> Why not just ship what's ready at the end of any particular cycle, and spill over whatever's not?
We do, but many times customers are paying many thousands of dollars for feature requests and have particular deadlines they want it by. Especially in our industry where there are a handful of competitors fighting to get huge contracts. We sell to insurance companies, so there is a lot of legacy software being replaced.
> and the only real solution that I've seen is to underpromise work on purpose, every time
Yep, we've become much better at underpromising in order to protect ourselves