McDonald’s tests Google-backed AI drive-thru ordering system
McDonald’s tests AI drive-thru system “ArchIQ” (Archy) with Google at 5 US locations. System processed >1M transactions, ~90% completed without staff escalation. Features include multilingual orders, repeat customer recognition, and operational monitoring. Follows a failed IBM AI pilot criticized for high error rates. Part of broader “McDonald’s > NEXT” growth and automation strategy.
Analysis
TL;DR
- McDonald’s tests AI drive-thru system “ArchIQ” (Archy) with Google at 5 US locations.
- System processed >1M transactions, ~90% completed without staff escalation.
- Features include multilingual orders, repeat customer recognition, and operational monitoring.
- Follows a failed IBM AI pilot criticized for high error rates.
- Part of broader “McDonald’s > NEXT” growth and automation strategy.
Key Data
| Entity | Key Info | Data/Metrics |
|---|---|---|
| ArchIQ System | AI drive-thru & ops tool; developed with Google | Processed >1M transactions; ~90% no staff escalation |
| McDonald’s Loyalty Program | 2025 sales to loyalty members | Sales rose 20% to ~US$37B |
| McDonald’s Loyalty Program | 2025 active users | 90-day active users rose 19% to ~210 million |
| Drive-Thru Traffic (2025) | Negative year-long trend | Hovered between -5% to -8% monthly |
| Previous IBM Pilot | Ended in 2024 after errors | Tested across >100 restaurants |
| IBM Pilot Error Example | Viral customer video | Added >$250 worth of chicken nuggets |
Deep Analysis
McDonald’s is finally admitting that the first-generation AI ordering experiment was a bust. The IBM partnership, which imploded after videos of a $250 nugget order went viral, was a case study in deploying technology without solving the core problems of context, nuance, and error recovery. Now, with ArchIQ, they’re making a more pragmatic, layered bet—not just replacing a cashier, but attempting to build a nervous system for the restaurant itself. The shift from a pure point-of-sale AI to one that also monitors freezers and kitchen bottlenecks is the real story here. It’s a move from a fragile, single-use tool to a potentially resilient, multi-purpose operations platform.
The partnership with Google makes strategic sense. Google’s edge cloud infrastructure, referenced via the “blades” being sent to restaurants, is critical. Processing orders and sensor data locally reduces latency and dependency on a distant cloud, which is essential for a time-sensitive, high-volume environment like a drive-thru. This isn’t just about a chatty voice bot; it’s about deploying real-time machine learning at the periphery of McDonald’s vast network. The 90% completion rate is impressive, but it must be viewed skeptically until we know the complexity of the orders included. Were they simple Big Macs, or complex customizations with substitutions?
The most telling data point isn’t from McDonald’s, but from the industry: drive-thru traffic is in structural decline. This automates a shrinking channel, which seems counterintuitive. The move is defensive. If fewer people are coming to the drive-thru, you must maximize the efficiency and margin of every remaining transaction. Automating the order is step one; using AI to optimize the kitchen flow behind that order is step two. This is about unit economics under pressure, not hospitality innovation.
Kempczinski’s memo about automation raising the “standard for hospitality” for human interactions is a clever reframing. It’s an admission that the human-touchpoint is becoming a premium, optional feature. This bifurcates the customer experience: a fast, cold, algorithmic default for most, and a warmer, slower, human one for those who demand it. It’s a direct echo of supermarket self-checkout lanes, which didn’t eliminate cashiers but redefined their role.
The test’s biggest vulnerability remains the edge cases that derailed the IBM system: strong accents, background noise, ambiguous slang, and the chaotic reality of a family screaming orders from the back seat. “Repeat customer recognition” is a particularly risky feature—convenient when it works, a surreal nightmare when the AI guesses wrong. McDonald’s is walking a tightrope between impressive automation and the uncanny valley of a machine that thinks it knows you. Their success hinges not on the 90% of simple orders, but on how gracefully they fail in the other 10%.
Industry Insights
- Edge computing becomes the new battleground for QSR tech, with data processing moving on-premise to ensure speed and reliability for AI-driven operations.
- AI’s role expands from customer interface to holistic “restaurant brain,” integrating order-taking with predictive maintenance and workflow optimization.
- Automation drives a hospitality split: fast, impersonal service as the default, with human interaction becoming a premium, higher-cost option.
FAQ
Q: How does ArchIQ know a repeat customer’s “usual order”?
A: The system likely links order data to a loyalty account or recognizes vehicle/payment identifiers over time, though McDonald’s has not disclosed the specific technical implementation.
Q: Why did the previous IBM AI ordering system fail?
A: It was plagued by high-profile order errors, like adding hundreds of dollars of items, which went viral and highlighted significant issues with accuracy and handling conversational nuance.
Q: Will this AI system replace McDonald’s workers?
A: McDonald’s states it’s designed to support crew by handling orders and alerts, freeing them for other tasks. However, it inherently reduces the need for a dedicated order-taker position.
Disclaimer: The above content is generated by AI and is for reference only.
Frequently Asked Questions
How does ArchIQ know a repeat customer’s “usual order”? ▾
The system likely links order data to a loyalty account or recogni