This talk presents high-level initial findings of a comprehensive study of caller behavior at a prompt-by-prompt level to determine which wording works the best. The study includes millions of calls across many vertical  domains. It focuses on many of the most common types of prompting, such as date of birth, credit card numbers, and language preference, and reveals the results of the best-performing wording for each.

As VUI designers, we strive to create a natural experience for our users that takes into account external distractions, the ephemeral nature of speech, and the unnaturalness of talking to a machine optimized for a small set of tasks.  This talk covers the mechanisms of human conversation that enable efficient and informative interaction and their adaptation to dialogues with a minimally sentient nonhuman partner.