Yahoo’s platform to create intelligent conversational assistants.
A conversational assistant needs to understand what a user is saying, what he means when he says that and what to do when he means that. Today, there are platforms that let us build assistants trained to have all these capabilities. Yahoo’s AI studio is one such platform. It was originally conceived as a public platform but is now used exclusively to build Yahoo’s own bots, skills and assistants.
My Role
I designed the overall navigation and information architecture for the platform and the ‘Flows’ functionality in detail among other smaller parts.
Navigation
When I joined the team, they had built a number of features for the platform but had not focused on it's information architecture and navigation.
I took a deep look at various components that made up the platform and how they worked together and also undertook a competitive analysis study reviewing other platforms like Meya, Api.ai and Wit.
Combining these learnings, the new IA and navigation emerged.
Need for Flows
We realized that while our platform made it easier to train an assistant to understand a user’s utterances and provide simple answers, it was unable to handle complex business logic or have a meaningful conversation.
“What time is my meeting with Alex?”
(Bot) “You are meeting Alex at 9:00am today”
“Can you ask him to move our meeting to 10:00am instead”
(Bot) “Whom should I ask?”
Get Simarjeet Singh’s stories in your inbox
Join Medium for free to get updates from this writer.
There needed to be a way to keep track of context in a conversation and convert product flowcharts into logic.
Design for Flows
I transferred my initial understanding of what flows were into the first design iteration. Feedback from users (developers, trainers and product managers) was that while it was easy to understand, it did not make context management much easier.
The second iteration broke flow creation and flow visualization into two separate views. It also emphasized that a flow was made up of triggers and states. Triggers made an assistant to go from one state to another. This concept got traction with users and I started refining it.
Refinement
A few cycles of proposals, feedback and refinement were spent to figure out the right way to create a flow and the best way to visualize it.
Finally a combination was achieved that seemed to perform well on all considered parameters.
Looking Ahead
Releasing AI Studio to the public (as was originally planned) would have given us a lot of insights on how people were using it and where it needed improvements. Although number of users are limited inside Yahoo, we plan to keep learning from them and keep improving the platform further.
