Taco Bell Scales Back AI Tests After Customer Complaints

A string of AI-driven mistakes echoes earlier problems seen at other fast-food outlets.

Taco Bell is rethinking its use of an AI ordering system at drive-through outlets after a wave of viral clips showed the ordering system repeatedly making mistakes.

This comes just a few months after Taco Bell's parent company, Yum Brands, partnered with Nvidia to develop and implement AI voice-ordering systems for its drive-through restaurants. The goal, the company said, was to move ordering through digital channels to make things more efficient.

However, customers have highlighted a string of glitches that have left them frustrated and sometimes out of pocket.

In one video shared a customer reportedly ordered 18,000 cups of water through the AI voice assistant. Another shows the AI stuck in a loop, repeatedly asking a man who ordered a Mountain Dew what drink he would like with his meal. Others have reported unwanted extras being added to their orders or being charged twice.

Taco Bell’s chief digital and technology officer, Dane Mathews, admitted to the imperfections of the AI solution and said the company is now considering where it makes sense to deploy it.

“I think like everybody, sometimes it lets me down, but sometimes it really surprises me,” Mathews told The Wall Street Journal.

A Familiar Pattern Of Errors

This news follows McDonald's decision to end its own AI ordering trial in the U.S. last year. The system, built with IBM, had generated headlines after customers reported being served hundreds of unwanted chicken nuggets or bizarre combinations such as bacon-topped ice cream.

Like McDonald’s, Taco Bell hoped AI could reduce wait times and staffing costs, but repeated errors have shown that it is too early to trust AI to fully handle some customer-facing services.

Mathews confirmed to The Wall Street Journal that the company has no plans to scrap the AI voice system entirely. Instead, they plan to learn from the millions of orders it processed. He added that the company is now deciding where to best deploy the technology and how to train staff to act as a human-in-the-loop to assist the AI.

This article originally appeared on our sister site Computing.