The chief executive of a 3,500-restaurant chain found out his portions were too small from TikTok. Not from sales. Not from his operators. From customers with phones, filming their burrito bowls on kitchen scales, months before a single number moved. He caught it in time. Most owners don't, because they're watching the wrong instrument.
In the spring and summer of 2024 a genre of video took off: people filming their Chipotle orders and complaining the portions had shrunk. Some weighed the food on camera. A few filmed the person behind the counter to see whether being watched made the scoop bigger. It seemed to, which made for better video, which made more videos.
It got serious enough that analysts at Wells Fargo ordered 75 identical burrito bowls from eight New York locations and weighed them. Portions varied widely some bowls came in roughly a third heavier than others.
Chipotle's numbers at that point were fine. Good, actually. Revenue up, traffic up, the sort of quarter where nobody in the building thinks there's a problem, because on every instrument the company was actually looking at, there wasn't one.
The signal existed. It just wasn't anywhere the P&L could see it.
Here's the part I'd give them credit for. On the July 24 earnings call a good one, the kind where a CEO gets to talk about beating expectations Brian Niccol brought it up himself. "There was never a directive to provide less to our customers," he said, and then told analysts the company was retraining on portions and had "re-emphasized generous portions across all of our restaurants."
That's a genuinely hard thing to do. Nobody wants to introduce the one bad topic into their own good news, in front of analysts, on a recorded line. The instinct is to let it ride and hope it burns out. He didn't, and the story stopped being about shrinking portions and started being about a company that listens.
The counterfactual is what makes it worth studying. Say he skips it. The videos keep coming, the story hardens into "Chipotle is quietly shorting you," and eventually a reporter asks the question on a call where the numbers aren't good. Now it isn't a portion problem, it's a trust problem, and he's answering it from behind.
On the same call, Chipotle told analysts its cost of sales would rise to just under 31% of revenue the following quarter, up from 29.4%, and attributed 40 to 60 basis points of that to the portion fix. At a company doing roughly $3 billion a quarter, that's somewhere between $12 million and $18 million a quarter my arithmetic, not theirs spent fixing something that had been visible on phones for months.
Now shrink it. You run four dental offices. A hygienist leaves in March, the schedule at one location tightens, and appointments start running fifteen minutes behind. Nobody tells you, because nobody complains to the person at the front desk. They complain in the car, then in a Google review, then in the neighborhood Facebook group when someone asks for a recommendation.
Your new-patient numbers don't move for two quarters. They're noisy anyway seasonality, insurance cycles, a competitor's mailer. By the time the trend is clear enough that you'd bet money on it, you're six months downstream of the hygienist and you have no idea the two are connected. So you spend the next quarter on a marketing fix for an operations problem, which is the most expensive kind of wrong answer there is. It costs money and buys nothing, because the reviews keep coming.
Three reviews that say "waited forty minutes" are not three unhappy patients. They're the recommendation you stop getting, from a source you can't see, for as long as those reviews sit there.
Chipotle's real failure wasn't the portions. It was that a company with a real budget had no instrument pointed at the place its customers were talking, and had to be told by the internet.
You have less budget and better odds, because your version is smaller and easier to read. Three things, none of which require hiring anybody:
Niccol's advantage wasn't that he's smarter than you. It's that somebody in his building was paid to watch a channel that didn't appear on any financial report, and they told him while it was still cheap.
Nobody is doing that for you. And the channels have multiplied. When a benchmark of 10,000 US local businesses looked at who ChatGPT and Perplexity actually recommend, the most-recommended tier averaged 304 reviews and the least-recommended averaged about 118. The machines are reading the same reviews you aren't and they're assembling an answer about your business from them, on a schedule nobody tells you about.
Sometimes you own the narrative and sometimes it owns you. Mostly, though, it just gets written somewhere you weren't looking, by people who were never going to call and tell you.
Adapted from a piece first published July 31, 2024.
