Caden AI Beta
Overview
Beta of Caden's first personal AI tool that has access to their data to enable more personalized responses. Deciding to release a beta version serves several purposes: it reduces initial engineering time and costs, provides insights into user interactions, and allows the data team to make some improvements before launching the full model. To ensure users receive meaningful responses, a natural language form model was implemented, offering controlled and reliable query handling.
Benmarking
Some intital research to see other products with natural language forms.
Building The Queries
Original Query Builder
Partnered with the data team to establish queries that would yield interesting answers to users. At the time we had retail, entertainment, travel and location data to pull from. We came up with some follow-up questions to mimic the feeling of a chat.
Updated Query Builder
We realized the above solution was a bit limited and the user would get through answers quickly, especially since not all users have all these data types connected. To make it more interesting, we decided to play around with different phrasing to give the illusion of endless questions.
User Flows
There were a few flows and variables to think through. For the ideal experience the user will need to have sufficient data from the corresponding source however, since the users' data was a huge variable, we had to think through all other scenarios. Another requirement worth noting is that we wanted to limit the amount of questions the user could ask per week.
Question Builder
This is how the user starts the experience and begins building their question.
Outputs
This is what the user see after they form their prompt. The answer will type out and then will show a possible follow-up question they could ask. If the user does not meet the data requirements, they will be directed to add more accounts or try another question.
Prototype
What's Next?
The beta feature will launch to a select group of users. Product and design will observe user behaviors and we will decide where to take the next iteration.
