$10B to place chips next to beer

Supermarkets know that chips sell better next to beer. They know it because they paid to find out. In American grocery alone, suppliers hand the shops roughly nine billion dollars a year for nothing more than a place on the shelf,¹ and then everybody pays again for the research that says which place. Call it ten billion. A year.

I got the same knowledge for free in 2020, when I joined a grocery delivery app where the bread was not next to the bread.

Bread next to bread

The app delivered groceries in fifteen minutes. I owned its catalog, on iOS and Android. And when I arrived, that catalog was organized the way the products sat in the database: by supplier. One company’s bread, beer and washing-up liquid lived together. Another company’s bread lived somewhere else.

Nobody had decided this. That is the point. A catalog that nobody designs takes the shape of the table underneath it.

It matters, because people do walk the aisles even when the aisles are pixels. In an eye-tracking study of forty online grocery shoppers in the UK, 95% found their way around by department, and 80% also used search.²

A shop is an interface, and shops needed a century to get theirs right. The first self-service grocery opened in Memphis in 1916 with 605 products, four aisles and a turnstile.³ Before that you handed a list to a man behind a counter. A supermarket today carries more than thirty thousand products,⁴ and nearly every one of them stands where it stands on purpose.

What shops paid for

Start with the rent. In 2003 the American competition regulator went and asked what a place on the shelf costs. Between $2,313 and $21,768, per product, per chain, per city. Suppliers told them a national launch needed one and a half to two million dollars before a single pack was sold.⁵

Then comes the question of which place. That one has been measured to the decimal point.

My favorite person in this history is the architect who drew the first enclosed shopping mall in 1956. The moment when you walk into a shop, lose the thread and start buying things you never came for is named after him. He hated what his idea turned into. In 1978 he said of the malls built in his image: “I refuse to pay alimony for those bastard developments.”⁹

Beer and diapers

Every product person has been told the one about beer and diapers. Diapers, in the original American. It is the founding story of the whole trade, and it is worth hearing properly.

I like the true version much more. The most famous finding in retail analytics never changed a single shelf in the shop it came from.¹⁰ It became true five years later, because somebody believed it.¹¹

Reading the answers

So we did not run a research program. We read the answers.

Bread went next to bread. Categories were rebuilt around what a person came to do, which is the oldest idea in the trade: a category is a need, not a supplier.¹² The top of the screen became the eye-level shelf, because on a phone the first screen is the only shelf everybody sees. Some positions became paid ones, as they are in every shop on earth.

And then the shelf that gave this piece its title. The rule was simple: whatever is likely to be bought together stands right next to each other. Chips next to beer, diapers next to the baby wipes, and sometimes, as the legend would have it, next to the beer as well.

Here is where online beats any supermarket. A shop has one wall, so the shelf next to the beer is the same for everybody who walks past it. An app has a wall for each person. We put a machine learning model behind that shelf, trained on what people actually put in their baskets together, so the row next to the beer was right for whoever was holding the phone.

It was not all borrowed taste. We A/B tested the new catalog against the old one, and the new one converted 20% better, which was worth about half a million pounds a month. Together with the ML suggestions, the categories built around intent lifted thirty-day retention by 40% and added about £5M a year in recurring revenue. And since nobody should be placing products on shelves by hand, we automated the boring part.

Ten billion dollars

None of this was clever, and I say that with some pride. It was product sense with a very expensive bibliography. Somebody else had paid for the shelves, the six-store experiments and the 1.2 million baskets. We showed up at the end and copied the homework.

Online we had exactly one unfair advantage. Moving a shelf cost us a row in a database. The same database that started all this.

The chips are next to the beer now. That will be ten billion dollars, please.