That only works under the assumption that you have time to come back to it. I've also seen cases where you just get 15 years of cruft, good intentions and bad implementations, at which point people never want to touch it again, because the last time they did they accidentaly broke a client's system.
There's a 6-month startup phase in just about every SW project I've led that sets a strong course for the lifetime of the project that is awfully hard to correct, both for better or for worse. My advice is always suck it up early and collect the dividends down the road.
Make it work, make it right, make it fast.
I've shared that with a couple teams in the past, and it always catches on.