Many IVR designs are based on reasonable assumptions about caller behavior, but before-and-after data from multiple deployments can determine if these assumptions are justified, or if the data actually challenges some long-held beliefs about VUI design. This session presents a recent case study with access to pre- and post-deployment metrics and examines several data-driven findings that raise questions about certain design assumptions, speech versus DTMF usage, global command usage, identification strategies, and more.