How McDonald's AI Drive-Thru Experiment Failed (And What Comes Next)

How McDonald's AI Drive-Thru Experiment Failed (And What Comes Next)

McDonald's spent three years testing AI-powered voice ordering at more than 100 drive-thru locations. The partnership with IBM, which began in 2021, ended abruptly in mid-2024 after customers documented a steady stream of catastrophic ordering failures on TikTok.

One viral video showed an AI order-taker adding McNuggets to a customer's order in an endless loop. Another featured ice cream topped with ketchup and butter. A third captured the system picking up orders from neighboring cars in the drive-thru lane.

The technology couldn't handle accents or dialects. It misheard menu items. It made mistakes so consistently that franchisees and executives concluded the experiment wasn't worth continuing.

McDonald's quietly pulled the plug in June 2024, giving franchisees until July 26 to shut off the system. But the company isn't giving up on AI voice ordering. In a memo to operators, chief restaurant officer Mason Smoot wrote that "a voice ordering solution for drive-thru will be part of our restaurants' future." The company promised to select a new vendor by the end of 2024.

That deadline has passed. As of early 2026, McDonald's has not publicly announced a replacement. The IBM experiment is over. What comes next is anyone's guess.

The IBM Partnership: Apprente to AOT

McDonald's bought Apprente, a Silicon Valley voice recognition startup, in 2019. The acquisition was part of a broader bet on restaurant tech - the same year, McDonald's also bought Dynamic Yield, a personalization platform for digital menu boards.

IBM acquired McDonald's tech lab, McD Tech Labs, in 2021 and rebranded Apprente's technology as Automated Order Taking (AOT). The two companies announced a strategic partnership to roll out voice AI across McDonald's global system.

The vision was straightforward: automate drive-thru ordering to reduce labor costs, speed up service, and free employees for other tasks. The technology would handle the most repetitive part of the job - taking orders - while humans focused on food prep and customer service.

IBM marketed AOT as having "some of the most comprehensive capabilities in the industry, fast and accurate in some of the most demanding conditions." McDonald's tested it in more than 100 U.S. locations over a two-year period.

The results were mixed at best.

What Went Wrong

The accuracy problems were well-documented. Customers posted dozens of videos showing the AI misunderstanding orders, adding unwanted items, or failing to process simple requests.

Sources familiar with the technology told CNBC that the system struggled with accents, dialects, and background noise. Drive-thrus are loud. Customers speak quickly. Menus are complex. The AI couldn't keep up.

McDonald's declined to release accuracy data. IBM insisted the technology worked well "in some of the most demanding conditions," but those conditions apparently didn't include real McDonald's drive-thrus during peak hours.

The business case also fell apart. Drive-thru AI is supposed to reduce labor costs or improve throughput. If the system makes mistakes 10% or 15% of the time, employees have to monitor orders, correct errors, and apologize to frustrated customers. That's not automation - it's just another task.

McDonald's concluded the technology wasn't ready. The company ended the IBM partnership and removed AOT from all test locations by late July 2024.

The Competitive Landscape

While McDonald's was pulling back, competitors were doubling down.

Wendy's partnered with Google Cloud to build "Wendy's FreshAI," a custom voice ordering chatbot. White Castle teamed up with SoundHound AI to deploy voice tech at more than 100 locations by the end of 2024. Checkers, Rally's, Hardee's, Carl's Jr., Krystal, Dunkin, and Taco John's all launched AI drive-thru pilots.

Popeyes UK claimed 97% accuracy in its AI pilot and launched the first "AI-powered drive-thru" in the UK market in 2024. The company named the system "Al" - a play on both "artificial intelligence" and the casual greeting.

Taco Bell tested its own AI ordering system, with similarly mixed results. Like McDonald's, Taco Bell encountered communication failures and technical glitches that frustrated customers.

The industry consensus is clear: AI voice ordering is inevitable. The only question is which vendor gets it right first.

