I find this a bit defeatist. Estimating software projects is definitely possible, especially for less novel projects, and high-effort estimates will have higher

I find this a bit defeatist. Estimating software projects is definitely possible, especially for less novel projects, and high-effort estimates will have higher fidelity than low-effort estimates. Few projects are truly novel. You can't make business decisions without having some idea of estimates, so I find that arguments like this are used to weasel out of making any sort of commitment. Go ahead, add a big buffer, try to take uncertainty into account, pad, pad, pad - at least after all that we'll have some idea. On the other hand a very real problem I've often faced is when the target date is an input into the estimation process. I.e. we want to ship by May 2017, go estimate until the estimation fits this target date. Of course, after estimation the target date can be an input into requirements scoping, after which you can re-estimate.