The Tuesday Story · Edition No. 1 · 4 min read
Seven Times Normal
In 2004, Walmart asked its transaction logs what a hurricane looks like before it arrives. The answer became the founding legend of retail analytics—and it rests on thinner evidence than almost anyone who retells it knows.
Late August 2004, Bentonville, Arkansas. Hurricane Frances is grinding across the Caribbean toward Florida's Atlantic coast, and a million people are being told to leave their homes. Twelve hundred miles away, on the home-office campus of the world's largest retailer, the chief information officer is asking her team a question no meteorologist can answer: What do people buy when a hurricane is coming?
Linda Dillman had something better than a hypothesis. She had a rehearsal. Three weeks earlier, Hurricane Charley had torn through southwest Florida—and every purchase made in every Walmart in its path was sitting in the company's data warehouse, timestamped, itemized, store by store.
The team's brief sounded almost boring. Everyone already knew the disaster-prep canon: water, flashlights, batteries, plywood. The point of the exercise was to find what wasn't on that list — the demand hiding in the transaction logs that no one's intuition would ever volunteer. They found it.
The Fold
When the analysts ran Charley's numbers, the survival goods were all there, as expected. But the pre-hurricane top seller wasn't water or flashlights—it was beer. And the strangest signal was a breakfast pastry, strawberry Pop-Tarts, which Dillman told the Times was selling at "like seven times its normal sales rate" before landfall. Note the like. The most-quoted number in retail analytics is a spoken estimate from one interview. The finding was almost certainly real. The precision is folklore.
Strip away the anecdote, and the shape of the work is instantly recognizable to anyone doing demand forecasting today. Charley was, in effect, a labeled training example. What Dillman's team did has a name now—demand sensing: mining transaction history for signals of demand that hasn't arrived yet.
The modern version is a gradient-boosted demand model with weather covariates—storm-track distance, days-to-landfall, and category—trained on every hurricane since. The 2004 version was analysts querying a warehouse by hand. The statistical humility required is the same; only the tooling grew up.
Twenty-two years later, the trick is so routine it doesn't have a story anymore. Every serious retailer feeds weather into demand forecasts. And the regional translation that matters for us: in MENA retail, the biggest demand signal isn't in the sky; it's in the calendar. Ramadan reshapes purchasing more predictably than any storm front.
So here's this week's question for you: What's the strawberry Pop-Tart of your business—the demand signal sitting in your data that nobody's intuition would ever volunteer?
Sources & further reading
Constance L. Hays, "What Wal-Mart Knows About Customers' Habits," The New York Times, November 14, 2004.
Walmart corporate newsroom on Hurricane Florence preparations (2018).
Steve Horwitz on Walmart's Hurricane Katrina response (ABC News, 2011).
Got a story worth telling?
Pitch it — published stories get a byline in the edition.
Subscribe to the PetaFold Journal