McDonald’s is testing another AI drive-thru system, two years after ending a much larger automated-ordering pilot that became known for spectacular mistakes. The new platform, ArchIQ—nicknamed “Archy”—is running at five unnamed US restaurants and was developed with Google.
The limited scope is significant. Voice ordering in a noisy lane looks simple until accents, children, engines, menu changes and corrections collide in one conversation. McDonald’s now has a chance to test whether a more capable system can improve speed without turning an ordering error into the customer’s defining experience.
ArchIQ combines ordering with restaurant awareness
A demonstration showed ArchIQ greeting customers, accepting changes, confirming the total and directing the vehicle to collection. It can operate in English and Spanish, and a franchisee account said roughly 90% of more than one million transactions were completed without staff escalation.
The system may also recognise a returning customer who asks for a usual order, although McDonald’s has not explained the identity or consent mechanism. That convenience needs a clear opt-in: a familiar order is helpful only when customers understand how it was remembered.
ArchIQ extends beyond the speaker post. It can alert managers when a freezer fails or a kitchen bottleneck develops, turning the same local computing infrastructure into an operational assistant. Google Edge Cloud blades are reportedly being installed in US restaurants, allowing more processing close to the location.
This broader role could create more value than voice automation alone. A correct order still arrives late if equipment fails or the kitchen becomes overloaded. Connecting demand with operational conditions lets managers respond to the whole service flow.
The previous pilot makes recovery the central test
McDonald’s ended its IBM-based trial in 2024 after testing at more than 100 restaurants. Customer videos documented bizarre additions, including an order reportedly inflated by hundreds of dollars of chicken nuggets. Those incidents became memorable because the system appeared unable to recognise that it had left ordinary reality.
The new evaluation should therefore measure correction and escalation, not merely the percentage of orders completed automatically. Staff need to enter a conversation instantly, see what the system understood and fix the basket without forcing the customer to begin again.
Accuracy should be broken down by language, menu complexity, background noise and requested modification. A single average can conceal poor performance for customers whose speech or orders differ from the dominant test set.
Restaurants also need a simple fallback during connectivity or hardware problems. Automation cannot become a new point of failure at a lane whose value depends on predictable throughput.
Better automation should make hospitality more visible
McDonald’s frames ArchIQ as part of its NEXT growth plan, which focuses on restaurant economics and operations. The company’s digital reach is already enormous, with nearly 210 million active loyalty users at the end of 2025.
Automation may free employees from wearing a headset in a noisy environment, but the saved time must appear somewhere customers value: quicker preparation, cleaner hand-offs or more attentive problem solving. Reducing labour without improving those moments would make the technology visible for the wrong reason.
The company’s chief executive has acknowledged that greater automation leaves fewer human interactions, raising the standard for each one that remains. That is the right lens for the pilot. A drive-through is not only a transaction pipeline; it is a compressed service encounter in which trust can vanish after one inexplicable mistake.
McDonald’s should expand ArchIQ only when it can prove consistent orders, fast human recovery and tangible support for restaurant teams. The most successful voice system will not be the one that never asks for help. It will know precisely when the conversation has exceeded its confidence—and hand it to an employee before frustration reaches the pickup window.
