What would you have done? McDonald's switched off its voice-ordering AI after 100 restaurants and three years of testing

In 2021, McDonald's began testing automated order taking at the drive-thru with IBM, rolled out to more than 100 restaurants. Three years on, accuracy was still short of expectations and videos of botched orders were circulating: push ahead with the rollout, drop voice ordering altogether, or drop the vendor without dropping the project? On 17 June 2024 the chain confirmed it was ending the partnership, with the technology switched off in every participating restaurant no later than 26 July — while stating that the test “has given us the confidence that a voice ordering solution for drive-thru will be part of our restaurants' future.” In June 2026 it began piloting a new voice assistant built with Google in five restaurants.

The lesson holds at any size: a disappointing pilot has not failed, it has answered — the expensive mistake is confusing ending a vendor with ending the project, or the reverse, extending a vendor because the years already invested weigh more than the results. Before launching your next project, write into the specification the numeric performance threshold that triggers either rollout or exit, the review date that will settle it, and the clause that keeps the vendor replaceable: data and models exportable in an open format.

Source: Restaurant Dive, “McDonald's ends IBM drive-thru voice order test”, 17 June 2024; AI News, “McDonald's tests Google-backed AI drive-thru ordering system”, 10 June 2026.