In this work, we present an experimental analysis of a Dialogue System for the automatization of simple telephone services. A rst evaluation of a preliminary version of the system was done based on the Speech Recognizer error rate and on the identi cation of two groups of users, that we refer to as group A and B. From this evaluation we conclude the necessity to design a robust and exible system suitable to have di erent dialogue control strategies depending on the characteristics of the user and the performance of the speech recognition module. A system adaptation procedure combining a normalized average number of utterances per task, the amount of information in some particular utterances, and an estimate of the recognition error rate. Our adaptation procedure de nes two di erent control management strategies for a exible mixed-initiative strategy: xed for high recognition errors and Group B users, and mixed for low error rates and A users. Experimental results following the PARADISE framework showed an important improvement.