Theleponic conversations with call centern agents follow a fixed pattern, commonly known as call flow. Each call flow is a sequence of states such as greet, gather details, provide options, confirm details, conclude. We present a mechanism for segmenting the calls into these states using transcription of the calls. We also evaluate the quality of segmentation againts a hand tagged corpus. The information about how the agents are performing in their calls in crucial to the calls center operations. In this paper we also show how the call segmentation can help in automating the monotoring process thereby increasing the efficiency of call center operations.