Definitely. And being both sides of that table at some point, it is always possible to give an estimate. The difference is that the estimate should be a range

Definitely. And being both sides of that table at some point, it is always possible to give an estimate. The difference is that the estimate should be a range. How tall is the Eiffel Tower? If you didn't know and gave an estimate you might say, 172m is quite high, maybe that's high enough to be the answer without over-shooting. You could give a range as the answer: Between 100m and 500m tall. It's definitely more than 100m high, for why else would there be lifts, such a wide base, etc, etc... and it's definitely less than 500m, because it certainly doesn't rise half a KM into the sky. That first estimate... wrong. Wildly wrong, but because it is a fixed and precise number it gives the illusion of accuracy and knowledge on my part where none was applied in coming to the number. As an estimate, the client will be upset. That second estimate... well now I've Googled and the answer is 300m and so it turns out the estimate was correct. That's all a client wants... the best and worst case, and when there is a huge variance between the two, what is the basis for that variance? How much of it comes back to the clients as requirements that could be made simpler? How much is the tech choice? The features? Always give an estimate, just make it a range.