It is projected that by the year 2033, 41% of the population in the United Kingdom will be aged 60 and over, compared to 17% in 2010. However, as people get older, age-related impairments to working memory and cognition, as well as sensory impairments, mean that older adults struggle to adopt new information technology systems, which have become integral to everyday life in recent years. The research presented here investigates, across a series of three large scale experiments, how Embodied Conversational Agents (ECAs) can be used in virtual world applications, addressing three important domains of application space, to understand how ECAs may be used to engage with older users. The Virtual Banking Experiment investigates whether or not ECA gender or age portrayal has an effect on overall usability of virtual world applications for older users. The results of the experiment show that older users do not exhibit a significant preference for a particular ECA gender, however they do exhibit a preference for interacting with younger ECAs when compared with older ECAs in these applications. Results of the Entertainment Experiment show that scripting the interaction, between older users and ECAs in a virtual world quiz game, significantly improved the usability of these virtual world gaming applications employing ECAs as interlocators. The Social Support Experiment investigated three styles of providing advice to older users as part of a virtual advice bureau service. The results of the experiment show that older users prefer to receive advice from ECAs that has not been presented as having been obtained from real-world experiences.