In this case the two points where we should have caught the issue would have been:
* Integration testing with a specific data set that was different from actual production data.
* Code review (they're not systematic yet but we're working on that) because the faulty code had a sketchy section and the conversation around it would have lead to a better version (that's what actually happened when fixing the bug)
The retort to this is usually along the lines of “we are moving fast and need people to write 100% correct code”. Perhaps that’s the case, but the solution is to put guidelines down (api review, unit testing, etc) instead of demanding people be infallible.
In this case a reasonable test case beyond “Did it compile?!?” Would have caught this.