Whilst Clintons use case might be valid — could it also be an edge case?
What if the majority of users need to calculate how much they need for the day/week/month i.e. I need $20 for coffee, $40 for lunches, $139 for birthday present etc. In this use case, it would be very difficult for the system to predict a suggested amount — thus making the calculator quite useful!
Moreover, I wonder if having a calculator actually decreases queue times — it would be interesting to test ATMs with and without one.
But I do agree — could be problem worth solving.