McDonald's Sent One Customer 515 Pages. It Predicted He'd Never Leave.
· CX Pulse
A Wired journalist asked McDonald's for his loyalty data and got 515 pages. The transactions were expected. The predictions were not.
McDonald's Sent One Customer 515 Pages. It Predicted He'd Never Leave.
Earlier this month, Wired's Reece Rogers asked McDonald's for the data it holds on him. He got back 515 pages.
Most of it is what you'd expect. Years of transactions, which restaurants, which items, every promotion he interacted with, app activity going back as far as the account does.
The part worth reading twice is the part McDonald's worked out on its own.
The predictions
The file estimated he would visit 2.16 times over the following six weeks. That he'd spend $13.49 on an average visit. That he'd spend $29.15 in total.
And it put his chance of leaving at zero percent.
Not low. Zero.
McDonald's says the obvious thing in response, and it's a reasonable thing to say: "We use information such as past purchases to provide a more engaging, personal customer experience." The loyalty programme has 210 million active users, and the CFO has called it the company's single most important digital metric.
None of this is a leak or a breach. He asked, and they sent it, which is how it's supposed to work.
Why this reaches you even at a fraction of the size
You are almost certainly doing a smaller version of this and may not have thought about it in these terms.
If you run a loyalty scheme, a CRM, an email platform with engagement scoring, a booking system that flags regulars, or a helpdesk that tags accounts by value, then you hold two different kinds of information about people. It's worth separating them in your head, because customers separate them instinctively.
The record is what they gave you. Orders, visits, tickets, emails. Customers expect you to have this. Nobody is upset that their coffee shop knows they order flat whites.
The inference is what you decided about them. A lead score. A churn risk flag. A lifetime value estimate. A tag that says "price sensitive" or "high maintenance." Nobody handed you those. A system produced them, usually without anyone reading the output.
The 515 pages became a story because of the second kind.
The question that actually matters
If a customer asked to see what you hold on them, could you explain the inferences out loud?
Not the orders. Those explain themselves. The score. The flag. The estimate. Could you say where the number came from, what it changes about how they get treated, and how someone gets out of a category once they've been put in one?
Most businesses can't answer the last one, and that's the uncomfortable bit. Inferred fields tend to be sticky. A customer flagged as low value two years ago after one bad quarter often stays flagged, quietly shaping which offers they see and how fast their emails get answered, long after the behaviour changed.
What this isn't
This isn't an argument against personalisation. Personalisation is mostly good and customers broadly like it. A shop that remembers your usual is a better shop.
It also isn't a McDonald's problem. Starbucks and Kroger run comparable programmes, and the only reason we're looking at this one is that a journalist filed a request and then wrote it up.
The five minute version
Open whatever tool holds your customer records and find one inferred field. A score, a segment, a risk flag, a tier. There will be one.
Then ask two questions about it. Where does the number come from, and what changes for the customer because of it?
If nobody at your company can answer both, you have a field that's affecting how real people get treated and nobody is reading it. That's worth ten minutes whether or not anyone ever asks you for their file.
Source: Morning Brew, reporting on Wired's Reece Rogers, 14 August 2026.