Maybe voice interfaces are simply a fundamentally flawed idea. If one can extend a hand to take the package with the food, they can also push a few buttons. If those buttons are with hercons or such, they'll even last longer than consumer-grade touchscreens.
Of course it's easier when a human takes the order. But then if the cost of N screens with physical buttons is equal to that, one can make their order, say, N/2 times slower without any hurry and, well, the throughput should be higher still.
For drive thrus - that'd be M lanes with such terminals and a bit slower than M lanes with people. So - depends on how the cost of asphalt and space and people and terminals work economically.
What's definitely idiotic is to think one can replace a human with an "AI" without losing in efficiency. But then again, maybe it's worth it.