How would you balance failure attribution of that bad scenario between "own billing" and "problematic implementation" ?

I am not going to go into every nuanced issue, but to give context rounding errors alone costed us hundreds of thousands of dollars in customer support and eng time. I am talking Office Space style fractions of pennies type problems. Mostly due to how long it was an issue before it was caught.

Sure we could have probably botched implementation of some vendor, but we made every dumb mistake you could make and instead of doing it slowly we did it at scale.

Don't worry, every jurisdiction and bank will do the rounding of pennies differently, so you know, that's always a fun time.

Speaking of fractions of pennies: I've configured my AWS Cloud invoice to be in Euros, and every month the Net amount plus the VAT amount is always 1 cent off the Total fee :P

I wrote my own arithmetic functions just to deal with the paranoia.

Because they're actually the same thing. A team that thinks it's the right thing to build their own billing system lacks the experience and knowledge to do it correctly.