A simple order doesn't prove a simple problem
A guest asks for a familiar meal by number, the system understands it and the demonstration looks convincing. Jason Riggs points to what happens next: modifiers, a request for a beverage brand the restaurant does not sell, a breakfast item after breakfast or a menu that has just changed.
In his first Season 2 ChatGTM conversation with Michael Beck and Chad Horn, Riggs discusses why restaurant AI ordering requires more than recognizing a sentence. The system must interpret the request within the restaurant's actual menu and carry an accurate result into the operating process.
The distinction matters to buyers. A successful exchange is encouraging, but the restaurant has to serve a varied stream of real guests throughout the day. The edge cases in a demonstration can become ordinary work in a busy drive-through.
Start with the problem the restaurant needs solved
Riggs recalls the recurring question of how a company can use AI and argues for reversing the order: identify the problem, then decide whether AI is an appropriate solution. That product-management discipline is especially useful when a technology attracts enough attention to become an objective in itself.
An ordering project might aim to reduce repetitive crew workload, improve order capture or support a more consistent interaction. Those are related aims, but they require specific measures and an understanding of the current process.
Write down the problem in terms a restaurant manager would recognize. Describe the moment where work becomes difficult and the change that would count as improvement. That gives the team a basis for evaluating the system without confusing an interesting capability with a useful outcome.
Test the menu that exists, including its exceptions
Riggs emphasizes how menus vary between brands and over time. Dayparts, seasonal items and modifiers create conditions that an ordering system must handle intelligently. The technology also needs to pass the resulting order into the point of sale in a usable form.
A practical evaluation should therefore include realistic changes and ambiguities. Test an unavailable choice, a changed preference, an item near a daypart boundary and a combination with several options. Check both the conversation and the order the restaurant receives.
This is also an ownership question. Someone must keep the menu information current and understand how changes reach the system. A polished initial setup does not answer what happens after an item is removed or a local exception is introduced. Maintenance belongs in the buying conversation alongside the demonstration.
An accurate conversation still needs an accurate meal
Riggs separates the cognitive work of taking an order from the work of assembling and handing it to a guest. His argument is that reducing the first burden can give employees more attention for the second. That is a plausible operational benefit to investigate, not a guaranteed result of installing the technology.
An end-to-end review should follow the order beyond the speaker. Was it represented correctly? Could the team understand it? Did the correct items reach the bag and the customer? A failure at any stage affects the experience, even when an earlier stage reports success.
That view also changes how time savings should be assessed. Observe whether staff can use the attention the system is meant to release. If they must continually supervise, correct or re-enter orders, the operational benefit may be smaller than the automated interaction count suggests.
Keep judgment in the evaluation, even as tools get faster
Riggs discusses AI's usefulness in organizing information, including large collections of support tickets. Faster synthesis can help a team identify themes, but he repeatedly returns to the need for critical thinking and review. An output can be useful without being automatically correct.
Apply that discipline to the pilot itself. Review a representative set of interactions, compare system measures with restaurant observations and investigate exceptions. Distinguish an actual improvement from a change in how a metric is counted. The people using the system need a clear route to report what the summary misses.
The strongest case for AI ordering is an operating result that remains convincing after the demonstration ends. The restaurant can handle the changing menu, employees understand their responsibilities and guests receive the meal they intended to order. That is a more demanding test than taking a number one, and a more useful one.
