How to test your AI menu assistant before launch
Test your AI menu assistant with category, budget, comparison and missing-information scenarios. Plan result records and checks after menu updates.
A well-written answer to one question does not prove that every important situation in your menu works correctly. A short but varied scenario list is more informative during launch preparation. Write down expected results from your own products in advance, then check whether responses meet those expectations for names, prices, availability and information limits.
Contents
Derive expected results from the menu record
Choose a few available products from three different categories. Write their names, published prices, currency and any verified nutrition values in a small check table. Calculate which products fit the budget yourself before asking the assistant. This reduces the risk of accepting the model's response as correct and interpreting the menu record to fit it. The current menu approved by the business should be the test's reference.
Distinguish the draft from the published menu. A new price may have been saved in the management screen while the guest menu still shows an older publication. Identify the version you intend to test and open its guest link. Comparing Orvunai's product cards and assistant responses against that published information gives the check the correct foundation.
Prepare short questions covering different behaviors
A category question tests product discovery, a budget question tests a numerical limit, and a comparison question tests bringing records together. You should also request an item absent from the menu and ask about a field without recorded information. Not every test should find a product; accurately explaining that no suitable product exists is also a successful result. Keep the question text and expected behavior on the same row.
Adapt the examples below to your actual category and product names. Choose numerical limits with at least one suitable and one unsuitable product in your menu to make the check more meaningful. Do not expect a product guarantee for special ingredient requests that have not been verified.
| Scenario | Check point |
|---|---|
| Show a category | Relevant, available products |
| Find options within a budget | Recorded prices and correct currency |
| Compare two products | Correct products and preserved missing fields |
| Request an item absent from the menu | No invented product |
| Ask about an unrecorded detail | Information limit and direction to the team |
Test changes after the first message
Guests do not start from scratch with every question. They may ask for a dessert and then reduce their budget, or compare two products and then request another alternative. Try at least a few scenarios as two- or three-message conversations. Observe whether results use the new budget while preserving the relevant category context. Note whether old products continue appearing in the wrong context.
When product availability changes, also check that a new question uses the current status. A price or currency change can affect numbers from the earlier conversation. Manage test changes separately from real service data or choose a controlled publication time. Do not leave the business's menu in an incorrect state merely for testing.
Record issues so they can be reproduced
When you see an error, do not write only 'The assistant recommended incorrectly'. Record details such as the menu or branch, question sequence, expected product, displayed response and check time. You do not need to collect a guest's personal information; an example illustrating the problem is enough. Distinguishing a missing product field, an incorrect publication and unexpected chat behavior helps resolve the issue faster.
After a correction, retest only the problem scenario and nearby flows likely to be affected. Choose checks according to the menu risk instead of repeating the same long list after every minor wording change. As the final prelaunch step, verify that cards and comparison tables are readable on a phone. The assistant's function is fulfilled when guests can understand the accurate information it provides.
Explore the relevant Orvunai product guide to put these steps into practice with your own menu.
Explore AI restaurant menu and Luca AI waiter