Leveraging AI To Power Up Speech
Yoodli is the top AI-powered speech coaching platform.
The Challenge.
Yoodli was already the go-to app for anyone serious about leveling up their speaking skills. The app leverages AI to grade your conference calls. From there, Yoodli directs you to relevant "next steps" for improvement.
We realized that Yoodli's "next step" was conversation crafting and categorization, and a necessary first step for that is an interface that allows users to pick what type of conversation they are having, along with specific talking points for the call.
Enter the AI-powered teleprompter. The design process for this feature required deep collaboration with engineering and heavy customer research.
A Starting Point
The list of asks within the AI-teleprompter epic was never-ending. But our job was to deliver value fast. That meant drilling down to what was the most important part of this feature set, and working with engineering to make sure we could pull it off. From customer research, we learned a few critical points:
People didn't know what the "best" version of an interview or speech looked like. Was there a way to get expert advice around a specific conversation?
Yoodlers planned their speeches out in Google Docs, Notion, and other text editing software. They often had one of these windows open as they talked.
In my design process, we built the empty state of this feature early to help define what an initial feature set for an AI teleprompter could look like.
Weighing Solutions
Many users said they wanted a Google Docs extension to be the primary interface for this feature. I pushed back on that notion (thinking that maybe some users use Notion), and asked them if a simple copy-and-paste would work just as well. In user testing, we also found that many users wanted to edit their talking points just before a call, or even during a call.
Pivoting
A big assumption we made was that users would want to tie their talking points directly to calendar events. In user research sessions, however, we found that many would modify their makeshift scripts just before a call or even during one. I made the decision to pivot to a more free-form approach.
Successes
We were able to build this feature in only a month and quickly scale to a full release from there. By collaborating heavily with engineering we were able to build an “impossibly huge” feature in record time.