[spongeb00b]

I repeatedly see features being placed into a "phase 2" that appears approx 90% into a project as the deadline looms.

Half of those features eventually get implemented, the rest it’s realised are not needed at all.

[alkonaut]

That's the correct way of going about any software development. The problem though is the customer will likely be unwilling to agree to a contract that doesn't have at least 75% of the scope or the product is "unusable". Their existing/previous product is usually better all the way until every single feature of it is available in the new system (I have not yet in my career ever seen a system that isn't replacing a previous system - at least if you count an excel sheet or a piece of paper as "previous system").

What you end up with when shipping an MVP is you also convert a minimum number of users and you end up with two systems. The "MVP" is thus often "whatever allows all existing users to switch" which is why projects snowball to be too big and too late. You won't realize what people used the old system for, until you try to replace it. And it always turns out it has 2000 features when your discovery had found 20.