We present the first experiment we conducted to evaluate the attention monitoring performance of Louise, following a Wizard of Oz method, during the interactions with a cohort of 8 elderly users in a day hospital environment. Louise is a new, semi-automatic prototype of an Embodied Conversational Agent (ECA), a virtual character interacting with users through social-like communication, adapted to the special needs of older adults with cognitive impairment; it is intended to ultimately provide assistance in their activities of daily living. We recorded and analyzed both videos of the conversation-like interactions and Louise's tracking data. In our experiment, Louise's attention estimation algorithm achieved about 80% accuracy; moreover, in almost all cases, the user's attention was successfully recaptured by Louise after a planned, experimenter-induced distraction. These results are in line with what was observed in previous experiments involving only younger adults, thus suggesting that attention measurement tools embedded in cognitive prostheses will not need to be adapted to elderly patients. Finally, to gain further insights on conversation management and provide evidence-based suggestions for future work, we performed an anthropological analysis of the whole experiment.