Incident file · Production
McDonald's rolls back drive-thru AI ordering (2024)
enterprise-scale reliability failure, full rollback after ~3 years
The system
McDonald's piloted IBM-built AI voice ordering across more than 100 US drive-thrus for roughly three years, taking spoken customer orders in real time at checkout without a human order-taker.
What was observed
Viral videos circulated showing comically wrong orders: hundreds of chicken nuggets, bacon added to ice cream, nine iced teas added to a single bill. In June 2024, McDonald's ended the pilot.
Stop and think
- Analysts cited an informal bar of 95%+ accuracy needed to beat a human order-taker on cost and reliability, while the system reportedly stuck in the "low-to-mid 80% range." What does that gap mean in terms of how many orders per shift go wrong, and who absorbs that cost?
- A drive-thru is a high-ambiguity, real-time audio environment (accents, dialects, background noise). What design choice determines whether a low-confidence transcription becomes a wrong order on someone's bill, or gets caught before it does?
- This ran successfully enough to stay deployed for roughly three years before rollback. What does that tell you about the difference between "works often enough to ship" and "works reliably enough to scale without a human safety net"?
Answer the questions above before you read on: what failed, why, and what would have stopped it.