What McDonald's Learned

Despite the failure, McDonald's executives maintain that the IBM experiment was valuable. In its statement ending the partnership, the company said the work "has given us the confidence that a voice ordering solution for drive-thru will be part of our restaurants' future."

That confidence is based on lessons learned, not performance metrics.

McDonald's now knows that accent recognition is critical. The system has to work in Miami, Minneapolis, and Los Angeles - across every demographic and every regional dialect. It has to handle background noise, mumbling, and customers who change their minds mid-order.

It also knows that accuracy matters more than speed. A system that's 85% accurate isn't useful. It creates more work than it eliminates. The bar is probably 95% or higher - on par with human order-takers.

The company also learned that integration matters. AOT has to work with existing POS systems, kitchen display screens, and payment processing. It can't be bolted on. It has to be part of the stack.

The Google Cloud Partnership

In December 2023, McDonald's announced a multi-year partnership with Google Cloud. The deal moves restaurant computations from on-premises servers to the cloud and applies generative AI "across a number of key business priorities."

The partnership is broader than drive-thru voice ordering. It includes operations, supply chain, and customer experience. But the timing is notable. McDonald's ended the IBM partnership in mid-2024, six months after signing with Google.

Google Cloud has deep AI expertise. It powers Wendy's FreshAI. It has experience with voice recognition, natural language processing, and machine learning at scale. It's a logical partner for McDonald's next attempt at automated order-taking.

McDonald's has not confirmed Google Cloud as its new AOT vendor. But the company said it would make a decision by the end of 2024. As of early 2026, no announcement has been made publicly.

The Economics of Drive-Thru AI

The business case for AI voice ordering is simple in theory: reduce labor costs and improve throughput.

Drive-thru accounts for 70% of McDonald's U.S. sales. Shaving 10 seconds off average service time could add millions in annual revenue. Automating order-taking could free one employee per shift to work on food prep, cleaning, or delivery.

But the math only works if the technology performs. If employees have to babysit the AI, correct mistakes, or apologize for errors, the ROI disappears.

Accuracy is the gating factor. At 95% accuracy, the system might work. At 85%, it's a liability.

Labor savings are also more complicated than vendors suggest. Fast-food employees don't just take orders. They assemble food, clean stations, manage delivery, and handle in-store customers. Automating one task doesn't eliminate a headcount. It reallocates time.

The real value might be consistency. AI doesn't forget upsells. It doesn't get tired. It doesn't skip the "would you like fries with that?" prompt. Done right, automated order-taking could increase check sizes by 5% to 10% through consistent suggestive selling.

McDonald's is betting that the economics will eventually work. But the technology has to get better first.

What's Next

McDonald's promised a decision on a new voice ordering vendor by the end of 2024. That timeline has passed with no public announcement.

The likely candidates are SoundHound AI, Google Cloud, and custom-built solutions from POS providers like Toast or Olo. SoundHound already works with White Castle and has integrations with major restaurant tech platforms. Google Cloud has the AI infrastructure and the McDonald's relationship.

Whoever wins the contract will face intense scrutiny. The IBM experiment was public, messy, and ultimately unsuccessful. McDonald's can't afford another high-profile failure.

The new system will need to hit 95%+ accuracy from day one. It will need to handle accents, dialects, and noisy environments. It will need to integrate with McDonald's POS, kitchen systems, and payment stack. And it will need to work at scale - across thousands of locations, in multiple languages, in markets around the world.

That's a tall order. IBM couldn't deliver. The next vendor will have to do better.

The Bigger Picture

McDonald's AI drive-thru experiment failed, but the company's commitment to automation hasn't wavered. The chain is testing robotic fryers, automated beverage dispensers, and AI-powered kitchen systems. It's investing in mobile ordering, curbside pickup, and delivery infrastructure.

Voice ordering is just one piece. But it's an important piece. Drive-thru generates the majority of McDonald's revenue. Automating that channel could reshape the entire operating model.

The question isn't whether McDonald's will deploy AI voice ordering. It's when - and with whom.

For now, human order-takers are safe. But the clock is ticking.