Just the fax - differentiating voice and fax phone lines using call billing data

We survey an experiment in distinguishing voice and fax customers in a long distance telephony network based on the customers' calling patterns. We highlight three elements of algorithmic interest. First, we indicate how the theory of I/O-efficient algorithms productively guided the design of our software, whereas other algorithmic formalisms were less useful. Next, we describe a hyperplane-separator heuristic we developed for this problem whose performance here is comparable to that of an established learning algorithm. Finally, we discuss some open problems whose solution would have direct practical value.