Yum Brands × Nvidia: rolling out AI drive-thru after McDonald's failed in public.
Taco Bell, KFC, and Pizza Hut are deploying AI ordering across hundreds of US locations — taking the slow, partner-led path their biggest competitor abandoned in 2024.
The Initiative
Yum Brands announced a multi-year partnership with Nvidia to deploy voice AI across drive-thrus, internal call centres, and back-of-house operations. Taco Bell is the lead pilot, with hundreds of US restaurants using AI ordering in 2025. Crucially, Yum is rolling out after McDonald's ended its IBM-powered drive-thru pilot in mid-2024 following viral TikTok failures — bacon on ice cream, 260 chicken nuggets, the whole reel.
Slide 2 — Objective, Challenge & Class Concepts
Speed, accuracy, and labour cost — at the loudest customer touchpoint there is.
Objective
Reduce labour cost in the toughest hire-and-retain segment of QSR, increase throughput at the drive-thru window, and free human staff to focus on food prep and complex orders.
Challenge
Drive-thru is the most public, most viral failure surface a fast-food brand has. One bad order becomes a TikTok with 30M views. McDonald's already paid the brand-equity cost. Yum has to be measurably better, not just measurably cheaper.
Class Concepts
AI in Operations: back-of-house cost reduction at customer-facing scale.
Customer Experience: a brand interaction is happening, like it or not.
Fail-Publicly Risk: learning curves are funnier on social media.
Late-Mover Advantage: watching the leader stumble can pay for itself.
Slide 3 — Viewpoint & Recommendations
The drive-thru is the worst place to learn AI in public.
"Voice ordering should be the third place you ship AI, not the first."
My recommendations
Lead with chat, not voice. Build AI ordering inside the app first — text is forgiving, voice is theatre. Get the menu logic right before you put it on a speaker.
Always-on human override. The cashier inside the store should be one button away from taking control of any order, with no penalty in throughput metrics.
Train on failure clips. McDonald's TikTok library is the world's best QA dataset for this exact problem. Use it.
Localize the voice. A Texas Taco Bell sounds different from a Brooklyn one. National-average voice assistants will fail in regional markets.
Publish a "we got it right" rate. If accuracy is the differentiator, make it the headline. Don't let TikTok set the narrative.