Ask for a venti iced coffee with light skim milk at a Starbucks counter and the transaction takes about 15 seconds. Ask the same thing through Starbucks' new ChatGPT-powered ordering interface, and it turns into a multi-step conversation.
Starbucks has been rolling out a conversational AI ordering feature in its app that lets customers describe what they want in plain English instead of navigating the standard menu. The idea is that natural language should be more intuitive than tapping through dozens of drink options and customization variables. For customers building a complex custom order from scratch, there's a reasonable case. For regulars with a fixed order they've placed hundreds of times, it's a slower path to the same drink.
The failure mode here is consistent with other AI ordering pilots across retail and fast food: conversational interfaces are optimized for discovery, but most daily customers already know exactly what they want. A regular who orders the same drink every morning doesn't need help navigating 50 milk options. They need the fewest possible taps between intent and payment. Menu-based ordering runs on muscle memory for repeat customers. Typing out your order, waiting for the AI to parse it, and reviewing whether the customizations registered correctly breaks that flow every time.
Fast-food chains and other retailers have been running similar experiments for over a year, with consistent results: the technology performs better as a discovery layer than as a default interface. A more practical implementation would be hybrid - menu navigation as the standard, with conversational AI available for customers who want help building something new. Replacing the menu entirely just adds friction for the 80% of customers who already know what they want.