McDonald’s ditches AI Tech, offers key lessons

Surprisingly, it took McDonald’s over two years to realize its AI-based order system wasn’t ready for deployment.

Recently, McDonald’s decided to end its much-publicized AI systems for voice ordering at more than 100 drive-thru restaurants in the US. This innovative service, launched with the hope of enhancing customer experience and reducing wait times, failed to deliver and instead became a source of frustration among McDonald’s customers.

For those unfamiliar, McDonald’s offers drive-thru service at over 25,000 locations worldwide. To make its service as efficient and convenient as possible, the fast-food giant launched the AI-based voice ordering service in the US with IBM in 2021. However, what began as an innovative AI initiative to streamline the order-taking process quickly turned into a major pain point. Instead of receiving positive feedback and increasing customer satisfaction, the initiative resulted in widespread mockery from customers, who shared viral stories of comical mishaps involving the AI-driven tool.

A company shutting down its AI and ML plans is not new, especially after receiving negative feedback. What’s truly surprising for many is that it took the world’s largest fast food restaurant chain over two years to realize that its AI-based automated order-taking system was not ready for deployment. The problem lies in the immense pressure from top management on enterprises, their CIOs, and IT leaders to implement AI and ML-led automation. While AI has advanced beyond the phase of FOMO or “me too,” an overreliance on its potential sometimes delays the recognition of its flaws, as admitted by a few CIOs in my interactions.

According to several media reports citing sources from McDonald’s, the problem with the AI system was its inability to correctly interpret different accents and dialects, especially for those for whom English is not the first language. While McDonald’s aimed to address rising labor costs and achieve significant savings by automating its order-taking operations, it likely failed to effectively utilize historical customer data in training the AI system. As a result, the AI struggled to provide custom recommendations or address recurring issues that customers faced.

That said, it is also true that every successful project is the result of failed experiments, and sooner or later AI will be an integral part of many operations in the industry. However, this episode is also a reminder that AI capabilities still need more effort in research, data learning, and guidance from technology partners.

“I can’t believe I’m yelling at a robot right now,” complained one user on Reddit. “Then an actual human being broke in, apologized for the confusion (not their fault), and took my order instead.” This sentiment encapsulates the broader customer frustration and the gap between AI expectations and reality.

Share on