Hacker News Reader

Comment by crazygringo | Hacker News Reader

> “We strictly prohibit employees from using non-public, seller-specific data to determine which private label products to launch,” a company spokesperson said in a statement.

So employees can use non-public, aggregate seller data? Or non-public manufacturer data?

Yeah. The "seller-specific" part makes the statement meaningless.

View on HN

The distinction between aggregated and non-aggregated data is moot. From the WSJ article:

> Some executives had access to data containing proprietary information that they used to research bestselling items they might want to compete against, including on individual sellers on Amazon’s website. If access was restricted, managers sometimes would ask an Amazon business analyst to create reports featuring the information, according to former workers, including one who called the practice “going over the fence.” In other cases, supposedly aggregated data was derived exclusively or almost entirely from one seller, former employees said.

> Amazon draws a distinction between the data of an individual third-party seller and what it calls aggregated data, which it defines as the data of products with two or more sellers. Because of the size of Amazon’s marketplace, most products have many sellers. Viewing the data of a product with a number of sellers wouldn’t give it insight into proprietary seller information because the figures would show lots of different seller behavior.

> Amazon said that if there is only one seller of an item, and Amazon is selling returned or damaged versions of that item through its Amazon Warehouse Deals clearance account, Amazon considers that “aggregate” data—and hence is permissible for its employees to review.

It's not moot at all.

Uber employees (and some cities) can get access to aggregated data showing where uber is driving.

The same employees should not be able to have access to where YOU are driving (rider specific).

This is particularly key for sellers with a unique product. Chairs are selling well? No problem. But detailed transaction detail for a specific seller and product and their decisions there - a problem.

If I'm not misinterpreting, the post you're replying to wasn't claiming that a general aggregation in principle was the same as individualized statistics, but rather saying that Amazon's definition of "aggregation" effectively mooted the normal argument for it. In particular that last bit, where Amazon would count it as "aggregated" even if the only "other seller" were Amazon themselves in the form of reselling returned/damaged/refurbed goods. In practice that would probably be pretty close to having individualized analysis generally available, if not identical (because employees could separately get Amazon Warehouse Deals alone, since that's "in house", and then simply delete it from the "aggregated" sales and have both).

But if your city only has one Uber driver, the fact that the data is aggregate is meaningless since it still tells you the exact location data of that one Uber driver. I think that's the point that's being made.

If you can find a way to make data "aggregate", you can abuse it. How about aggregate data on all Uber drivers named Bill who drive a 2014 beige Camry with a license plate ending in "DEF". Yes technically this could return multiple results, but in practice it's irrelevant.

You read a lot from aggregated seller and product data. And if it is really a unique product, just aggregate data for similar, or equal products. And you are 95% there. More than enough to make decisions. And in most cases, you have more than one seller per product, even more so for suitable private brand products.

Not to mention, they could simply subcontract the new private label product recommendation work to a consulting company and provide the data to them. Then, an "employee" wouldn't be using seller-specific data (a contractor would).

Reply on news.ycombinator.com