McDonald's Ends AI Drive-Thru Ordering Trial After Viral Order Errors (2024)

Incident Summary

In June 2024, McDonald’s announced it was ending its Automated Order Taker (AOT) partnership with IBM after an AI voice-ordering system deployed at over 100 U.S. drive-thru locations produced persistent ordering errors that went viral on social media.^[1]

TikTok videos documented the system adding $222 worth of chicken McNuggets to a single order, putting bacon on ice cream, substituting stacks of butter for ice cream orders, and picking up orders from adjacent drive-thru lanes and combining them into one bill. McDonald’s CEO Chris Kempczinski had previously cited an 85% accuracy rate, with approximately 20% of orders requiring staff intervention. The system also struggled with accents and dialects.^[2]

McDonald’s Senior Vice President Mason Smoot sent a memo to franchisees stating the company had decided to end the IBM partnership, while noting it would continue evaluating alternative voice-ordering solutions. IBM issued a joint statement confirming the “collaborative” nature of the parting. The technology was removed from all test locations by July 26, 2024.^[3]

Key Facts

• System: Automated Order Taker (AOT), developed by IBM from McDonald’s 2019 acquisition of startup Apprente
• Deployment: Over 100 U.S. drive-thru locations
• Accuracy: 85% (per McDonald’s CEO); approximately 20% of orders required staff intervention
• Documented errors: $222 of McNuggets, bacon on ice cream, butter substituted for ice cream, cross-lane order mixing
• Decision: McDonald’s ended IBM partnership via franchisee memo from SVP Mason Smoot (June 2024)
• Removal: Technology removed from all test locations by July 26, 2024

Threat Patterns Involved

Primary: Overreliance and Automation Bias — An AI voice-ordering system was deployed in customer-facing roles despite a 15% error rate and inability to handle accents and dialects, with errors compounding into absurd order totals before staff could intervene.

Significance

1. Consumer-facing AI accuracy threshold — The incident demonstrated that an 85% accuracy rate is insufficient for customer-facing ordering systems, where errors are immediately visible and shareable on social media
2. Viral reputational damage — TikTok videos of absurd AI ordering failures generated significant negative attention for both McDonald’s and IBM, illustrating the reputational risk of deploying imperfect AI in high-visibility consumer interactions
3. Dialect and accent limitations — The system’s struggles with linguistic diversity highlighted a known but underaddressed limitation of voice-recognition AI in diverse customer populations
4. Orderly retreat — McDonald’s decision to remove the technology rather than continue iterating in production represents a responsible deployment correction, though the multi-year pilot raises questions about pre-deployment testing adequacy