Pizza Hut’s AI Lesson for Airline Retailing

Eric Marketts

Welcome to the V1 Airline Retailing Report. I'm Eric Marketts. A $100 million lawsuit against Pizza Hut sounds like a story for the restaurant trade press, but my guest argues it's really an airline retailing story. Jim Hetzel is VP of Product Management at ALTOUR, a travel technology veteran with more than three decades across Cirium, FlightStats and Amadeus, and he produces this show. On May 24 he published a LinkedIn article called Airlines Are Making the Same AI Mistake That Just Cost Pizza Hut $100 Million. Jim, let's start with the pizza. What happened?

Jim Hetzel

Thanks, Eric. One framing point first. Everything about the lawsuit is an allegation, and Pizza Hut has denied the claims in court filings. With that said, here's the story. A franchisee called Chaac Pizza Northeast, which runs 111 Pizza Hut locations across New York, New Jersey, Maryland, Washington D.C. and Pennsylvania, sued on May 6 in the Texas Business Court. It claims more than $100 million in lost business and enterprise value. The cause is an AI system called Dragontail, which Yum Brands bought in 2021 for $72.5 million, the figure I use in the article. It does kitchen management, order flow and delivery dispatch. Chaac says it was one of the strongest franchisees before the rollout. More than 90% of deliveries arrived within 30 minutes, and New York sales were growing at positive 10.19%. Afterward, more than half of its orders ran past 45 minutes, and New York sales growth fell to negative 9.78%.

Eric Marketts

That's a brutal swing. But you say the system worked exactly as designed. How does a system work as designed and still make pizza arrive cold?

Jim Hetzel

Because the design was wrong for that business. Before Dragontail, Chaac's managers keyed order ready status into a dedicated DoorDash tablet. It looked inefficient, but it gave them two things: control over the pace of orders leaving the kitchen, and the ability to block poorly rated drivers. Dragontail connected the kitchen directly to DoorDash, and that removed the human gate. Drivers could suddenly see order status, tip amounts, whether the customer was paying cash or card, and what was coming out of the oven next. They're gig workers. They optimize their own earnings. So they did the rational thing. They held completed orders for up to 15 minutes waiting for better jobs, and they skipped low tip and cash orders. The lawsuit says rack time, the time a pizza sits between oven and out the door, went from under five minutes to as much as 20.

Eric Marketts

So the information went to the party whose incentives didn't match the franchisee's. I can already hear the airline version of that.

Jim Hetzel

That's lesson one. NDC pushes richer fare logic, bundling rules, availability data and dynamic pricing parameters into channels run by aggregators and OTAs. That transparency is necessary. But handing your full offer intelligence to parties who don't share your revenue goals, with no governance over how they use it, isn't a distribution strategy. It's a data leak. An aggregator that has your offer logic can rank your products below a competitor's, arbitrage your pricing, or use your own bundling rules to sharpen their counter offer. The intelligence benefit flows outward. Your yield doesn't.

Eric Marketts

I'll push on that, because I'm optimistic about NDC. Richer content and transparency are the whole promise of modern retailing. Are you arguing against that?

Jim Hetzel

No. I'm saying transparency creates leverage for whoever receives it, and that isn't automatically you. Share the data, but know who gets it, what they can do with it, and what you get back.

Eric Marketts

Your second lesson is about those store managers with the tablets. What's the airline equivalent?

Jim Hetzel

Every airline operation has them. Revenue managers who step in where the system handles a market badly. Agents who know which exceptions to flag. Operations staff who override automated schedules when they collide with real conditions. The advice I give is simple: before you automate a human workflow, map what it was actually doing. The answer is almost never creating unnecessary friction. It's usually managing an edge case the system doesn't handle. And American Airlines is the expensive example. In April 2023 it pulled more than 40% of its fare content from legacy GDS channels, forcing agencies to connect through NDC or lose access. It had also cut its agency facing sales team and ended incentive contracts. Fourteen months later, CEO Robert Isom stood up at the Bernstein Strategic Decisions Conference and said, quote, we moved faster than we should have, and we didn't execute well. He also said, we've used a lot of sticks, we've got to put some more carrots in place. Travel Weekly reported the strategy was expected to cost American $1.5 billion in revenue.

Eric Marketts

And the chief commercial officer who built that strategy, Vasu Raja, announced his departure the day before. But back to the human side. If I'm a traveler whose NDC booking goes wrong, who's standing at that control gate for me?

Jim Hetzel

That's the real question, and it's lesson three. Dragontail worked in theory. Nobody adequately modeled what drivers would do with the information it generated. NDC does the same thing. The Offer and Order model is sound, and it works in certification and test scenarios. It breaks when it meets legacy mid office systems, PNR write back requirements, corporate travel policy engines, and all the workflows nobody documented because that was just the way things worked. Jim Davidson at Accelya has made the point that the real cost isn't search and book. It's what happens after ticketing: exchanges, refunds, EMD handling, disruption management. That's where the theory hits the real world. It worked in testing is not a distribution answer.

Eric Marketts

That connects to your fourth lesson on mandates. Chaac alleges Pizza Hut gave managers little training beyond how to enter an order. Is that what airlines do to agencies?

Jim Hetzel

Too often, yes. A compliance deadline gets set, certification requirements go out, and the support structure is a documentation portal and a quarterly webinar. That creates the appearance of adoption, not operational capability. Viktor Nekrylov at DRCT, an IATA certified aggregator, has noted the tension between management that wants NDC compliance and agents who don't trust a process that hasn't been proven in their environment. That's not resistance to change. It's a reasonable worry about servicing when something breaks.

Eric Marketts

To be fair to Pizza Hut, a franchise attorney, Ryan Palmer, told Franchise Times that franchisors have broad authority to impose system changes, and that disputes like this usually reflect implementation and communication gaps rather than flawed strategy.

Jim Hetzel

I'd say that supports the argument. Implementation is the whole game. The precedent will likely clarify training, timelines and accommodations, not whether you can mandate technology. For airlines, that's the lesson. You can have the right strategy and still fail on the operational details that decide whether it works.

Eric Marketts

Your last lesson is dependency. Chaac relied entirely on DoorDash, and Pizza Hut took over the national contract.

Jim Hetzel

Right. Dragontail didn't create that dependency. It made it catastrophic by giving DoorDash more information and leverage at the same time. Chaac then had to live with terms it had no part in negotiating, through a system it couldn't modify. Airlines can end up there with a single dominant GDS relationship, a sole source aggregator, or an OTA that represents 30% of indirect bookings. AI layered on top doesn't reduce the concentration. It makes you more legible to the party you depend on, and less controllable by you. Lufthansa Group is the better model. It pushed NDC above 30% of indirect distribution by pairing incentives with real investment in servicing, reducing GDS dependency progressively while keeping agent capability intact.

Eric Marketts

So if an airline executive listening wants one takeaway, what is it?

Jim Hetzel

Optimization is not alignment. A system can perfectly optimize a process while shifting value and control to someone whose incentives conflict with yours. So ask the questions first. What does this aggregator do with our full offer logic? What happens to agency behavior if we remove the fallback before servicing is reliable? What's our recourse if a mandated implementation fails in a partner's environment? And who actually benefits from the transparency this system creates? Yum, as far as the public record shows, never asked what a DoorDash driver does with that information. As I wrote, the lesson doesn't get more expensive. It gets more embarrassing.

Eric Marketts

Jim Hetzel, thanks for walking us through it. The article is on LinkedIn, and we'll see how the Pizza Hut case plays out in court. For the V1 Airline Retailing Report, I'm Eric Marketts.