- The Cautionary
- Posts
- Learning AI #27
A Pizza Hut franchisee just told a court that AI cost it $100 million.
Pizza Hut is structured as a franchise model, where restaurants are owned by individuals, but are part of the Pizza Hut system. Franchise owners are required to display proper signage, adhere to menu standards, and buy supplies from Yum Brands, which owns Pizza Hut.
Chaac Pizza Northeast runs 111 Pizza Hut locations across the northeastern US. In May 2026 it sued Yum Brands, over Dragontail, a mandatory AI system that all franchise owners were required to install.
Dragontail was supposed to route orders more efficiently. It routed Chacc’s Pizza Huts into oblivion.
Pizza Hut was a model of efficiency, until it incorporated a new AI ordering and scheduling system.
Dragontail handed DoorDash drivers a live feed of order details: what the tip was, whether it was cash, what else was waiting at the same store. Drivers used that information the way you'd expect.
Drivers picked the high-tip orders and left the rest.
Chaac says drivers held ready-to-go pizzas for up to 15 minutes while cherry-picking better deliveries, and often skipped low-tip or cash orders entirely.
The numbers in the lawsuit are specific. Rack time, how long a pizza sits after it's made, went from under 5 minutes to as much as 20. Delivery time went from under 30 minutes to more than 45.
Only about half of orders arrived within the 30-minute window Pizza Hut advertises. In New York City, year-over-year sales growth flipped from positive 10% to negative 10%. Ouch.
None of that is a hallucination or a chatbot inventing a policy. It's a system built with good access to data and no thought about incentives. Dragontail knew everything about every order. It just handed that knowledge to the one party in the chain with a reason to use it against the franchise owner: the driver deciding which orders were worth their time.
The failure here was the deployment. Someone decided a delivery-dispatch AI should be mandatory across every franchise location, including ones like Chaac's that ran entirely on DoorDash drivers rather than in-house staff. Nobody asked what happens when a gig worker sees a tip amount before deciding whether to grab the order.
The AI system did not account for the basic human instinct of greed.
You don't need to run 111 pizza stores to hit this same problem. Any AI tool that's handed real operating data, whether that's a scheduling assistant, a pricing model, or a customer-routing bot, will get used by whoever has access to it in whatever way benefits them most. (This is called human nature.)
If the incentives aren't checked before the rollout, the AI doesn't fix the process, but makes it worse.
Chaac is asking for $100 million plus fees and interest. Yum Brands hasn't said publicly how it plans to respond. Whatever a judge decides, the lesson is already in the numbers: a 30-minute delivery time became a 45+ minutes because the software worked exactly as designed, and nobody checked what "designed" meant for the people using it.
More AI gone wrong
As I was finalizing this post, I ran across the following.
Meta employees claim their livelihoods are at the mercy of an algorithm. Again. |
Twenty-six workers have accused Meta of using faulty, AI-driven performance metrics to gauge employee productivity and dictate layoffs. That’s because the metrics didn’t account for workers dealing with illness, pregnancy, bereavement, or disabilities. |
TL;DR — This flawed scoring system (allegedly) led to illegal layoffs, since medical and family leave is protected at the federal and state level. |
Things I think about
There are more than 40,000 Google searches performed every second.
**********