It might not even be a prompt problem, but an overkill to a simple job. Sounds like the breaking point might be voice recognition - which I really thought would be much further along by now with all the other advancements. It seems figuring out what unique voices are saying is one of the hardest things to do well, and then add in background noises. Probably why the drive-thru AI attempts were terrible, working that job as a human can be difficult.