McDonald's and IBM ended their AI drive-thru test

The conversation

McDonald's automated order-taking (AOT) technology traces to Apprente, a voice-AI startup acquired in 2019 and folded into an internal unit called McD Tech Labs. In October 2021, McDonald's and IBM announced, in a joint statement on IBM's newsroom, that IBM would acquire McD Tech Labs to “further accelerate the development and deployment” of AOT. On 17 June 2024 the two companies said they were ending that arrangement. Nation's Restaurant News reported McDonald's stated, “After thoughtful review, McDonald's has decided to end our current global partnership with IBM on AOT beyond this year,” adding “IBM remains a trusted partner and we will still utilize many of their products across our global system.”

What the documents show

The test ran at roughly 100 US locations from October 2021 until the 2024 announcement, per the outlet's reporting, which cites a 2022 BTIG analyst report placing voice-ordering accuracy in the “low 80% range,” short of a 95%-plus figure it describes as McDonald's target; that figure is BTIG's own estimate as reported, not one McDonald's or IBM published. Separately, ordering errors became widely shared on TikTok during the same period. IBM's quoted response describes the technology, not the decision to end the test, as sound: “This technology is proven to have some of the most comprehensive capabilities in the industry, fast and accurate in some of the most demanding conditions,” and said it remained “in discussions and pilots with several quick-service clients interested in the AOT technology.” Neither statement attributes the end to the videos specifically.

The system boundary

AOT was built to convert a spoken drive-thru order into a digital order without a person relaying it over an intercom; McDonald's frames its limits as a matter of accuracy and scale, not the underlying premise, saying it “see[s] tremendous opportunity in advancing our restaurant technology.” Where the system fell short, the fallback was a crew member correcting or re-taking the order at the window — the same fallback any drive-thru already has, so failures were visible as mistakes rather than unrecoverable errors.

Where it fails

The documented limitation is accuracy under real drive-thru conditions — background noise, regional accents, free-form phrasing — falling short of a target both the analyst report and the companies' framing treat as necessary before wider rollout. A builder evaluating a similar voice-agent deployment should treat a multi-year, hundred-site pilot ending in further evaluation, not expansion, as evidence the accuracy bar, not the concept, was the blocker.

• What accuracy rate, measured against real (not synthetic) orders, does the vendor report, and who verifies that figure?
• What happens to an order the system gets wrong, and is the fallback fast enough not to slow the line?
• Is a highly visible public pilot the right test bed, given that errors can shape perception independent of the actual error rate?

McDonald's and IBM's statements describe an ended partnership on one technology path, not an abandonment of voice ordering: both said they expect voice ordering to remain part of future restaurant technology, a forward claim from the companies themselves, not a documented plan.