In this talk, we tell the story of how we built Calendar.help at Microsoft Research. Calendar.help schedules people’s meetings just like an executive assistant. People interact with the system by emailing their virtual assistant, Cortana. For example, if Bob wants to meet with Alice, he could email her and cc Cortana. Cortana then follows up with Alice, handling the back-and-forth emails until a meeting time is finalized.
To handle any complex scheduling request, we introduce a novel three-tiered hybrid-intelligent architecture, combining automation with humans-in-the-loop. Common scheduling scenarios are broken down using workflows and completed as a series of microtasks that are automated when possible and executed by humans otherwise. Unusual scenarios fall back to trained human assistants who execute them as unstructured macrotasks.
We describe the approach we used to develop Calendar.help, and how we were able to turn research into a beta product in collaboration with the Outlook team. Our findings provide insight into how complex information tasks can be broken down into repeatable components that can be executed efficiently to improve productivity.
We share the lessons learned from scheduling thousands of meetings during over two years of real-world usage.
Andrés is a researcher at Microsoft AI & Research and an affiliate professor at the University of Washington. His research focuses on the design and study of social computing systems.
Justin Cranshaw is an HCI researcher at Microsoft AI & Research, where he’s been focused on building intelligent systems to make people more productive and effective at work.