion fills out the meal without staff involvement.
Seasonal Dish Discovery
You add a new summer cocktail. Within a week, the engine picks up that grilled dishes pair well with it. The cocktail starts appearing as a suggestion on grill orders.
Delivery Orders
Delivery customers cannot be upsold by a waiter. Suggestions fill that gap. The popup and cart section do the work that staff would do in a dine-in setting.
Why Automated Suggestions Outperform Manual Upselling
Manual upselling depends on staff memory, training, and motivation. On a busy Friday night, suggesting extras drops to the bottom of the priority list. An automated system runs on every order regardless of how busy the kitchen is. It never forgets, never feels awkward, and never skips a table.
Data From Your Own Customers
Generic recommendation engines use broad industry data. Your customers are not generic. A steakhouse and a sushi bar have completely different pairing patterns. The Ordering.Tools suggestion engine uses only your venue's order history. The recommendations reflect what your specific customers actually buy together.
The Right Moment Matters
A suggestion shown at the wrong time is noise. Shown at the right time, it is helpful. The post-add popup catches the customer while they are still browsing and in buying mode. The cart page section catches them during the final review. Both moments are natural decision points where an extra item feels like a good idea, not a pushy pitch.
Measurable Impact
Every suggestion click is tracked. You can see which products generate the most add-ons, which suggestions convert, and how average order value changes over time. The analytics dashboard ties directly into the suggestion engine so you can measure the return without guesswork.
Let Your Menu Do the Upselling
Smart suggestions run on every order. No setup, no training, no scripts.