Behavior analysis of low-literate users of a viral speech-based telephone service

We studied the behavior of users of a simple telephone-based voice modification and forwarding system, which has operated in Pakistan for about a year, attracting 165,000 users who interacted with the system by telephone over 636,000 times and generated very rich interaction data. Trying to cluster the users based on their activity profile, we found that they form a continuum rather than truly distinct clusters. We did discover that, with experience, users respond faster to menus (using more barge-in) and make fewer mistakes and abortive attempts. Finally we studied how users' choice of activity evolved over time, and found that with experience users show an increasing interest in message sending, become more explorative of the system's capabilities, and better adapt themselves to its constraints. Many new users seem to arrive with some preexisting knowledge of Polly's functionality, presumably through some back-channel information from their friends. Long-term users engage in lengthier calls from the start, and take a more active interest in voice modification and forwarding features.