If you read my article about Amazon’s new pay structure, and you also read my article about how talent is a power curve, the insightful among you may have noticed an interesting correlation between them. Specifically, could Amazon’s talent curve, when combined with the target compensation ranges, actually be a good way of compensating people “fairly?”

Well, let’s start with questioning the word “fairly.” As noted in the paper I cited in the talent post [https://www.hermanaguinis.com/pdf/PPsych2012.pdf], people don’t really like it if someone else is paid three or four times what they’re paid… even though the “someone else” is objectively three or four times more productive than they are. So “fairly” ends up a loaded term.

But I COULD believe an argument that asserts that’s what Amazon’s compensation ranges were trying to accomplish. (I’m not saying I do believe it – I’m just saying it’s possibly true.) Going back to my post, I had an example where the 50^th percentile for performance was 41% more productive than the least productive person, whereas the 90^th percentile was 216% more productive.

I’m pretty sure the top of Amazon’s compensation ranges are NOT three times higher than the bottom of the range, at least at levels 7 and below (Sr. Manager, Principal Engineer, etc.). Therefore, by at least one measure of “fair,” it seems it can’t be. The fact that talent is a power curve means you WILL have people that are two or three times more productive than another person in the same role. But the compensation ranges mean that you can only compensate them, what, maybe 50% more?

On the other hand, if you’re constrained by the market and by industry averages, and you want to try to pay those superstars closer to “fairly,” then maybe this new compensation model gets you closer to that goal. And by the same argument, if you can’t tell the difference between the 5^th and 10^th percentile, and the 20^th percentile isn’t much better than either of them, then maybe lumping them in at the bottom of the range is close to the best you can do? I don’t know, though – having 65% of the people making 20% or less of the pay range would imply a very steep power curve (i.e. an alpha well above 3). Maybe that’s justified? I don’t have the data to tell.

My gut and experience tells me that software engineers aren’t that polarized – there are some ineffective ones, a lot of somewhat effective ones, and while there are only a few hyper-effective ones, I’m thinking 1 out of ten, not 1 or 2 out of 100. That would suggest a flatter power curve than the one implied by Amazon’s compensation targets.