The projections are done by the city and state in this case, not Amazon. The city/state agents involved worked with a third-party (REMI) to perform a comprehensive forecasting and economic analysis. Also, most of the subsidies here are not in the form of capital investments, but rather things like tax credits which are provided for some action already taken (for example, having created a job that meets certain requirements).
This deal is very safe because it is paid out incrementally, over time, and mostly in-step with incremental actions taken by Amazon, and because the ROI is so big (even if projections are significantly off target, this will still be a win).
I don't think it matters whether Amazon or the city/state engaged the third party - the personal career/campaign finance incentives of the individual politicians/functionaries who negotiate for the city/state are aligned with Amazon's wishes and against those of the taxpayers.
And the third party's incentives are to please the paying customer in showing a good return on investment.
But if the subsidies won't actually get paid unless Amazon delivers on their promises, I agree this deal is way better than most of its type.
This deal is very safe because it is paid out incrementally, over time, and mostly in-step with incremental actions taken by Amazon, and because the ROI is so big (even if projections are significantly off target, this will still be a win).