The short answer
Yes, for phone orders. The famous AI failures happened in drive thrus, where open microphones, engine noise, and crosstalk break speech recognition. A phone call is a clean, single speaker audio channel, and current phone ordering agents post order accuracy in the mid to high 90s on par with trained staff.
Why drive thru AI famously struggled
The skepticism is earned. The most publicized test of AI order taking was McDonald's drive thru pilot with IBM, which ended in 2024 after viral videos of the system adding hundreds of chicken nuggets to bewildered customers' orders. Other chains quietly scaled back similar pilots.
The failure mode was the environment, not the concept. A drive thru microphone is an open mic aimed at a running engine, with wind, rain, car stereos, and back seat passengers all bleeding into the signal. Speech recognition degrades sharply in that setting, and every recognition error compounds into an order error. The lesson the chains took away was narrow but real: the microphone environment decides the outcome.
Why phone ordering is an easier problem
A phone call removes nearly every drive thru failure condition at once. The audio channel is engineered for a single voice. The caller holds the microphone to their mouth. There is no lane pressure timer, so the agent can confirm ambiguous items instead of guessing. And the caller chose to make the call, so intent is high and cooperation is natural.
The menu constraint does the rest. AI order taking works when the agent's understanding is bounded by a structured menu, because every utterance must resolve to a real item, size, or modifier rather than free text. A caller cannot accidentally order 260 nuggets when the menu logic rejects it and the agent reads the order back before payment.
The accuracy numbers today
Current phone ordering agents report order accuracy in the mid to high 90s. That compares favorably with humans answering a busy restaurant phone, where noise on the restaurant's end, multitasking, and handwritten tickets produce their own steady error rate. The readback and payment steps add two verification passes no rushed employee performs consistently. Deployment scale supports the numbers: phone ordering agents now run in production at thousands of restaurants, a quiet contrast with the publicity around the drive thru retreats.
The honest caveat is that accuracy depends on setup. An agent running against a stale or poorly structured menu will fail no matter how good the underlying model is. Vendors who insist on proper menu onboarding are signaling seriousness, not friction.
What working actually means
Working means more than transcribing correctly. It means the call ends with a paid, accurate ticket in the point of sale, and it means the agent answers calls the restaurant was previously missing entirely, up to 43 percent of them at peak. Against a baseline of voicemail, the bar is not perfection. It is answering.
X1 Voice, Batch Group's AI phone agent for restaurants, is built for the full definition: complete orders with modifiers, payment collected on the call, tickets injected natively into Square, Clover, and OrderCounter, in more than 8 languages, 24/7.
Commonly Asked Questions
- Why did McDonald's drop its drive thru AI?
- The IBM pilot ended in 2024 after persistent recognition errors in the noisy drive thru environment. Open microphones, engine noise, and crosstalk broke the speech recognition, a problem phone calls do not share.
- Is AI order taking more accurate than a human?
- On a busy phone line, generally yes. Voice agents post accuracy in the mid to high 90s and the mandatory readback and payment confirmation catch errors a rushed employee misses.
- What makes an AI order taking deployment fail?
- Almost always the menu data. Stale items, unpriced modifiers, and unstructured menus produce bad orders regardless of the AI. Proper menu onboarding and live sync prevent most failures.
Keep reading