Hacker News Reader

I think you’re right that is the core of the model. Good customers always get treated better than Joe Rando.

But Amazon has more information and uses that too. They know loss rates by zip code or by block by carrier.

Usually items I order for them that are higher shrink risk don’t go through their Amazon distribution network and come UPS or USPS. Anecdotally, I’ve noticed those returns are easier — they probably dump cost on the 3rd party shipper.

View on HN

The model isn’t aware of that stuff. They feed groups into the model, it might be grouped by zip code. But there isn’t a variable for zip code in the model. The model just forecasts LTV and nothing else, from transaction histories.

Do you know this because you work on the model?

That sounds like an unanswerable question.

Should be a yes or no answer? Either you worked on the algorithm or you didn't.

I asked the question because they spoke with the assurance of an insider... so I wondered if they worked on it.

Reply on news.ycombinator.com