The Bellini Paradox
A drink that looked like a money-loser, and the assumption that was lying about it.
I built a report that does one small, honest job. For every cocktail on the bar menu, it works out the pour cost — the cost of the liquor in the glass divided by what we charge for the drink. If a cocktail sells for ten dollars and the liquor in it cost two, that is twenty percent. Twenty percent is a healthy number. Most of our drinks land right around there, which is the boring, good result you want from a report. You glance at it, nothing is on fire, you move on.
One line was not boring. The Bellini lit up red at about forty-two percent. More than double everything around it. On paper, a drink we were practically paying people to drink. Every time a bartender made one, the math said the margin bled out of it.
The instinct, when a number is that loud, is to act on it. Raise the price. Trim the recipe. Maybe pull the drink off the menu entirely and stop the bleeding. That instinct is fast, it feels responsible, and in this case it was completely wrong. The number was not telling me the truth about the drink. It was telling me the truth about something else, and I almost missed which thing.
What the number was actually measuring
Here is the part that is easy to forget when you are staring at a clean percentage. The report does not know what a Bellini is. It does not taste anything. All it knows is what we typed into the system. To get a pour cost, it has to know what a bottle of Prosecco costs per ounce, and to know that, it has to know how big the bottle is. Cost of the bottle, divided by ounces in the bottle, gives cost per ounce. Pour an ounce and a half into a glass, multiply, and you have the cost of the drink.
Somewhere along the way, the Prosecco had been entered as the little single-serve mini bottles — the four-packs you see at the grocery checkout. Per ounce, those minis are expensive. You pay a premium for the small format and the convenience. So the system did its arithmetic faithfully on a false premise. It believed every Bellini was being poured from those tiny, costly bottles, and it dutifully reported the drink as a disaster.
The bar does not pour from minis. Whoever is making a Bellini reaches for a standard 750ml bottle, the way you would expect. And a 750 is dramatically cheaper per ounce than a four-pack of minis — same liquid, fraction of the cost, because you are not paying for all that glass and packaging. The drink in the glass had not changed. The recipe had not changed. Only the number in a field had, and that one field was wrong.
Correct the bottle size to what the bar actually uses, and the Bellini drops from about forty-two percent to roughly twelve. Not a money-loser. One of the best margins on the whole menu. If I had trusted the red number and raised the price, I would have made a perfectly profitable drink less appealing, blamed a recipe that was never at fault, and felt good about it the whole time. The fix would have been worse than the problem, because the problem was imaginary.
The number tells you where to look. The looking tells you what it means.
What resolved this was not a smarter report. The report had already done its entire job — it flagged the one drink out of all of them that did not fit, and it pointed me at exactly the right question. That is genuinely all you can ask of data. It cannot walk back to the bar and check what is on the shelf. I had to do that part. I had to go see what we actually pour. The anomaly was the question; the bottle on the floor was the answer.
I think a lot of people, when they finally get the dashboard or the report they always wanted, expect it to hand them conclusions. It does not. A surprising number is not a verdict. It is a question wearing a verdict's clothes. The forty-two percent did not mean "fix this drink." It meant "go look at this drink, because something here does not match what you believe." Those are very different instructions, and the gap between them is where most bad decisions get made — confidently, from a chair, acting on a figure nobody walked out and verified.
There is an old, tired argument that the data people and the floor people are somehow rivals — that the numbers will eventually replace the gut, or that the operators who know the room can wave the spreadsheet away. The Bellini says both sides have it backwards. The report could see, instantly, the one drink out of dozens that was off. No human scanning a menu would have caught that. But the report could not tell me why, and it could not stop me from drawing the obvious, wrong conclusion. Only standing at the bar could. The data found the anomaly. The floor knowledge explained it. Take away either one and I am either blind to the problem or wrong about its cause.
So now I treat my own reports a little differently. When a number surprises me, I no longer ask what to do about it. I ask what it is really measuring, and whether I have actually gone and looked at the thing underneath. Most of the time the surprise is not a hidden problem in the business. It is a stale assumption I typed in months ago and stopped questioning. The Bellini was never the problem. An assumption about a bottle was. The report did not find a bad recipe. It found a bad belief — and it could only find it because I went and checked what the bar actually pours.
/ar/