They weren't "Lovin' it"
When AI doesn't meet McDonald's standards
For two years, McDonald’s put AI in the one place it could measure clearly – the drive-thru. Voice ordering, built with IBM, ran in over 100 restaurants, testing against real-time traffic to see what worked.
In June 2024, they pulled it; by the end of July, it was back to business with the normal human crew.
With all the confidence of a ~$190B1 company, they killed a “successful” voice-ordering solution.
I’ve watched leaders make this exact move. The pilot didn’t clear the bar, so instead of saying so, they promoted the goal to a “Future State”. The next partner will get it there. The win is always in the next phase, which is exactly far enough that no one has to call the current attempt what it was. Undercooked!
Do you want fries with that?
Two years, a hundred restaurants, and scrapping the program is not a “successful” proof of concept. What’s hard is a spoken order. What’s harder is a system trained to understand the complexity of languages and accents. A Stanford study tested ASR systems from five providers- Amazon, Apple, Google, Microsoft, and IBM, the same IBM behind McDonald’s ordering- and found error rates for Black speakers nearly double those for white speakers, even when age, gender, and the words spoken were identical.2 A 2024 Georgia Tech study found the same gap for Spanglish and Chicano English speakers, and traced it to one cause: the models were never trained on enough of those voices to begin with.3
Whatever the system couldn’t do, it still can’t.
So here’s the one tool to “carry out” into your next planning meeting. When a test ends and someone reaches for “this gives us confidence for the future,” say this:
Before we file this as a future state — what do we call today? And what do we keep, either way?
It takes very little to find secondary value in the test you worked hard to put together. Even if it’s for your personal profile.
McDonald’s can afford to run the test again when AI has improved its linguistic range. Most teams can’t. They get one shot at the budget, and “Phase 2 will fix it” is the lie.
Thank you. Seeing you next time!
McDonald's franchisee memo via Restaurant Business and Fox Business, June 2024; corroborated by Restaurant Dive and CIO Dive. Market capitalization (~$190B) via StockAnalysis, accessed July 2026.
Koenecke et al., "Racial disparities in automated speech recognition," PNAS, 2020, via Stanford Engineering.
Harris et al., Georgia Tech/Stanford dialect study, November 2024, via Georgia Tech News.