As skills became more popular and grew in number, the requirement for users to speak the names of individual skills became unwieldy and difficult, subject to names and invocations that weren’t connected to the mental models users had of their experiences.  Working with product and technical managers, I helped develop several fallback mechanisms to go beyond skill name and invocation models to allow users to interact with skills more naturally.

Working with NLU, ML and product teams, these interactions included name-free utterances that were associated exclusively with a previously used skill (“when is my ride coming?”), are likely intended for a skill (“make ocean sounds”), or couldn’t be fulfilled by Alexa, but can be with a skill (“order a pizza.”)  This also included using elastic search to target candidate utterances that failed Q&A but were possibly fulfilled with a skill.