Restaurants · June 20, 2026
How to increase restaurant average order value with AI
Test AI upselling and cross-selling in your restaurant. A hypothetical calculation explains order value, contribution, metrics and integration requirements.
To increase a restaurant's average order value, test one relevant suggestion, confirm the revised total and measure its contribution. An AI assistant can support that workflow in a channel connected to your operation. Results depend on the menu, pricing, acceptance and costs; installing an agent does not guarantee a percentage increase.
Consider a WhatsApp order: the customer chooses a dish and the assistant asks whether they would like an available drink. Acceptance updates the order summary before payment; a refusal ends the suggestion without pressure. That specific intervention can be measured against the current process. This guide covers five opportunities, a hypothetical calculation and a supervised pilot.
1 · Average order value, spend per guest and contribution are different
Average order value = period sales ÷ orders or transactions in the same period. Use a consistent basis for discounts, taxes, tips, delivery charges and refunds. Spend per guest divides by diners; it is different from a table check or a family delivery order.
Separate WhatsApp, dine-in and delivery when their operations differ. A higher average may reflect new prices, larger groups or a different product mix. Do not automatically attribute it to the assistant. Also assess the variable cost of supplying the additional item.
Upselling and cross-selling are related, but distinct
A higher-value version of the same item is an upsell. Adding a drink, side or dessert is cross-selling. Both can raise order value; this guide groups them as commercial suggestions while preserving customer choice. No item should be added without acceptance.
2 · Five opportunities: choose where to make a suggestion
These are alternative moments for designing a pilot, not five offers to send in one order. Available integrations determine which are feasible.
1. When the dish is selected
Offer an upgrade or side listed on the menu: “Would you like to add fries at the listed price?” The assistant needs current prices and stock; it must not invent bundles or promotions.
2. Before the total is confirmed
Suggest a relevant drink once. If accepted, show items, quantities, charges and the updated total before payment. Do not call an item a bestseller without supporting restaurant data.
3. During table service
A staff tool can suggest approved combinations. It requires an interface and a specific integration; it does not appear automatically when a WhatsApp agent is purchased.
4. When the customer asks about finishing the meal
Coffee or dessert may be relevant when someone expresses interest. Knowing that a main course is finished requires a staff or system signal; the AI does not observe the table on its own.
5. Before a delivery order is closed
Check stock, packaging and cutoff time. Adding an item after payment or kitchen dispatch may require changing the order and charge. Prefer a suggestion before confirmation, with clear delivery terms.
3 · Rules the restaurant needs to approve
- One suggestion per order during the pilot. This is a design rule to test, not a guarantee of acceptance or satisfaction.
- No repeated offer after refusal. The original order must be able to proceed smoothly.
- Verified items and prices. Use an authorized source and confirm availability.
- Preferences based on available, authorized data. Customer history and weather require sources and integrations; they are not implicit model capabilities.
- Ingredients or allergy questions: use verified information and refer to staff when it is insufficient. Do not invent food-safety assurances.
AI can draft and interpret messages; commercial actions need tools and rules. Saving preferences does not automatically train the model. Suggestion changes and A/B tests must be designed, supervised and approved within the contracted service.
4 · Hypothetical example: 25 of 100 orders accept
This coffee shop is fictional. Values are editorial assumptions explaining a calculation, not an industry benchmark or a client's result. Volume stays at 100 orders a day, with a base average of COP 18,500. A COP 4,200 add-on is offered on each order and accepted by 25 people.
| Example variable | Calculation | Result |
|---|---|---|
| Base daily sales | 100 × COP 18,500 | COP 1,850,000 |
| Daily additional sales | 25 × COP 4,200 | COP 105,000 |
| Modeled average order value | (1,850,000 + 105,000) ÷ 100 | COP 19,550 |
| Average order value change | 1,050 ÷ 18,500 | Approximately 5.68% |
| Additional sales over 30 identical days | 105,000 × 30 | COP 3,150,000 |
If each add-on had a variable cost of COP 1,600, additional contribution would be 25 × (4,200 − 1,600) = COP 65,000 a day, or COP 1,950,000 across those 30 hypothetical days. Technology, payment, operating and other applicable costs still need deduction. This is neither net profit nor a sales forecast for your restaurant.
The example's 25% measures suggestion acceptance; 5.68% measures average order value growth. Those metrics are not interchangeable. If suggestions reach only some orders, change the denominator and recalculate. Real trading days are not assumed to be identical.
5 · Implementation phases and owners
- Prepare the menu: select combinations and check prices, variable cost, stock and exceptions.
- Configure: define the trigger, suggestion limit, acceptance, order summary and human handoff.
- Test: simulate refusals, unavailable products, quantity changes, payment failures and repeated messages.
- Pilot and review: use a controlled channel and volume, reconcile orders and decide whether to expand.
POS, staff tablets, payments and kitchen systems require checking APIs, permissions and supported operations. If an integration is unavailable, describe the manual step or alternative in the proposal. Read the restaurant AI agent guide and our WhatsApp-to-kitchen walkthrough, in Spanish, before defining scope.
A suggestion during an active order and a later campaign are different workflows. For subsequent contact, review permissions, preferences and template rules in the WhatsApp Business policy. Meta messaging and other usage also contribute to cost.
6 · What to measure before continuing
The visual above is a mockup using fictional data, not a connected dashboard or functional demo. Its conversation shows a separate illustrative order, not the coffee shop used in the calculation.
- Coverage: eligible orders shown a suggestion ÷ eligible orders.
- Acceptance: accepted suggestions ÷ suggestions shown.
- Order value and contribution: sales per order and contribution after variable cost, by channel and period.
- Friction: time to confirmation, abandonment, complaints, cancellations and refunds.
Compare equivalent periods and record promotions, price changes and menu changes. A well-defined comparison group helps interpret the effect; a simple before/after comparison does not establish causation. Dashboards and reporting frequency depend on instrumentation and agreed scope.
7 · Choosing the next step
Start with a combination the restaurant can supply, charge and measure. Avoid introducing five suggestions or several channels at once. A pilot helps decide whether to expand, adjust or stop the automation.
Explore the Cocina Conectada demo with sample data or book a free 30-minute discovery call. The proposal must separate implementation, operation and usage, and confirm whether upselling and each integration are included. Current MD fees require a signed-in account.