Bonded Wine Trade
Why fine wine prices are a data problem
Buy a case of Bordeaux and you quickly notice something odd: the identical wine, in the identical format, is listed all over the market at prices that can differ by double digits. Some of that is genuine supply and demand. A lot of it is just noise — merchants pricing off stale data, or off each other.
The catch is that you can’t see the noise until you can compare listings properly, and you can’t compare them until the data is clean. That turned out to be the whole project.
Three problems hiding in one
- Naming. “Ch. Léoville Barton 2016”, “Leoville-Barton 2016 12x75”, and “Leoville Barton, St Julien 2016” are the same wine to a human and three different strings to a computer.
- Format. Prices quote per bottle, per case of six, per case of twelve, in bond or duty-paid. Comparing them means normalising all of it to one basis.
- Freshness. A brilliant price that’s three weeks old is not a price.
Everything else in this project is downstream of getting those three under control. The next post covers the naming problem, which is the hardest.
(Draft — expand with a real example screenshot and your own numbers.)