The Tornado Guarantee

In 1884 an American army sergeant began forecasting tornadoes twice a day, and was right 96.6% of the time. John Finley, of the Signal Corps, issued a tornado-or-no-tornado call for eighteen districts of the Midwest every morning and every afternoon.

Across 2,803 forecasts he was wrong just 95 times. He was, on his own numbers, very nearly a guarantee.

A geologist called G.K. Gilbert looked at it from a different perspective. Tornadoes are rare. If Finley had forecast "no tornado" every day, without looking out of the window, he would have scored 98.2%.

Last week a retail media network published their own 94% outlook. And just like the difference between Finley's score and Gilbert's, there is a gap between predictive targeting and a retail media network selling you an outcome.

Sam's Club Connect, the advertising arm of the warehouse chain Walmart owns, announced Predictive Precision Targeting on Wednesday of last week. It uses the membership data, machine learning and "more than 400 intent signals" to identify members by what they are likely to buy next, and what they will be worth over the twelve months after they do. Harvey Ma, who runs it, said: "Membership changes what advertising can do."

The proof point is a P&G test. Sam's Club identified about 800,000 households that had never bought Cascade Platinum Plus dishwasher pods but were, on the model, likely to. P&G reached them with connected TV and online video. The case study reports three numbers: a 2.3x conversion rate, a 71% higher projected twelve-month value for the buyers it acquired, and 94% accuracy. Jody Johnson, who runs P&G's Sam's Club team, said the campaign showed how much more precise they could be, and P&G has since extended the product across more brands.

This is not a small announcement from a small company. Walmart's global advertising business grew 38% in the quarter reported five weeks ago, with Walmart Connect in the US up 43%, against revenue growth of 5.9%. Last fiscal year it grew 46% to nearly $6.4 billion. The retailer everyone thinks of as a 4% margin business is, on its own disclosure, a company where advertising and membership together produce about a third of operating income. Advertising is to Walmart what popcorn is to a cinema.

So the forecast comes from the most consequential retail media business outside Amazon. It deserves to be taken seriously, and taken apart.

In a membership club even a cash purchase is tied to a household account, and the record runs for years across every category. If any company can tell you that a household buying soundbars and games consoles is a household about to buy a television, it is this one.

The test was also the right test. It went after households that had never bought the product, and the best evidence we have on where advertising actually works points exactly there. When two economists at eBay switched off the company's paid search in 2012 and watched, the ads turned out to have no measurable effect on people who already shopped there. The effect that survived the experiment was among new and infrequent customers. Sam's Club aimed P&G's money at the one audience where the science says money can do something.

P&G does not join any external comms lightly, so I trust their version of the numbers. A brand that can be told which of its buyers are drifting before they have gone has been handed something of Ehrenberg-Bass level value.

Last year, in The $179.5 Billion Tollbooth, I wrote: "In many ways, my first job was in retail media." I was describing the Unilever sales-rep years, the wobblers and the stapler and the aisle-end of John West tuna. It made a good opening, and it wasn't true. That was my fifth job. After working at Pizza Hut, a video store and as a mobile DJ (yes, you read that right) through university.

The first was the cool room at the back of a Woolworths in Sydney. I was fourteen, as young as the law then allowed me to work, and my Saturday began with carrying the bananas out to the stand before the doors opened and then working along the melons with a labelling gun, pricing each one by hand. This was way back when there was one type of lettuce and an avocado was very fancy indeed.

In the produce department, someone forecast what would sell. They knew, by Thursday, how many boxes of bananas that suburb would move on a Saturday, and they ordered to it. When they were right nothing went brown in the cool room. That was a prediction.

As a Unilever sales rep my job was the opposite. Nobody had forecast that the John West display at the end of aisle six would sell four cases more than the shelf alone. It sold four cases more because I built it, stapled the posters to it, and talked the manager into leaving it there for the week. That was causation. The store had a forecaster and the supplier had a rep, and neither of us would have dreamed of claiming the other's number.

Which job is Sam's Club doing with this retail media product?

Read the P&G case study's footnote and the three numbers separate into the two jobs. The 71% and the 94% belong to the forecaster. The model predicted how much each newly acquired buyer would spend over twelve months, then tracked them, and the prediction came within 94% of what happened. That is a creditable forecast. It is the produce manager, right about the bananas.

It is also a forecast about households that had already started buying the product. It says nothing about whether the advertising made them start. The number that would tell you that, the share of the 800,000 who would have upgraded to the pods anyway, appears nowhere. There is no holdout group in the account. The 2.3x is a conversion rate with no baseline attached. And the measuring was done by Omni-Impact, Sam's Club's own tool, launched in June last year specifically, in its own words, to quantify incrementality. A company that says it can measure incrementality published its flagship case with a conversion rate and a forecast score, and no incremental figure. The no tornado.

If a model can tell you with 94% confidence who is about to buy dishwasher pods, then most of those sales were coming whether or not P&G ran the ad. The advertising's contribution is whatever is left once you subtract the people who were going to buy anyway, and only a holdout reveals it. A perfect prediction leaves nothing for the advertising to have caused. Finley's guarantee was on the wrong thing, and so, until someone shows the counterfactual, is this one.

Predictive targeting is one step from the guarantee: not "we reached 800,000 households" but "we will deliver you 400,000 new buyers, and you will pay per buyer". Every marketer reading this would sign that contract tomorrow. And its price will be set by the party that made the forecast, using a measurement tool it owns, against a baseline nobody outside has seen. Finley, selling tornadoes, marking his own homework.

The last thing the case study tells us is perhaps more interesting. The media was connected TV and online video, the kind of buy a media agency would ordinarily plan. No agency is named. P&G buys almost all its US media in-house, as Marc Pritchard has said for years. And Johnson's own line about what the test means is not about media at all: "This could revolutionize our joint planning and be the foundation for future JBPs." To be clear, this means that in the P&G/Walmart JBP, which is about volume commitments, pricing, innovation, unique Walmart SKUs etc… will include retail media investment planning too. No agency layer needed.

That is the structure I wrote about last August, when I said Walmart would join Meta and Google in bypassing the agency. I was right about the outcome and wrong about the route. In June it started pushing its audiences and closed-loop measurement into the third-party buying platforms agencies already use. Now it has simply moved the decision into the joint business plan, where the retailer sits as store, media owner, forecaster and measurer, the supplier sits as buyer, and the budget is agreed in a meeting no agency attends.

Today I do as much shopping as I can from a screen. But I still like walking a supermarket, and last weekend I walked Sainsbury's, slowly, looking at brands and discovering new ones. The peas, fish, pizza, pasta sauce, condiments, tea, coffee and soft drinks that I care for now and in the past. It's all 'clean store' now, so few wobblers and no reps with staplers.

The shelf never told me why someone reached for something. It told me that they had, and when I was fourteen a man with a clipboard could tell me roughly how many would, tomorrow.

Finley was right 96.6% of the time. The weather did not need him. A guarantee is only worth what the seller can tell you about the days the tornado never came.

Brandflow is written by someone who has priced the melons, built the display and now signs the retail media budget, which is three ways of learning the difference between a forecast and a result.

First published 24 September 2026 in the Brandflow newsletter on LinkedIn.