Comment by mabbo | Hacker News Reader

Comment by mabbo | Hacker News Reader

Preface with the usual "I just work here and don't speak for the company", but from my point of view it always just comes down to numbers with Amazon. Make the presentation as pretty or splashy as you want but you're only doing that for your own sake, not Amazon's. You could have sent a handwritten letter with the same information.

Somewhere in the company will be a database. In it, there will be one column for every property of a city, one column for each property of the bid received (tax breaks included of course). I'll bet there will be hundreds of columns. And then there will be a very large formula that takes every property, assigns it a value to Amazon over the next 25 years, and adds it all up.

It's probably quite a complex formula. I have a friend with a master's degree in computational geometry who used to work on fulfillment center placement optimization. Lots of software dedicated to that I hear. Amazon is a data-driven company- it would never leave such problems to human intuition.

Once the data entry was done, and probably a few dozen formula updates were made as new factors were realized and added in, the decision would have been made automatically. The only problem left would be announcing it at the best time to further profit Amazon- another factor the formula probably spit out.

My guess is that we'll know when Q4 results are published. I'd love it to be my home city of Toronto, but I think the lack of tax breaks offered may push it down in the formula's ranking.

I agree as a matter of general principle. But these kind of decisions are hard to completely automate. Even if the final output is a number, you need to make judgement calls that go into the number. What's the value to Amazon to the goodness and population of schools within X distance vs. walkability vs. tax breaks? And then, what are the real deal-breakers?

You can certainly use numbers to cull the field. But you probably end up with a few cities that all look good on paper and on the numbers but have different strengths and weaknesses that are more qualitative.

Correct me if I'm wrong, please. Doesn't any sufficiently-complex non-linear optimization problem degenerate into heuristic trial and error? There's a progression of better and better solutions. With the unpredictability of the real world, and the low number of trials they'll get (this is their second HQ), I wonder greatly what they've got going on.

I studied some of this stuff at one point and have been involved in evaluations of different system architectures and the like as an analyst. You can assign weights, impute values to qualitative variables, define "must haves," etc. to create a somewhat quantitative analytical framework. This has the virtue of at least being consistent and ostensibly objective.

But once you get beyond the tax breaks and the cost of living and whatever decisions Amazon has probably already made about geographical dispersion, it will probably come down to Bezos making a call. This isn't siting a new Marriott hotel or a distribution center.

I'd expect for a decision like this, there would in fact be some subconscious (or even conscious!) ideas like "Obviously Detroit should be high on the list, if it isn't then our weights must be off," or alternatively "Detroit's proposal just doesn't look good for X reason, if it's high on the list then we should add a column about X." It's not totally unreasonable to expect that these ideas can be influenced by the usual ways that you influence humans, even if those humans work at Amazon.

Right, but I think the point was that those judgement calls were made before the proposals were sent, so at that point the human factor was done.

If this was true, they should make this spreadsheet public (google doc) with the columns set by any mayor. Basically open an auction. Why hide it?

When you are trying to extract value, you don't want to reveal your estimate of the value extracted, or even give people more information than necessary about how you are estimating the amount of value you can extract.

I.e. you'll get gamed.

Uh. So when you go to an auction and bid, the auction house gets gamed? I am not following. Each mayor could put their best offers in compared with the others.

You dont negotiate with an open book.

As the UK is finding in negotiations with the EU...

Parts of the UK are finding that negotiating _without_ any kind of playbook is hard - nothing to do with it being open or closed, it simply doesn't exist.

Reply on news.ycombinator.com