In this study, we combine behavior-driven structural model with machine learning tools and apply them to data on a large scale. We investigate the important question of channel interdependence in the context of multi-device advertising. In particular, we explore how the distribution of ads on multiple digital devices (i.e. tablet, smartphone and PC) works together to affect the final conversion. We model the marginal impact of an advertisement impression on an individual consumer’s behavior based on the conversion funnel theory. To handle the sheer volume of the impression level data and the iterative nature of the estimation procedure, we develop a novel estimation algorithm that can distribute the data and computational burden in parallel on cloud computing infrastructure. Our preliminary results show channel diversity stimulates disengaged consumers to transition to engaged stages. But it does not encourage the already engaged consumers to stay engaged. In addition, the channel diversity is more effective for the early impressions than for the late impressions.