Back
AI Feedback Tool
As feedback came in from varied sources, from product demos to user interviews, we were faced with the challenge of tracking and understanding all the notes and feedback.
In numerous meetings between scientists, customers and the product team, crucial details often got lost in the flood of notes. We needed a way to identify key points and actionable items repeatedly discussed across these sessions, streamlining our focus on what truly mattered. We saw an opportunity that AI would help us group, prioritize and turn insights in to actionable items.
Role
UX Researcher
UX Research Methods
User interviews, reviewing feedback documents
Collaborators
CTO, fellow design team members
Tools
Figma, Coda, ChatGPT, Monday
Discovery
Problem Identfication: Notes, feedback, and customer insights came from disparate sources and distilling raw notes into actionable insights is essential. However, the manual process of doing this is time-consuming and prone to error.
User Interviews: Through discussions with professionals who often work with notes (e.g., scientists, students, and internal team memeber), a common theme emerged: the need for quick, organized data extraction from raw notes.
User interview w/ a scientist and my Product Manager
Cardsorting the menus and potential categories
User Feedback
Users expressed a desire for a feature that would allow them to tag or categorize notes, especially when dealing with larger datasets. They highlighted the potential benefits of a system that could automatically recognize and categorize notes based on content or context.
Design Decision
AI Integration: I collaborated with the dev team to train the AI. We wanted the tool to learn how to categorize notes, analyze sentiment, and determine sources. We used CODA as the platform as the interface of the tool. We manage to hook in to ChatGPT on the backend of the tool for AI analysis.
Using CODA as the Interface: CODA is notes tool akin to Notion and its flexible document interface was adapted as the primary platform for the tool. It modularity and ease of use enabled a customized workspace tailored to the tool's needs. Check the video below if you are unfamiliar with CODA.
CODA Video
Here is the flow we brainstormed for the back end:
Here is the flow for what we want our internal users to achieve:
Design
Sketches: Initial designs showed a simple interface with upload options and an AI-activation button.
Wireframes
Initial wireframes were designed to emphasize simplicity and clarity.
What we Built
Utilizing CODA's Interface - We employed CODA's dynamic document interface as the foundation for our tool. The adaptability of CODA permitted a user-friendly workspace tailored to seamlessly accept and present notes.
Note Input - CODA's tables and input forms became the primary methods for users to upload or type in their raw notes. The drag-and-drop functionality made this process intuitive.
Note Analysis with ChatGPT - Once notes were ready for AI analysis, we integrated ChatGPT with CODA to review the notes, categorizing notes, analyzing sentiment, and even identifying potential sub-categories and sources.
Notes Detail - Analysis line by line the category of the note, tagging, sentiment, and context of the note.
Table Before AI Insights
Annotations
Notes - User generated, copy and pasted or uploaded content broken down line by line in a table
High level Code - AI determines the category of the note
Set subcode - AI determines the sub-category of the note
Auto Summary - AI summarizing the note
Sentiment scale - AI determining the sentiment of the note
Meeting - user generated, origin of feedback
Table After AI Insights
Now watch ChatGPT generate insights after the push of a button
Report Generation
Using CODA's views and publishing capabilities, analyzed data was transformed into comprehensive reports. Users were presented with a summary view first, showcasing the primary insights derived from ChatGPT's deep dive into their notes.
Conclusion
After developing an alpha build of the AI Notes Aggregation Tool, it was immediately rolled out for internal use within the company. Because its an internal facing tool, our team members actively tested its features, providing real-time feedback to refine and enhance its capabilities. This hands-on approach ensured that the tool was not only innovative but also tailored to meet the practical needs of of the product team.